box3: fix gamebook audio, accents, and crashes

The BOX-3 touch gamebook had several blocking bugs that made the
voiced+foley story packs unusable. This fixes them end to end.

- Stack overflow on story start: gb_play_task kept a 2 KB buffer on
  its 4 KB stack and rebooted the board. Buffers are now static (the
  task is a singleton) and the stack is 6 KB.
- No sound at all: the ES8311 codec was never initialised — main.c
  wrote raw I2S to a muted DAC. Output (and the mic) now go through
  the BSP esp_codec_dev (bsp_audio_codec_speaker/microphone_init,
  esp_codec_dev_write/read). The power amplifier (GPIO46) is also
  forced on and volume set to 100, as the es8311 pa_pin handling did
  not drive it in practice.
- First passage silent: gb_play re-mounted the SD that the gamebook
  had already mounted; the failed second bsp_sdcard_mount tore down
  the VFS so the first fopen failed. Now we open the file first and
  only mount if that fails. A short silence lead-in also primes the
  codec/PA ramp.
- No accents: the gamebook used ASCII-only lv_font_montserrat_14/24.
  Added custom Montserrat fonts (font_fr_14/24, Latin-1 + œŒŸ « » ’ …,
  uncompressed so they render without LV_USE_FONT_COMPRESSED).
- Mic/voice streaming gated behind CONFIG_BOX3_VOICE_STREAMING
  (default off): as a gamebook the BOX-3 needs no mic, and the
  constant codec reads + bridge reconnects interfered with playback
  and flooded the log.

plip_virtual + cmd_exec handle types switched from i2s_chan_handle_t
to esp_codec_dev_handle_t accordingly.

box3_voice/main/CMakeLists.txt

@@ -1,5 +1,5 @@
 idf_component_register(
-    SRCS "main.c" "voice_ws_client.c" "scenario_server.c" "plip_virtual.c" "plip_ui.c" "stimulus.c"
+    SRCS "main.c" "voice_ws_client.c" "scenario_server.c" "plip_virtual.c" "plip_ui.c" "stimulus.c" "cmd_exec.c" "gamebook.c" "font_fr_14.c" "font_fr_24.c"
     INCLUDE_DIRS "."
     PRIV_REQUIRES
         driver
@@ -13,5 +13,6 @@ idf_component_register(
         spiffs
         scenario_mesh
         espressif__esp-box-3
+        espressif__esp_codec_dev
         lvgl__lvgl
 )

box3_voice/main/Kconfig.projbuild

@@ -1,5 +1,15 @@
 menu "Zacus BOX-3 Voice Configuration"
 
+    config BOX3_VOICE_STREAMING
+        bool "Enable microphone + voice-bridge streaming"
+        default n
+        help
+            Start the mic capture task and the WebSocket voice-bridge client.
+            Off by default: as a touch gamebook the BOX-3 needs no microphone,
+            and the continuous codec reads + bridge reconnect attempts interfere
+            with speaker playback (the first narration clip came out silent) and
+            flood the log. Enable only when using the BOX-3 as a voice device.
+
     config ZACUS_WIFI_SSID
         string "WiFi SSID"
         default "zacus-net"

box3_voice/main/cmd_exec.c

@@ -0,0 +1,667 @@
+// cmd_exec.c — BOX-3 executor for structured ESP-NOW CMD frames (D5 contract).
+// See cmd_exec.h for the wire format spec.
+//
+// Objective 1 — screen: full-scene faithful rendering
+//   Palette: bg #0055AA / symbol #FF8800 / title #FFFFFF / subtitle #AAAAAA
+//   Effects:  pulse  (opacity breathing on symbol via LVGL anim)
+//             glitch (rapid opacity flicker on title via lv_timer)
+//             gyro   (rotating arc ring around symbol)
+//             none   (static)
+//
+// Objective 2 — play: real WAV streaming to I2S
+//   1. Mount SD card (best-effort, /sdcard). Parse WAV header (PCM-16).
+//   2. Stream PCM samples via the ES8311 codec (esp_codec_dev), same path as TTS.
+//   3. If SD absent / file not found → fallback bip + log.
+//   4. Embedded test cue always available at virtual path "embedded://" or
+//      when the requested path starts with "embedded:" — proves WAV decode + I2S.
+
+#include "cmd_exec.h"
+
+#include <string.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <inttypes.h>
+
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+
+#include "bsp/esp-bsp.h"
+#include "cJSON.h"
+#include "esp_log.h"
+#include "esp_vfs_fat.h"
+#include "lvgl.h"
+
+#include "embedded_wav.h"   // s_embedded_wav[], EMBEDDED_WAV_SIZE
+
+// audio_play_tone is defined in main.c (writes through the ES8311 codec).
+extern void audio_play_tone(float frequency, int duration_ms);
+
+// Speaker codec (ES8311) handle — obtained from main.c after speaker_init().
+// Audio is streamed through esp_codec_dev so the ES8311 DAC is actually driven
+// (a raw I2S write leaves the codec muted).
+#include "esp_codec_dev.h"
+static esp_codec_dev_handle_t s_spk_handle_local = NULL;
+
+void cmd_exec_set_spk_handle(esp_codec_dev_handle_t h)
+{
+    s_spk_handle_local = h;
+}
+
+#define TAG "cmd_exec"
+
+// ─── Palette (matches idf_zacus/components/display_ui reference) ─────────────
+#define COL_BG      0x0055AA  // Workbench blue
+#define COL_SYMBOL  0xFF8800  // Workbench orange
+#define COL_TITLE   0xFFFFFF  // white
+#define COL_SUB     0xAAAAAA  // mid grey
+
+#define SCREEN_LOCK_MS 1000
+
+// ─── Screen overlay state ─────────────────────────────────────────────────────
+
+static lv_obj_t *s_overlay    = NULL;
+static lv_obj_t *s_title_lbl  = NULL;
+static lv_obj_t *s_sub_lbl    = NULL;
+static lv_obj_t *s_sym_lbl    = NULL;
+static lv_obj_t *s_gyro_arc   = NULL;
+
+// Active animations / timers (so we can stop them on next call)
+static lv_timer_t *s_glitch_timer = NULL;
+
+// ─── Effect callbacks ─────────────────────────────────────────────────────────
+
+static void pulse_anim_cb(void *obj, int32_t v)
+{
+    lv_obj_set_style_opa((lv_obj_t *)obj, (lv_opa_t)v, 0);
+}
+
+static void gyro_anim_cb(void *obj, int32_t v)
+{
+    lv_arc_set_end_angle((lv_obj_t *)obj, (uint16_t)(v % 360));
+}
+
+static void glitch_timer_cb(lv_timer_t *t)
+{
+    (void)t;
+    static bool flip = false;
+    flip = !flip;
+    lv_obj_set_style_opa(s_title_lbl,
+                         flip ? LV_OPA_20 : LV_OPA_COVER, 0);
+    // Slight X jitter: move label 0 or 3 pixels right alternately
+    lv_obj_set_x(s_title_lbl, flip ? 3 : 0);
+}
+
+// ─── Stop all active effects ──────────────────────────────────────────────────
+
+static void stop_effects_locked(void)
+{
+    // Stop pulse animation on symbol
+    lv_anim_del(s_sym_lbl, pulse_anim_cb);
+    if (s_sym_lbl) {
+        lv_obj_set_style_opa(s_sym_lbl, LV_OPA_COVER, 0);
+    }
+
+    // Stop glitch timer
+    if (s_glitch_timer) {
+        lv_timer_pause(s_glitch_timer);
+    }
+    if (s_title_lbl) {
+        lv_obj_set_style_opa(s_title_lbl, LV_OPA_COVER, 0);
+        lv_obj_set_x(s_title_lbl, 0);
+    }
+
+    // Stop gyro arc
+    if (s_gyro_arc) {
+        lv_anim_del(s_gyro_arc, gyro_anim_cb);
+        lv_obj_add_flag(s_gyro_arc, LV_OBJ_FLAG_HIDDEN);
+    }
+}
+
+// ─── Build overlay (lazy, first CMD only) ────────────────────────────────────
+
+static void build_overlay_locked(void)
+{
+    if (s_overlay) return;
+
+    lv_obj_t *scr = lv_screen_active();
+
+    // Full-screen panel with Workbench blue background
+    s_overlay = lv_obj_create(scr);
+    lv_obj_set_size(s_overlay, LV_PCT(100), LV_PCT(100));
+    lv_obj_set_pos(s_overlay, 0, 0);
+    lv_obj_set_style_bg_color(s_overlay, lv_color_hex(COL_BG), 0);
+    lv_obj_set_style_bg_opa(s_overlay, LV_OPA_COVER, 0);
+    lv_obj_set_style_border_width(s_overlay, 0, 0);
+    lv_obj_set_style_radius(s_overlay, 0, 0);
+    lv_obj_set_style_pad_all(s_overlay, 10, 0);
+
+    // Title — top, white, font 24
+    s_title_lbl = lv_label_create(s_overlay);
+    lv_label_set_text(s_title_lbl, "");
+    lv_obj_set_style_text_font(s_title_lbl, &lv_font_montserrat_24, 0);
+    lv_obj_set_style_text_color(s_title_lbl, lv_color_hex(COL_TITLE), 0);
+    lv_label_set_long_mode(s_title_lbl, LV_LABEL_LONG_WRAP);
+    lv_obj_set_width(s_title_lbl, LV_PCT(90));
+    lv_obj_set_style_text_align(s_title_lbl, LV_TEXT_ALIGN_CENTER, 0);
+    lv_obj_align(s_title_lbl, LV_ALIGN_TOP_MID, 0, 14);
+
+    // Symbol — center, orange, font 48
+    s_sym_lbl = lv_label_create(s_overlay);
+    lv_label_set_text(s_sym_lbl, "");
+    lv_obj_set_style_text_font(s_sym_lbl, &lv_font_montserrat_48, 0);
+    lv_obj_set_style_text_color(s_sym_lbl, lv_color_hex(COL_SYMBOL), 0);
+    lv_obj_align(s_sym_lbl, LV_ALIGN_CENTER, 0, -10);
+
+    // Gyro arc: hidden rotating ring around symbol (gyro effect)
+    s_gyro_arc = lv_arc_create(s_overlay);
+    lv_obj_set_size(s_gyro_arc, 90, 90);
+    lv_obj_align(s_gyro_arc, LV_ALIGN_CENTER, 0, -10);
+    lv_arc_set_rotation(s_gyro_arc, 270);
+    lv_arc_set_bg_angles(s_gyro_arc, 0, 360);
+    lv_obj_set_style_arc_color(s_gyro_arc, lv_color_hex(COL_SYMBOL), LV_PART_INDICATOR);
+    lv_obj_set_style_arc_width(s_gyro_arc, 4, LV_PART_INDICATOR);
+    lv_obj_set_style_arc_opa(s_gyro_arc, LV_OPA_TRANSP, LV_PART_MAIN);
+    lv_obj_set_style_bg_opa(s_gyro_arc, LV_OPA_TRANSP, 0);
+    lv_obj_set_style_border_width(s_gyro_arc, 0, 0);
+    lv_obj_add_flag(s_gyro_arc, LV_OBJ_FLAG_HIDDEN);
+
+    // Subtitle — bottom, grey, font 14
+    s_sub_lbl = lv_label_create(s_overlay);
+    lv_label_set_text(s_sub_lbl, "");
+    lv_obj_set_style_text_font(s_sub_lbl, &lv_font_montserrat_14, 0);
+    lv_obj_set_style_text_color(s_sub_lbl, lv_color_hex(COL_SUB), 0);
+    lv_label_set_long_mode(s_sub_lbl, LV_LABEL_LONG_WRAP);
+    lv_obj_set_width(s_sub_lbl, LV_PCT(90));
+    lv_obj_set_style_text_align(s_sub_lbl, LV_TEXT_ALIGN_CENTER, 0);
+    lv_obj_align(s_sub_lbl, LV_ALIGN_BOTTOM_MID, 0, -14);
+
+    // Glitch timer: created paused, resumed by glitch effect
+    s_glitch_timer = lv_timer_create(glitch_timer_cb, 120, NULL);
+    lv_timer_pause(s_glitch_timer);
+}
+
+// ─── op=screen ────────────────────────────────────────────────────────────────
+
+static esp_err_t exec_screen(const cJSON *a)
+{
+    const cJSON *title = a ? cJSON_GetObjectItemCaseSensitive(a, "t") : NULL;
+    const cJSON *sub   = a ? cJSON_GetObjectItemCaseSensitive(a, "u") : NULL;
+    const cJSON *sym   = a ? cJSON_GetObjectItemCaseSensitive(a, "y") : NULL;
+    const cJSON *eff   = a ? cJSON_GetObjectItemCaseSensitive(a, "e") : NULL;
+
+    const char *t_str = cJSON_IsString(title) ? title->valuestring : "";
+    const char *u_str = cJSON_IsString(sub)   ? sub->valuestring   : "";
+    const char *y_str = cJSON_IsString(sym)   ? sym->valuestring   : "";
+    const char *e_str = cJSON_IsString(eff)   ? eff->valuestring   : "none";
+
+    ESP_LOGI(TAG, "CMD op=screen t=\"%s\" u=\"%s\" y=\"%s\" e=\"%s\"",
+             t_str, u_str, y_str, e_str);
+
+    if (!bsp_display_lock(SCREEN_LOCK_MS)) {
+        ESP_LOGW(TAG, "screen: display lock timeout");
+        return ESP_FAIL;
+    }
+
+    build_overlay_locked();
+    stop_effects_locked();
+
+    // Update labels
+    lv_label_set_text(s_title_lbl, t_str);
+    lv_label_set_text(s_sym_lbl,   y_str);
+    lv_label_set_text(s_sub_lbl,   u_str);
+
+    // Re-align after text update
+    lv_obj_align(s_title_lbl, LV_ALIGN_TOP_MID, 0, 14);
+    lv_obj_align(s_sym_lbl,   LV_ALIGN_CENTER,  0, -10);
+    lv_obj_align(s_sub_lbl,   LV_ALIGN_BOTTOM_MID, 0, -14);
+
+    // Bring overlay to front
+    lv_obj_move_foreground(s_overlay);
+
+    // Apply effect
+    if (strcmp(e_str, "pulse") == 0) {
+        // Opacity breathing: COVER → 50 → COVER, 600ms per half
+        lv_anim_t a;
+        lv_anim_init(&a);
+        lv_anim_set_var(&a, s_sym_lbl);
+        lv_anim_set_exec_cb(&a, pulse_anim_cb);
+        lv_anim_set_values(&a, LV_OPA_COVER, LV_OPA_50);
+        lv_anim_set_time(&a, 600);
+        lv_anim_set_playback_time(&a, 600);
+        lv_anim_set_repeat_count(&a, LV_ANIM_REPEAT_INFINITE);
+        lv_anim_start(&a);
+        ESP_LOGI(TAG, "screen: pulse effect started");
+
+    } else if (strcmp(e_str, "glitch") == 0) {
+        // Rapid opacity + X-jitter on title
+        if (s_glitch_timer) {
+            lv_timer_resume(s_glitch_timer);
+        }
+        ESP_LOGI(TAG, "screen: glitch effect started");
+
+    } else if (strcmp(e_str, "gyro") == 0) {
+        // Rotating arc ring around the symbol
+        if (s_gyro_arc) {
+            lv_obj_clear_flag(s_gyro_arc, LV_OBJ_FLAG_HIDDEN);
+            lv_anim_t a;
+            lv_anim_init(&a);
+            lv_anim_set_var(&a, s_gyro_arc);
+            lv_anim_set_exec_cb(&a, gyro_anim_cb);
+            lv_anim_set_values(&a, 0, 360);
+            lv_anim_set_time(&a, 1200);
+            lv_anim_set_repeat_count(&a, LV_ANIM_REPEAT_INFINITE);
+            lv_anim_start(&a);
+        }
+        ESP_LOGI(TAG, "screen: gyro effect started");
+
+    } else {
+        // "none" or unrecognised → static
+        ESP_LOGI(TAG, "screen: static (no effect)");
+    }
+
+    bsp_display_unlock();
+    return ESP_OK;
+}
+
+// ─── WAV player ───────────────────────────────────────────────────────────────
+//
+// WAV header layout (canonical PCM, 44 bytes):
+//   RIFF header : 12 bytes  (RIFF, chunk size, WAVE)
+//   fmt  chunk  : 24 bytes  (fmt , 16, pcm=1, channels, rate, byte_rate, align, bits)
+//   data chunk  : 8 bytes   (data, data_size)
+//   PCM samples : data_size bytes
+
+#define WAV_HDR_RIFF_OFS   0
+#define WAV_HDR_FMT_OFS    12
+#define WAV_HDR_DATA_OFS   36
+#define WAV_MIN_HEADER     44
+
+typedef struct {
+    uint16_t audio_format;
+    uint16_t channels;
+    uint32_t sample_rate;
+    uint32_t byte_rate;
+    uint16_t block_align;
+    uint16_t bits_per_sample;
+    uint32_t data_offset;   // offset of PCM data in the buffer/file
+    uint32_t data_size;
+} wav_info_t;
+
+static esp_err_t parse_wav_header(const uint8_t *buf, size_t buf_len, wav_info_t *out)
+{
+    if (buf_len < WAV_MIN_HEADER) return ESP_ERR_INVALID_SIZE;
+
+    // RIFF
+    if (memcmp(buf, "RIFF", 4) != 0) return ESP_ERR_INVALID_ARG;
+    if (memcmp(buf + 8, "WAVE", 4) != 0) return ESP_ERR_INVALID_ARG;
+
+    // Walk chunks to find fmt and data (handles non-standard ordering / extra chunks)
+    size_t pos = 12;
+    bool got_fmt = false, got_data = false;
+
+    while (pos + 8 <= buf_len && !(got_fmt && got_data)) {
+        uint32_t chunk_size;
+        memcpy(&chunk_size, buf + pos + 4, 4);
+
+        if (memcmp(buf + pos, "fmt ", 4) == 0 && chunk_size >= 16) {
+            memcpy(&out->audio_format,   buf + pos + 8,  2);
+            memcpy(&out->channels,       buf + pos + 10, 2);
+            memcpy(&out->sample_rate,    buf + pos + 12, 4);
+            memcpy(&out->byte_rate,      buf + pos + 16, 4);
+            memcpy(&out->block_align,    buf + pos + 20, 2);
+            memcpy(&out->bits_per_sample,buf + pos + 22, 2);
+            got_fmt = true;
+        } else if (memcmp(buf + pos, "data", 4) == 0) {
+            out->data_offset = (uint32_t)(pos + 8);
+            out->data_size   = chunk_size;
+            got_data = true;
+        }
+        pos += 8 + chunk_size;
+    }
+
+    if (!got_fmt || !got_data) return ESP_ERR_NOT_FOUND;
+    if (out->audio_format != 1) {
+        ESP_LOGW(TAG, "WAV format %d is not PCM (1) — unsupported", out->audio_format);
+        return ESP_ERR_NOT_SUPPORTED;
+    }
+
+    return ESP_OK;
+}
+
+// Stream raw PCM-16 samples to the speaker I2S channel.
+// Handles any sample rate (simple rate-adaptation via repeat/skip is NOT done here;
+// the test WAV and typical cues should be 16 kHz mono = native rate).
+static esp_err_t stream_pcm16(const uint8_t *pcm, size_t byte_len,
+                               bool loop, int max_loops)
+{
+    if (!s_spk_handle_local) {
+        ESP_LOGW(TAG, "play: no speaker handle — skip stream");
+        return ESP_ERR_INVALID_STATE;
+    }
+
+    int loops = 0;
+    do {
+        size_t offset = 0;
+        while (offset < byte_len) {
+            size_t chunk = (byte_len - offset < 1024) ? (byte_len - offset) : 1024;
+            int ret = esp_codec_dev_write(s_spk_handle_local,
+                                          (void *)(pcm + offset), (int)chunk);
+            if (ret != ESP_CODEC_DEV_OK) {
+                ESP_LOGW(TAG, "codec write error: %d", ret);
+                return ESP_FAIL;
+            }
+            offset += chunk;
+        }
+        loops++;
+    } while (loop && (max_loops == 0 || loops < max_loops));
+
+    return ESP_OK;
+}
+
+// ─── SD card mount (best-effort, guarded) ────────────────────────────────────
+
+static bool s_sd_mounted = false;
+
+static void ensure_sd_mounted(void)
+{
+    if (s_sd_mounted) return;
+    ESP_LOGI(TAG, "play: attempting SD card mount...");
+    esp_err_t ret = bsp_sdcard_mount();
+    if (ret == ESP_OK) {
+        s_sd_mounted = true;
+        ESP_LOGI(TAG, "play: SD card mounted at %s", BSP_SD_MOUNT_POINT);
+    } else {
+        ESP_LOGW(TAG, "play: SD card mount failed: %s (will use fallback)", esp_err_to_name(ret));
+    }
+}
+
+// ─── play task (off main task to allow loop without blocking CMD recv) ────────
+
+typedef struct {
+    char path[128];
+    bool loop;
+} play_args_t;
+
+static void play_task(void *arg)
+{
+    play_args_t *pargs = (play_args_t *)arg;
+    char path[128];
+    bool loop = pargs->loop;
+    strncpy(path, pargs->path, sizeof(path) - 1);
+    path[sizeof(path) - 1] = '\0';
+    free(pargs);
+
+    ESP_LOGI(TAG, "play_task: path=\"%s\" loop=%d", path, loop);
+
+    // Check for embedded asset path
+    bool use_embedded = (strncmp(path, "embedded:", 9) == 0 ||
+                         strcmp(path, "<none>") == 0 ||
+                         strlen(path) == 0);
+
+    if (!use_embedded) {
+        // Try SD card
+        ensure_sd_mounted();
+        if (!s_sd_mounted) {
+            ESP_LOGW(TAG, "play: SD not mounted, falling back to embedded asset");
+            use_embedded = true;
+        }
+    }
+
+    if (!use_embedded) {
+        // Build full path if not absolute
+        char full_path[160];
+        if (path[0] == '/') {
+            snprintf(full_path, sizeof(full_path), "%s", path);
+        } else {
+            snprintf(full_path, sizeof(full_path), "%s/%s", BSP_SD_MOUNT_POINT, path);
+        }
+
+        FILE *f = fopen(full_path, "rb");
+        if (!f) {
+            ESP_LOGW(TAG, "play: file not found: %s — falling back to embedded", full_path);
+            use_embedded = true;
+        } else {
+            // Read WAV from file
+            fseek(f, 0, SEEK_END);
+            long fsize = ftell(f);
+            fseek(f, 0, SEEK_SET);
+
+            if (fsize < WAV_MIN_HEADER || fsize > 512 * 1024) {
+                ESP_LOGW(TAG, "play: file size %ld bytes — out of range, fallback", fsize);
+                fclose(f);
+                use_embedded = true;
+            } else {
+                uint8_t *fbuf = malloc((size_t)fsize);
+                if (!fbuf) {
+                    ESP_LOGW(TAG, "play: alloc failed for %ld bytes — fallback", fsize);
+                    fclose(f);
+                    use_embedded = true;
+                } else {
+                    size_t nread = fread(fbuf, 1, (size_t)fsize, f);
+                    fclose(f);
+
+                    wav_info_t wi;
+                    esp_err_t ret = parse_wav_header(fbuf, nread, &wi);
+                    if (ret != ESP_OK) {
+                        ESP_LOGW(TAG, "play: WAV parse error %s — fallback", esp_err_to_name(ret));
+                        free(fbuf);
+                        use_embedded = true;
+                    } else {
+                        ESP_LOGI(TAG, "play: WAV ok — %" PRIu32 " Hz %" PRIu16 "-bit %" PRIu16 " ch, %" PRIu32 " bytes PCM",
+                                 wi.sample_rate, wi.bits_per_sample,
+                                 wi.channels, wi.data_size);
+                        if (wi.bits_per_sample != 16) {
+                            ESP_LOGW(TAG, "play: %" PRIu16 "-bit WAV — only 16-bit supported, fallback",
+                                     wi.bits_per_sample);
+                            free(fbuf);
+                            use_embedded = true;
+                        } else {
+                            // Real streaming from file
+                            stream_pcm16(fbuf + wi.data_offset, wi.data_size, loop, 3);
+                            free(fbuf);
+                            use_embedded = false;
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    if (use_embedded) {
+        // Play embedded C5-E5-G5 cue (proves WAV decode + I2S path)
+        wav_info_t wi;
+        esp_err_t ret = parse_wav_header(s_embedded_wav, EMBEDDED_WAV_SIZE, &wi);
+        if (ret == ESP_OK && wi.bits_per_sample == 16) {
+            ESP_LOGI(TAG, "play: streaming embedded cue (%" PRIu32 " Hz %" PRIu16 "-bit, %" PRIu32 " bytes PCM)",
+                     wi.sample_rate, wi.bits_per_sample, wi.data_size);
+            stream_pcm16(s_embedded_wav + wi.data_offset, wi.data_size, false, 1);
+        } else {
+            // Last-resort: beep
+            ESP_LOGW(TAG, "play: embedded WAV parse error %s — beep fallback",
+                     esp_err_to_name(ret));
+            audio_play_tone(880.0f, 200);
+        }
+    }
+
+    vTaskDelete(NULL);
+}
+
+// ─── Gamebook narration player (async, streamed, interruptible) ──────────────
+// One persistent task plays one WAV at a time, streaming straight from the SD
+// in small chunks (no full-file malloc → any length, RAM-safe). A new request
+// or a stop flag breaks the stream so tapping a choice cuts the narration.
+
+static TaskHandle_t   s_gb_task = NULL;
+static char           s_gb_path[160];
+static volatile bool  s_gb_new  = false;
+static volatile bool  s_gb_stop = false;
+
+static void gb_play_task(void *arg)
+{
+    (void) arg;
+    /* The task is a singleton (guarded by s_gb_task), so these large buffers
+     * are kept static — on the stack they ate ~2.1 KB and, together with the
+     * FATFS fread internals + i2s + logging, overflowed the 4 KB stack and
+     * rebooted the board when a story started. */
+    static uint8_t hdr[64];
+    static uint8_t buf[2048];
+    for (;;) {
+        if (!s_gb_new) { vTaskDelay(pdMS_TO_TICKS(20)); continue; }
+        char path[160];
+        strncpy(path, s_gb_path, sizeof(path) - 1);
+        path[sizeof(path) - 1] = '\0';
+        s_gb_new  = false;
+        s_gb_stop = false;
+
+        /* The SD is normally already mounted (the gamebook mounts it at init).
+         * Try the file directly first — re-mounting an already-mounted card via
+         * bsp_sdcard_mount() FAILS and tears down the existing VFS, which made
+         * the very first narration clip silent. Only mount if the open fails. */
+        FILE *f = fopen(path, "rb");
+        if (!f) {
+            ensure_sd_mounted();
+            f = fopen(path, "rb");
+        }
+        if (!f) { ESP_LOGW(TAG, "gb: open %s failed", path); continue; }
+        size_t hn = fread(hdr, 1, sizeof(hdr), f);
+        wav_info_t wi;
+        if (parse_wav_header(hdr, hn, &wi) != ESP_OK || wi.bits_per_sample != 16) {
+            ESP_LOGW(TAG, "gb: bad/usupported WAV %s", path);
+            fclose(f);
+            continue;
+        }
+        fseek(f, wi.data_offset, SEEK_SET);
+        /* Wake the codec/PA before the real audio: after idle the ES8311 mutes
+         * and ramps up on the first write, swallowing the start of the clip.
+         * A short silent lead-in lets the unmute settle so the narration plays
+         * from the very first word. */
+        memset(buf, 0, sizeof(buf));
+        for (int i = 0; i < 4 && s_spk_handle_local && !s_gb_stop && !s_gb_new; i++) {
+            esp_codec_dev_write(s_spk_handle_local, buf, (int)sizeof(buf));
+        }
+        size_t remaining = wi.data_size;
+        ESP_LOGI(TAG, "gb: play %s (%u B, %u Hz)", path,
+                 (unsigned)wi.data_size, (unsigned)wi.sample_rate);
+        while (remaining > 0 && !s_gb_stop && !s_gb_new) {
+            size_t want = remaining < sizeof(buf) ? remaining : sizeof(buf);
+            size_t got = fread(buf, 1, want, f);
+            if (got == 0) break;
+            if (s_spk_handle_local) {
+                esp_codec_dev_write(s_spk_handle_local, buf, (int)got);
+            }
+            remaining -= got;
+        }
+        fclose(f);
+    }
+}
+
+void cmd_exec_play_file_async(const char *path)
+{
+    if (!path || !path[0]) return;
+    strncpy(s_gb_path, path, sizeof(s_gb_path) - 1);
+    s_gb_path[sizeof(s_gb_path) - 1] = '\0';
+    s_gb_stop = true;    // interrupt whatever is playing
+    s_gb_new  = true;    // … and pick up the new clip
+    if (!s_gb_task) {
+        xTaskCreate(gb_play_task, "gb_play", 6144, NULL, 4, &s_gb_task);
+    }
+}
+
+void cmd_exec_stop_play(void)
+{
+    s_gb_stop = true;
+    s_gb_new  = false;
+}
+
+// ─── op=play ──────────────────────────────────────────────────────────────────
+
+static esp_err_t exec_play(const cJSON *a)
+{
+    const cJSON *path_j = a ? cJSON_GetObjectItemCaseSensitive(a, "p") : NULL;
+    const cJSON *loop_j = a ? cJSON_GetObjectItemCaseSensitive(a, "l") : NULL;
+    const char *path = cJSON_IsString(path_j) ? path_j->valuestring : "embedded://cue";
+    bool loop        = cJSON_IsNumber(loop_j)  ? ((int)loop_j->valuedouble != 0) : false;
+
+    ESP_LOGI(TAG, "CMD op=play path=\"%s\" loop=%d", path, loop);
+
+    play_args_t *pargs = malloc(sizeof(play_args_t));
+    if (!pargs) {
+        ESP_LOGW(TAG, "play: alloc failed — beep fallback");
+        audio_play_tone(880.0f, 200);
+        return ESP_ERR_NO_MEM;
+    }
+    strncpy(pargs->path, path, sizeof(pargs->path) - 1);
+    pargs->path[sizeof(pargs->path) - 1] = '\0';
+    pargs->loop = loop;
+
+    // Spawn a dedicated task (4 KB stack) so we don't block the CMD receive path.
+    // Priority 4 = same as mic_capture.
+    if (xTaskCreate(play_task, "cmd_play", 4096, pargs, 4, NULL) != pdPASS) {
+        ESP_LOGW(TAG, "play: task create failed — beep fallback");
+        free(pargs);
+        audio_play_tone(880.0f, 200);
+        return ESP_FAIL;
+    }
+
+    return ESP_OK;
+}
+
+// ─── op=evt ───────────────────────────────────────────────────────────────────
+
+static esp_err_t exec_evt(const cJSON *a)
+{
+    const cJSON *name_j = a ? cJSON_GetObjectItemCaseSensitive(a, "n") : NULL;
+    const char *name = cJSON_IsString(name_j) ? name_j->valuestring : "<none>";
+    ESP_LOGI(TAG, "CMD op=evt n=\"%s\" (evt received on BOX-3 — logging only)", name);
+    return ESP_OK;
+}
+
+// ─── op=led ───────────────────────────────────────────────────────────────────
+
+static esp_err_t exec_led(const cJSON *a)
+{
+    const cJSON *pat_j = a ? cJSON_GetObjectItemCaseSensitive(a, "p") : NULL;
+    const char *pat = cJSON_IsString(pat_j) ? pat_j->valuestring : "<none>";
+    ESP_LOGI(TAG, "CMD op=led pattern=\"%s\" (stub — no LED driver in box3_voice)", pat);
+    return ESP_OK;
+}
+
+// ─── public entry point ───────────────────────────────────────────────────────
+
+esp_err_t cmd_exec_handle(const char *data, size_t len)
+{
+    if (!data || len == 0) return ESP_ERR_INVALID_ARG;
+
+    cJSON *root = cJSON_ParseWithLength(data, len);
+    if (!root) {
+        ESP_LOGW(TAG, "CMD parse error: not valid JSON (%.40s...)", data);
+        return ESP_ERR_INVALID_ARG;
+    }
+
+    const cJSON *op_j = cJSON_GetObjectItemCaseSensitive(root, "op");
+    if (!cJSON_IsString(op_j) || !op_j->valuestring || !op_j->valuestring[0]) {
+        ESP_LOGW(TAG, "CMD missing or invalid 'op' field");
+        cJSON_Delete(root);
+        return ESP_ERR_INVALID_ARG;
+    }
+
+    const char  *op = op_j->valuestring;
+    const cJSON *a  = cJSON_GetObjectItemCaseSensitive(root, "a");
+
+    esp_err_t err;
+    if (strcmp(op, "screen") == 0) {
+        err = exec_screen(a);
+    } else if (strcmp(op, "play") == 0) {
+        err = exec_play(a);
+    } else if (strcmp(op, "evt") == 0) {
+        err = exec_evt(a);
+    } else if (strcmp(op, "led") == 0) {
+        err = exec_led(a);
+    } else {
+        ESP_LOGW(TAG, "CMD op=\"%s\" unknown — ignored", op);
+        err = ESP_ERR_NOT_SUPPORTED;
+    }
+
+    cJSON_Delete(root);
+    return err;
+}

box3_voice/main/cmd_exec.h

@@ -0,0 +1,47 @@
+// cmd_exec.h — BOX-3 executor for structured ESP-NOW CMD frames (D5 contract).
+//
+// Wire format (SCENARIO_MESH_TEXT_CMD, ≤200 bytes):
+//   {"op":"<verb>","a":{<args>}}
+//
+// Supported ops:
+//   screen  {"t":"title","u":"subtitle","y":"symbol","e":"effect"} — display on LCD
+//             effect: "pulse"|"glitch"|"gyro"|"none"
+//   play    {"p":"<path>","l":0|1} — stream WAV via I2S; embedded cue if no file
+//   evt     {"n":"<name>"}         — log the event (future: relay to master)
+//   led     {"p":"<pattern>"}      — log the pattern (stub)
+//   <any>   unknown op             — log warn, ignore
+//
+// Tolerant by design: malformed JSON, missing keys, or unknown ops log a warning
+// and return without crashing. Never asserts on network input.
+
+#pragma once
+
+#include <stddef.h>
+#include "esp_err.h"
+#include "esp_codec_dev.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Parse and execute one CMD frame. `data` is the raw UTF-8 payload received via
+// scenario_mesh text callback (NUL-terminated). Returns ESP_OK if the op was
+// recognised and dispatched, ESP_ERR_INVALID_ARG on bad JSON or missing `op`,
+// ESP_ERR_NOT_SUPPORTED for unknown ops (all logged, no crash).
+esp_err_t cmd_exec_handle(const char *data, size_t len);
+
+// Provide the speaker codec (ES8311) handle so exec_play() can stream audio.
+// Must be called from main.c after speaker_init(), before any CMD arrives.
+void cmd_exec_set_spk_handle(esp_codec_dev_handle_t h);
+
+// Play a 16 kHz mono 16-bit WAV from the SD asynchronously, streamed in chunks
+// (RAM-safe for long narration) and INTERRUPTIBLE: a new call stops the current
+// clip and starts the new one. Used by the touch gamebook for passage audio.
+void cmd_exec_play_file_async(const char *path);
+
+// Stop the async narration (no-op if nothing is playing).
+void cmd_exec_stop_play(void);
+
+#ifdef __cplusplus
+}
+#endif

box3_voice/main/gamebook.c

@@ -0,0 +1,238 @@
+// gamebook.c — see gamebook.h. Touch CYOA for the ESP32-S3-BOX-3.
+#include "gamebook.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "cJSON.h"
+#include "esp_log.h"
+#include "esp_heap_caps.h"
+
+#include "bsp/esp-bsp.h"
+#include "lvgl.h"
+#include "cmd_exec.h"   // async WAV narration player
+
+static const char *TAG = "gamebook";
+
+// Custom fonts with French glyphs (ASCII + Latin-1 + ’ + …). See main/font_fr_*.c.
+LV_FONT_DECLARE(font_fr_14);
+LV_FONT_DECLARE(font_fr_24);
+
+#define GB_DIR      "/sdcard/gamebook"
+#define GB_LIBRARY  GB_DIR "/library.json"
+#define GB_JSON_MAX (256 * 1024)        // expanded books are ~80 KB of JSON
+
+// Colours (Workbench-ish palette).
+#define COL_BG     0x101820
+#define COL_TITLE  0xFFCC55
+#define COL_TEXT   0xF0F0F0
+#define COL_BTN    0x224466
+
+static cJSON   *s_lib_root  = NULL;     // owns library.json
+static cJSON   *s_lib       = NULL;     // borrowed: root->"library"
+static int      s_lib_n     = 0;
+
+static cJSON   *s_book      = NULL;     // owns the current <id>.json
+static cJSON   *s_passages  = NULL;     // borrowed: book->"passages"
+
+static lv_obj_t *s_root     = NULL;     // scrollable full-screen column
+
+// ── PSRAM allocators for cJSON (book trees are large) ───────────────────────
+static void *gb_malloc(size_t sz) { return heap_caps_malloc(sz, MALLOC_CAP_SPIRAM); }
+static void  gb_free(void *p)      { heap_caps_free(p); }
+
+static cJSON *load_json(const char *path)
+{
+    FILE *f = fopen(path, "rb");
+    if (!f) { ESP_LOGW(TAG, "open %s failed", path); return NULL; }
+    fseek(f, 0, SEEK_END);
+    long sz = ftell(f);
+    rewind(f);
+    if (sz <= 0 || sz > GB_JSON_MAX) { fclose(f); ESP_LOGW(TAG, "%s size %ld", path, sz); return NULL; }
+    char *buf = heap_caps_malloc((size_t)sz + 1, MALLOC_CAP_SPIRAM);
+    if (!buf) { fclose(f); return NULL; }
+    size_t rd = fread(buf, 1, (size_t)sz, f);
+    fclose(f);
+    buf[rd] = '\0';
+    cJSON *root = cJSON_Parse(buf);
+    heap_caps_free(buf);
+    if (!root) ESP_LOGW(TAG, "malformed JSON: %s", path);
+    return root;
+}
+
+// ── Root container (created once, cleared per view) ─────────────────────────
+// All UI functions below assume the LVGL lock is held (gamebook_init takes it;
+// button callbacks already run inside the LVGL task / lock).
+static void ensure_root(void)
+{
+    if (s_root) { lv_obj_clean(s_root); return; }
+    lv_obj_t *scr = lv_screen_active();
+    s_root = lv_obj_create(scr);
+    lv_obj_set_size(s_root, LV_PCT(100), LV_PCT(100));
+    lv_obj_set_pos(s_root, 0, 0);
+    lv_obj_set_style_bg_color(s_root, lv_color_hex(COL_BG), 0);
+    lv_obj_set_style_bg_opa(s_root, LV_OPA_COVER, 0);
+    lv_obj_set_style_border_width(s_root, 0, 0);
+    lv_obj_set_style_radius(s_root, 0, 0);
+    lv_obj_set_style_pad_all(s_root, 8, 0);
+    lv_obj_set_style_pad_row(s_root, 8, 0);
+    lv_obj_set_flex_flow(s_root, LV_FLEX_FLOW_COLUMN);
+    lv_obj_set_scroll_dir(s_root, LV_DIR_VER);
+}
+
+static lv_obj_t *add_title(const char *txt)
+{
+    lv_obj_t *l = lv_label_create(s_root);
+    lv_obj_set_style_text_font(l, &font_fr_24, 0);
+    lv_obj_set_style_text_color(l, lv_color_hex(COL_TITLE), 0);
+    lv_label_set_long_mode(l, LV_LABEL_LONG_WRAP);
+    lv_obj_set_width(l, LV_PCT(100));
+    lv_label_set_text(l, txt);
+    return l;
+}
+
+static lv_obj_t *add_text(const char *txt)
+{
+    lv_obj_t *l = lv_label_create(s_root);
+    lv_obj_set_style_text_font(l, &font_fr_14, 0);
+    lv_obj_set_style_text_color(l, lv_color_hex(COL_TEXT), 0);
+    lv_label_set_long_mode(l, LV_LABEL_LONG_WRAP);
+    lv_obj_set_width(l, LV_PCT(100));
+    lv_label_set_text(l, txt);
+    return l;
+}
+
+static lv_obj_t *add_button(const char *txt, lv_event_cb_t cb, void *user)
+{
+    lv_obj_t *b = lv_button_create(s_root);
+    lv_obj_set_width(b, LV_PCT(100));
+    lv_obj_set_style_bg_color(b, lv_color_hex(COL_BTN), 0);
+    lv_obj_set_style_pad_all(b, 10, 0);
+    lv_obj_add_event_cb(b, cb, LV_EVENT_CLICKED, user);
+    lv_obj_t *l = lv_label_create(b);
+    lv_obj_set_style_text_font(l, &font_fr_14, 0);
+    lv_label_set_long_mode(l, LV_LABEL_LONG_WRAP);
+    lv_obj_set_width(l, LV_PCT(100));
+    lv_label_set_text(l, txt);
+    return b;
+}
+
+// Forward decls
+static void show_library(void);
+static void enter_passage(const char *pid);
+
+static void tile_cb(lv_event_t *e);
+static void choice_cb(lv_event_t *e);
+static void menu_cb(lv_event_t *e);
+
+// ── Library menu ────────────────────────────────────────────────────────────
+static void show_library(void)
+{
+    cmd_exec_stop_play();   // silence any passage narration
+    if (s_book) { cJSON_Delete(s_book); s_book = NULL; s_passages = NULL; }
+    ensure_root();
+    add_title("Choisis ton histoire");
+    for (int i = 0; i < s_lib_n; i++) {
+        const cJSON *t = cJSON_GetObjectItem(cJSON_GetArrayItem(s_lib, i), "title");
+        add_button(cJSON_IsString(t) ? t->valuestring : "?",
+                   tile_cb, (void *)(intptr_t)i);
+    }
+    lv_obj_scroll_to_y(s_root, 0, LV_ANIM_OFF);
+}
+
+// ── Passage ─────────────────────────────────────────────────────────────────
+static void enter_passage(const char *pid)
+{
+    const cJSON *p = cJSON_GetObjectItem(s_passages, pid);
+    if (!cJSON_IsObject(p)) { ESP_LOGW(TAG, "passage '%s' missing", pid); return; }
+    const cJSON *screen  = cJSON_GetObjectItem(p, "screen");
+    const cJSON *text    = cJSON_GetObjectItem(p, "text");
+    const cJSON *choices = cJSON_GetObjectItem(p, "choices");
+
+    ensure_root();
+    add_title(cJSON_IsString(screen) ? screen->valuestring : "");
+    add_text(cJSON_IsString(text) ? text->valuestring : "");
+
+    // Narration: play this passage's WAV from the SD if the pack has audio.
+    // Interruptible — tapping a choice cuts it and plays the next passage.
+    const cJSON *wav = cJSON_GetObjectItem(p, "wav");
+    if (cJSON_IsString(wav) && wav->valuestring[0]) {
+        char path[160];
+        snprintf(path, sizeof(path), "%s/%s", GB_DIR, wav->valuestring);
+        cmd_exec_play_file_async(path);
+    }
+    int n = cJSON_IsArray(choices) ? cJSON_GetArraySize(choices) : 0;
+    if (n == 0) {
+        add_button("Revenir au menu", menu_cb, NULL);
+    } else {
+        for (int i = 0; i < n; i++) {
+            const cJSON *c = cJSON_GetArrayItem(choices, i);
+            const cJSON *lbl = cJSON_GetObjectItem(c, "label");
+            const cJSON *g   = cJSON_GetObjectItem(c, "goto");
+            // goto valuestring stays valid while s_book is alive → use as user_data
+            add_button(cJSON_IsString(lbl) ? lbl->valuestring : "?",
+                       choice_cb, cJSON_IsString(g) ? g->valuestring : NULL);
+        }
+    }
+    lv_obj_scroll_to_y(s_root, 0, LV_ANIM_OFF);
+    ESP_LOGI(TAG, "passage '%s' (%d choices)", pid, n);
+}
+
+static void load_book(int idx)
+{
+    const cJSON *e = cJSON_GetArrayItem(s_lib, idx);
+    const cJSON *file = cJSON_GetObjectItem(e, "book");
+    if (!cJSON_IsString(file)) return;
+    char path[128];
+    snprintf(path, sizeof(path), "%s/%s", GB_DIR, file->valuestring);
+    cJSON *book = load_json(path);
+    if (!book) return;
+    const cJSON *passages = cJSON_GetObjectItem(book, "passages");
+    const cJSON *start = cJSON_GetObjectItem(book, "start");
+    if (!cJSON_IsObject(passages) || !cJSON_IsString(start)) {
+        cJSON_Delete(book); return;
+    }
+    if (s_book) cJSON_Delete(s_book);
+    s_book = book;
+    s_passages = (cJSON *)passages;
+    ESP_LOGI(TAG, "load book #%d @ '%s'", idx, start->valuestring);
+    enter_passage(start->valuestring);
+}
+
+// ── Touch callbacks (run inside the LVGL task → lock already held) ───────────
+static void tile_cb(lv_event_t *e)   { load_book((int)(intptr_t)lv_event_get_user_data(e)); }
+static void menu_cb(lv_event_t *e)   { (void)e; show_library(); }
+static void choice_cb(lv_event_t *e)
+{
+    const char *goto_id = (const char *)lv_event_get_user_data(e);
+    if (goto_id) enter_passage(goto_id);
+}
+
+// ── Public init ─────────────────────────────────────────────────────────────
+void gamebook_init(void)
+{
+    cJSON_Hooks hooks = { .malloc_fn = gb_malloc, .free_fn = gb_free };
+    cJSON_InitHooks(&hooks);   // keep the big book trees in PSRAM
+
+    if (bsp_sdcard_mount() != ESP_OK) {
+        ESP_LOGW(TAG, "no SD card — gamebook disabled");
+        return;
+    }
+    s_lib_root = load_json(GB_LIBRARY);
+    if (!s_lib_root) { ESP_LOGW(TAG, "no library.json — gamebook disabled"); return; }
+    s_lib = cJSON_GetObjectItem(s_lib_root, "library");
+    if (!cJSON_IsArray(s_lib) || cJSON_GetArraySize(s_lib) == 0) {
+        cJSON_Delete(s_lib_root); s_lib_root = NULL; s_lib = NULL;
+        ESP_LOGW(TAG, "empty library");
+        return;
+    }
+    s_lib_n = cJSON_GetArraySize(s_lib);
+
+    bsp_display_brightness_set(80);
+    if (bsp_display_lock(1000)) {
+        show_library();
+        bsp_display_unlock();
+    }
+    ESP_LOGI(TAG, "touch gamebook up (%d stories)", s_lib_n);
+}

box3_voice/main/gamebook.h

@@ -0,0 +1,27 @@
+#pragma once
+// gamebook — "livre dont tu es le héros" TACTILE pour l'ESP32-S3-BOX-3.
+//
+// Écran 320x240 + tactile : on lit le passage à l'écran et on TOUCHE un bouton
+// pour choisir. Réutilise le pack gamebook du master sur la SD :
+//   /sdcard/gamebook/{library.json, <id>.json}
+// où chaque passage porte {screen, text, choices:[{label, goto}]} (le champ
+// "wav" éventuel est ignoré — version texte/tactile, pas d'audio requis).
+//
+// S'appuie sur la stack LVGL déjà démarrée par bsp_display_start() : on
+// construit une UUI (menu d'histoires en liste, puis pages tactiles) et la
+// navigation se fait dans les callbacks de boutons LVGL. 100% local, hors-ligne.
+
+#include <stdbool.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Monte la SD, charge la bibliothèque et affiche le menu tactile des histoires.
+// À appeler une fois depuis app_main, après bsp_display_start() et le serveur
+// de fichiers (pour que la SD soit accessible). No-op si aucun pack présent.
+void gamebook_init(void);
+
+#ifdef __cplusplus
+}
+#endif

box3_voice/main/main.c

@@ -28,8 +28,10 @@
 #include "scenario_server.h"
 #include "plip_virtual.h"
 #include "plip_ui.h"
+#include "gamebook.h"
 #include "stimulus.h"
 #include "scenario_mesh.h"
+#include "cmd_exec.h"
 
 /* BSP header — provided by espressif/esp-box component */
 #include "bsp/esp-bsp.h"
@@ -38,8 +40,12 @@ static const char *TAG = "zacus-voice";
 
 /* --------------- Shared state --------------- */
 
-static i2s_chan_handle_t s_spk_handle = NULL;   /* Speaker TX channel */
-static i2s_chan_handle_t s_mic_handle = NULL;   /* Microphone RX channel */
+/* Audio now goes through the ES8311 (speaker) / ES7210 (mic) codecs via the
+ * BSP + esp_codec_dev. bsp_audio_init() owns the single shared I2S port; we
+ * must NOT create our own raw I2S channels (that left the ES8311 DAC unpowered
+ * and the speaker silent). */
+static esp_codec_dev_handle_t s_spk_codec = NULL;   /* ES8311 output (DAC) */
+static esp_codec_dev_handle_t s_mic_codec = NULL;   /* ES7210 input (ADC)  */
 static volatile bool     s_wifi_connected = false;
 static volatile bool     s_voice_streaming = false;
 
@@ -112,11 +118,10 @@ static esp_err_t wifi_init_sta(void)
 // generator can sequence a melody for the master's microphone.
 void audio_play_tone(float frequency, int duration_ms)
 {
-    if (!s_spk_handle || duration_ms <= 0) return;
+    if (!s_spk_codec || duration_ms <= 0) return;
     const int total_samples = AUDIO_SAMPLE_RATE * duration_ms / 1000;
     const float amplitude = 16000.0f;
     int16_t buffer[256];
-    size_t bytes_written = 0;
     int sample_idx = 0;
     while (sample_idx < total_samples) {
         int chunk = (total_samples - sample_idx < 256)
@@ -125,29 +130,27 @@ void audio_play_tone(float frequency, int duration_ms)
             float t = (float)(sample_idx + i) / (float)AUDIO_SAMPLE_RATE;
             buffer[i] = (int16_t)(amplitude * sinf(2.0f * M_PI * frequency * t));
         }
-        i2s_channel_write(s_spk_handle, buffer, chunk * sizeof(int16_t),
-                          &bytes_written, portMAX_DELAY);
+        esp_codec_dev_write(s_spk_codec, buffer, chunk * sizeof(int16_t));
         sample_idx += chunk;
     }
 }
 
 static void audio_test_tone(void)
 {
-    if (!s_spk_handle) {
+    if (!s_spk_codec) {
         ESP_LOGW(TAG, "Speaker not initialized, skipping test tone");
         return;
     }
 
     ESP_LOGI(TAG, "Playing 440 Hz test tone (1 second)...");
 
-    /* Generate 1 second of 440 Hz sine wave using persistent speaker handle */
+    /* Generate 1 second of 440 Hz sine wave through the ES8311 codec */
     const int duration_ms = 1000;
     const int total_samples = AUDIO_SAMPLE_RATE * duration_ms / 1000;
     const float frequency = 440.0f;
     const float amplitude = 16000.0f;
 
     int16_t buffer[256];
-    size_t bytes_written = 0;
     int sample_idx = 0;
 
     while (sample_idx < total_samples) {
@@ -156,7 +159,7 @@ static void audio_test_tone(void)
             float t = (float)(sample_idx + i) / (float)AUDIO_SAMPLE_RATE;
             buffer[i] = (int16_t)(amplitude * sinf(2.0f * M_PI * frequency * t));
         }
-        i2s_channel_write(s_spk_handle, buffer, chunk * sizeof(int16_t), &bytes_written, portMAX_DELAY);
+        esp_codec_dev_write(s_spk_codec, buffer, chunk * sizeof(int16_t));
         sample_idx += chunk;
     }
 
@@ -172,15 +175,13 @@ static void tts_audio_callback(const uint8_t *data, size_t len)
      * Called from WebSocket event context with incoming TTS PCM16 data.
      * Write directly to the speaker I2S channel.
      */
-    if (!s_spk_handle || len == 0) {
+    if (!s_spk_codec || len == 0) {
         return;
     }
 
-    size_t bytes_written = 0;
-    esp_err_t ret = i2s_channel_write(s_spk_handle, data, len,
-                                       &bytes_written, pdMS_TO_TICKS(200));
-    if (ret != ESP_OK) {
-        ESP_LOGW(TAG, "TTS write to speaker failed: %s", esp_err_to_name(ret));
+    int ret = esp_codec_dev_write(s_spk_codec, (void *)data, (int)len);
+    if (ret != ESP_CODEC_DEV_OK) {
+        ESP_LOGW(TAG, "TTS write to speaker failed: %d", ret);
     }
 }
 
@@ -188,34 +189,42 @@ static void tts_audio_callback(const uint8_t *data, size_t len)
 
 static esp_err_t speaker_init(void)
 {
-    /* Enable power amplifier */
+    /* Init the ES8311 output codec via the BSP. This also runs bsp_audio_init()
+     * (shared duplex I2S port) + bsp_i2c_init() and configures the power
+     * amplifier (BSP_POWER_AMP_IO = GPIO46). The ES8311 DAC is configured over
+     * I2C here — without this the speaker stays muted no matter what we push on
+     * I2S. */
+    s_spk_codec = bsp_audio_codec_speaker_init();
+    if (!s_spk_codec) {
+        ESP_LOGE(TAG, "bsp_audio_codec_speaker_init failed — speaker unavailable");
+        return ESP_FAIL;
+    }
+
+    /* All speaker audio in this app is 16 kHz / 16-bit / mono PCM. */
+    esp_codec_dev_sample_info_t fs = {
+        .bits_per_sample = AUDIO_BITS,
+        .channel         = AUDIO_CHANNELS,
+        .channel_mask    = 0,
+        .sample_rate     = AUDIO_SAMPLE_RATE,
+        .mclk_multiple   = 0,
+    };
+    int ret = esp_codec_dev_open(s_spk_codec, &fs);
+    if (ret != ESP_CODEC_DEV_OK) {
+        ESP_LOGE(TAG, "esp_codec_dev_open (speaker) failed: %d", ret);
+        return ESP_FAIL;
+    }
+
+    /* Output volume 0..100. */
+    esp_codec_dev_set_out_vol(s_spk_codec, 100);
+
+    /* Belt-and-suspenders: the es8311 driver is configured with pa_pin=GPIO46
+     * (pa_reverted=false) and should raise it on open, but the speaker stays
+     * silent in practice — force the power amplifier on explicitly. */
     gpio_set_direction(BOX3_PA_ENABLE, GPIO_MODE_OUTPUT);
     gpio_set_level(BOX3_PA_ENABLE, 1);
 
-    i2s_chan_config_t chan_cfg = I2S_CHANNEL_DEFAULT_CONFIG(I2S_NUM_0, I2S_ROLE_MASTER);
-    ESP_ERROR_CHECK(i2s_new_channel(&chan_cfg, &s_spk_handle, NULL));
-
-    i2s_std_config_t std_cfg = {
-        .clk_cfg  = I2S_STD_CLK_DEFAULT_CONFIG(AUDIO_SAMPLE_RATE),
-        .slot_cfg = I2S_STD_PHILIPS_SLOT_DEFAULT_CONFIG(I2S_DATA_BIT_WIDTH_16BIT, I2S_SLOT_MODE_MONO),
-        .gpio_cfg = {
-            .mclk = BOX3_I2S_MCLK,
-            .bclk = BOX3_I2S_BCLK,
-            .ws   = BOX3_I2S_WS,
-            .dout = BOX3_I2S_DOUT,
-            .din  = I2S_GPIO_UNUSED,
-            .invert_flags = {
-                .mclk_inv = false,
-                .bclk_inv = false,
-                .ws_inv   = false,
-            },
-        },
-    };
-
-    ESP_ERROR_CHECK(i2s_channel_init_std_mode(s_spk_handle, &std_cfg));
-    ESP_ERROR_CHECK(i2s_channel_enable(s_spk_handle));
-
-    ESP_LOGI(TAG, "Speaker I2S initialized (I2S_NUM_0)");
+    ESP_LOGI(TAG, "Speaker codec (ES8311) initialized @ %d Hz, vol=100, PA on",
+             AUDIO_SAMPLE_RATE);
     return ESP_OK;
 }
 
@@ -225,43 +234,45 @@ static void mic_monitor_task(void *arg)
 {
     ESP_LOGI(TAG, "Starting mic capture task...");
 
-    i2s_chan_config_t chan_cfg = I2S_CHANNEL_DEFAULT_CONFIG(I2S_NUM_1, I2S_ROLE_MASTER);
-    ESP_ERROR_CHECK(i2s_new_channel(&chan_cfg, NULL, &s_mic_handle));
+    /* Init the ES7210 input codec via the BSP. It shares the same I2S port
+     * (already created by speaker_init -> bsp_audio_init); creating a separate
+     * raw I2S channel here would conflict with that shared port. */
+    s_mic_codec = bsp_audio_codec_microphone_init();
+    if (!s_mic_codec) {
+        ESP_LOGE(TAG, "bsp_audio_codec_microphone_init failed — mic unavailable");
+        vTaskDelete(NULL);
+        return;
+    }
 
-    i2s_std_config_t std_cfg = {
-        .clk_cfg  = I2S_STD_CLK_DEFAULT_CONFIG(AUDIO_SAMPLE_RATE),
-        .slot_cfg = I2S_STD_PHILIPS_SLOT_DEFAULT_CONFIG(I2S_DATA_BIT_WIDTH_16BIT, I2S_SLOT_MODE_MONO),
-        .gpio_cfg = {
-            .mclk = I2S_GPIO_UNUSED,
-            .bclk = BOX3_I2S_BCLK,
-            .ws   = BOX3_I2S_WS,
-            .dout = I2S_GPIO_UNUSED,
-            .din  = BOX3_I2S_DIN,
-            .invert_flags = {
-                .mclk_inv = false,
-                .bclk_inv = false,
-                .ws_inv   = false,
-            },
-        },
+    esp_codec_dev_sample_info_t fs = {
+        .bits_per_sample = AUDIO_BITS,
+        .channel         = AUDIO_CHANNELS,
+        .channel_mask    = 0,
+        .sample_rate     = AUDIO_SAMPLE_RATE,
+        .mclk_multiple   = 0,
     };
-
-    ESP_ERROR_CHECK(i2s_channel_init_std_mode(s_mic_handle, &std_cfg));
-    ESP_ERROR_CHECK(i2s_channel_enable(s_mic_handle));
+    int oret = esp_codec_dev_open(s_mic_codec, &fs);
+    if (oret != ESP_CODEC_DEV_OK) {
+        ESP_LOGE(TAG, "esp_codec_dev_open (mic) failed: %d", oret);
+        vTaskDelete(NULL);
+        return;
+    }
+    /* ES7210 input gain (dB). 30 dB is a sane default for the BOX-3 mic array. */
+    esp_codec_dev_set_in_gain(s_mic_codec, 30.0f);
 
     int16_t buffer[AUDIO_FRAME_SAMPLES];
-    size_t bytes_read = 0;
     int rms_log_counter = 0;
 
     while (1) {
-        esp_err_t ret = i2s_channel_read(s_mic_handle, buffer,
-                                          AUDIO_FRAME_SAMPLES * sizeof(int16_t),
-                                          &bytes_read, pdMS_TO_TICKS(1000));
-        if (ret != ESP_OK) {
-            ESP_LOGW(TAG, "Mic read error: %s", esp_err_to_name(ret));
+        int ret = esp_codec_dev_read(s_mic_codec, buffer,
+                                     AUDIO_FRAME_SAMPLES * sizeof(int16_t));
+        if (ret != ESP_CODEC_DEV_OK) {
+            ESP_LOGW(TAG, "Mic read error: %d", ret);
+            vTaskDelay(pdMS_TO_TICKS(20));
             continue;
         }
 
-        int samples = bytes_read / sizeof(int16_t);
+        int samples = AUDIO_FRAME_SAMPLES;
 
         /* If voice streaming is active, send PCM to bridge */
         if (s_voice_streaming) {
@@ -372,6 +383,28 @@ static void button_task(void *arg)
     }
 }
 
+/* --------------- ESP-NOW CMD text callback --------------- */
+
+// Called from the scenario_mesh text worker task for every CMD/EVT frame
+// received via ESP-NOW. On the BOX-3 we only care about CMD (master→us).
+static void on_mesh_text(uint8_t kind, const uint8_t src_mac[6], const char *text)
+{
+    if (kind == SCENARIO_MESH_TEXT_CMD) {
+        ESP_LOGI(TAG, "ESP-NOW CMD from %02x:%02x:%02x:%02x:%02x:%02x: %.80s",
+                 src_mac[0], src_mac[1], src_mac[2],
+                 src_mac[3], src_mac[4], src_mac[5], text);
+        esp_err_t err = cmd_exec_handle(text, strlen(text));
+        if (err != ESP_OK && err != ESP_ERR_NOT_SUPPORTED) {
+            ESP_LOGW(TAG, "cmd_exec_handle: %s", esp_err_to_name(err));
+        }
+    } else {
+        // EVT arriving at the BOX-3 (unusual — master echoing back).
+        ESP_LOGI(TAG, "ESP-NOW EVT from %02x:%02x:%02x:%02x:%02x:%02x: %.80s",
+                 src_mac[0], src_mac[1], src_mac[2],
+                 src_mac[3], src_mac[4], src_mac[5], text);
+    }
+}
+
 /* --------------- Application entry point --------------- */
 
 void app_main(void)
@@ -399,11 +432,19 @@ void app_main(void)
     /* Initialize persistent speaker output (used for test tone + TTS playback) */
     speaker_init();
 
+    /* Pass speaker codec handle to CMD executor for real WAV playback */
+    cmd_exec_set_spk_handle(s_spk_codec);
+
     /* Play test tone to verify audio output */
     audio_test_tone();
 
-    /* Start mic capture task (reads I2S mic, streams to bridge when active) */
+#if CONFIG_BOX3_VOICE_STREAMING
+    /* Start mic capture task (reads I2S mic, streams to bridge when active).
+     * Off by default: as a touch gamebook the BOX-3 needs no mic, and the
+     * continuous codec reads + bridge reconnects interfered with the speaker
+     * (the first narration clip came out silent) and spammed the log. */
     xTaskCreatePinnedToCore(mic_monitor_task, "mic_capture", 4096, NULL, 5, NULL, 1);
+#endif
 
     /* Start button handler (BOOT button toggles voice streaming) */
     xTaskCreate(button_task, "button", 2048, NULL, 4, NULL);
@@ -412,8 +453,10 @@ void app_main(void)
     ESP_LOGI(TAG, "Connecting to WiFi...");
     wifi_init_sta();
 
+#if CONFIG_BOX3_VOICE_STREAMING
     /* Start voice bridge connection task (waits for WiFi, then connects WS) */
     xTaskCreate(voice_bridge_task, "voice_bridge", 6144, NULL, 5, NULL);
+#endif
 
     /* Start the scenario hot-load HTTP server (POST /game/scenario).
      * httpd_start binds to all netifs — works as soon as the WiFi STA has an IP. */
@@ -423,7 +466,7 @@ void app_main(void)
         /* Phone-less PLIP annex: same REST contract as PLIP_FIRMWARE
          * (POST /ring /stop /play, GET /status), ring on the speaker,
          * BOOT button as the virtual hook switch. */
-        if (plip_virtual_init(scenario_server_handle(), s_spk_handle) != ESP_OK) {
+        if (plip_virtual_init(scenario_server_handle(), s_spk_codec) != ESP_OK) {
             ESP_LOGW(TAG, "plip_virtual_init failed — virtual phone unavailable");
         } else if (plip_ui_init() != ESP_OK) {
             /* REST/ESP-NOW phone still works headless if the UI fails. */
@@ -437,6 +480,11 @@ void app_main(void)
         } else {
             ESP_LOGW(TAG, "stimulus_init failed — QR/melody generator off");
         }
+
+        /* Touch gamebook: if a story pack is on the SD (/sdcard/gamebook/),
+         * take over the screen with a tap-to-choose "livre dont tu es le
+         * héros". No-op (and the phone UI stays) when no pack is present. */
+        gamebook_init();
     }
 
     /* Start the ESP-NOW receiver so the master can relay scenarios to us even
@@ -450,6 +498,15 @@ void app_main(void)
                  esp_err_to_name(mesh_err));
     } else {
         ESP_LOGI(TAG, "ESP-NOW scenario receiver active");
+        /* Register CMD text callback so the master can drive screen / audio / leds
+         * via POST /game/espnow/cmd {peer, command:<json>} (D5 contract). */
+        esp_err_t tcb_err = scenario_mesh_set_text_cb(on_mesh_text);
+        if (tcb_err != ESP_OK) {
+            ESP_LOGW(TAG, "scenario_mesh_set_text_cb: %s — CMD executor unavailable",
+                     esp_err_to_name(tcb_err));
+        } else {
+            ESP_LOGI(TAG, "ESP-NOW CMD executor registered (op: screen/play/evt/led)");
+        }
     }
 
     /* TODO: Initialize ESP-SR WakeNet for wake-word detection

box3_voice/main/plip_virtual.c

@@ -38,7 +38,7 @@ typedef struct {
     char reason[32];  /* "pickup" | "hangup" | "ring_timeout" | "dial:<n>" */
 } hook_event_t;
 
-static i2s_chan_handle_t        s_spk;
+static esp_codec_dev_handle_t   s_spk;
 static volatile plip_hook_state_t s_state = PLIP_HOOK_ON;
 static volatile bool            s_ring_stop = false;
 static TaskHandle_t             s_ring_task = NULL;
@@ -107,7 +107,6 @@ static void ring_burst(int duration_ms)
 {
     const int total = AUDIO_SAMPLE_RATE * duration_ms / 1000;
     int16_t buffer[256];
-    size_t written = 0;
     int idx = 0;
     while (idx < total && !s_ring_stop) {
         int chunk = (total - idx < 256) ? (total - idx) : 256;
@@ -116,8 +115,9 @@ static void ring_burst(int duration_ms)
             buffer[i] = (int16_t) (RING_AMPLITUDE *
                                    sinf(2.0f * (float) M_PI * RING_FREQ_HZ * t));
         }
-        i2s_channel_write(s_spk, buffer, chunk * sizeof(int16_t), &written,
-                          pdMS_TO_TICKS(500));
+        if (s_spk) {
+            esp_codec_dev_write(s_spk, buffer, chunk * sizeof(int16_t));
+        }
         idx += chunk;
     }
 }
@@ -279,7 +279,7 @@ plip_hook_state_t plip_virtual_state(void)
     return s_state;
 }
 
-esp_err_t plip_virtual_init(httpd_handle_t server, i2s_chan_handle_t spk)
+esp_err_t plip_virtual_init(httpd_handle_t server, esp_codec_dev_handle_t spk)
 {
     if (!server || !spk) return ESP_ERR_INVALID_ARG;
     s_spk = spk;

box3_voice/main/plip_virtual.h

@@ -16,7 +16,7 @@
 #pragma once
 
 #include <stdbool.h>
-#include "driver/i2s_std.h"
+#include "esp_codec_dev.h"
 #include "esp_err.h"
 #include "esp_http_server.h"
 
@@ -34,7 +34,7 @@ typedef enum {
 // Register the REST handlers on `server` and keep `spk` for the ring tone.
 // Call once after scenario_server_start(); the speaker channel must be
 // enabled (speaker_init() in main.c).
-esp_err_t plip_virtual_init(httpd_handle_t server, i2s_chan_handle_t spk);
+esp_err_t plip_virtual_init(httpd_handle_t server, esp_codec_dev_handle_t spk);
 
 // Forward a BOOT-button press. Returns true when the press was consumed as
 // a hook transition (pickup/hangup) — the caller should then skip its own

box3_voice/main/scenario_server.c

@@ -14,6 +14,7 @@
 #include <unistd.h>
 
 #include "cJSON.h"
+#include "cmd_exec.h"
 #include "esp_err.h"
 #include "esp_http_server.h"
 #include "esp_log.h"
@@ -22,6 +23,9 @@
 #include "freertos/FreeRTOS.h"
 #include "freertos/task.h"
 
+/* BSP SD card support (bsp_sdcard_mount, BSP_SD_MOUNT_POINT) */
+#include "bsp/esp-bsp.h"
+
 #define TAG "scenario_srv"
 
 #define MAX_SCENARIO_BYTES (64 * 1024)
@@ -216,6 +220,132 @@ static esp_err_t handle_scenario_post(httpd_req_t *req) {
     return send_json(req, "200 OK", buf);
 }
 
+// ---------- POST /game/cmd ----------
+//
+// WiFi-direct CMD endpoint: receives {op,a} JSON frames (≤512 bytes) and
+// forwards them to cmd_exec_handle(). Replaces ESP-NOW CMD path for annexes.
+
+#define MAX_CMD_BYTES 512
+
+static esp_err_t handle_cmd_post(httpd_req_t *req) {
+    if (req->content_len <= 0 || req->content_len > MAX_CMD_BYTES) {
+        ESP_LOGW(TAG, "POST /game/cmd: bad content_len=%d", (int)req->content_len);
+        return send_error(req, "400 Bad Request", "body must be 1..512 bytes");
+    }
+    char body[MAX_CMD_BYTES + 1];
+    int total = 0;
+    while (total < (int)req->content_len) {
+        int got = httpd_req_recv(req, body + total, req->content_len - total);
+        if (got <= 0) {
+            if (got == HTTPD_SOCK_ERR_TIMEOUT) continue;
+            return send_error(req, "400 Bad Request", "recv failed");
+        }
+        total += got;
+    }
+    body[total] = '\0';
+
+    ESP_LOGI(TAG, "POST /game/cmd (%d B): %.*s", total, total < 80 ? total : 80, body);
+    esp_err_t err = cmd_exec_handle(body, (size_t)total);
+    if (err == ESP_ERR_INVALID_ARG) {
+        return send_error(req, "400 Bad Request", "malformed cmd json or missing op");
+    }
+    return send_json(req, "200 OK", "{\"ok\":true}");
+}
+
+// ---------- POST /game/file ----------
+//
+// Write binary assets directly to the SD card over HTTP.
+// Query param: ?path=sd/<relative/path>  (only "sd/" prefix accepted)
+// Route: sd/<…> → /sdcard/<…>  (mkdir -p, up to 8 MiB, anti-traversal).
+// Returns 503 if SD is not mounted, 403 on bad path, 200 JSON on success.
+
+#define MAX_FILE_BYTES (8L * 1024L * 1024L)
+
+static bool s_sd_mounted = false;
+
+static esp_err_t ensure_sd_mounted(void) {
+    if (s_sd_mounted) return ESP_OK;
+    esp_err_t ret = bsp_sdcard_mount();
+    if (ret == ESP_OK || ret == ESP_ERR_INVALID_STATE /* already mounted */) {
+        s_sd_mounted = true;
+        ESP_LOGI(TAG, "file: SD mounted at %s", BSP_SD_MOUNT_POINT);
+        return ESP_OK;
+    }
+    ESP_LOGW(TAG, "file: SD mount failed: %s", esp_err_to_name(ret));
+    return ret;
+}
+
+static esp_err_t handle_file_post(httpd_req_t *req) {
+    char query[160], path_param[96];
+    if (httpd_req_get_url_query_str(req, query, sizeof(query)) != ESP_OK ||
+        httpd_query_key_value(query, "path", path_param,
+                              sizeof(path_param)) != ESP_OK) {
+        return send_error(req, "400 Bad Request", "missing path param");
+    }
+
+    // Only accept sd/<…> prefix; reject traversal and bare directory paths.
+    if (strncmp(path_param, "sd/", 3) != 0 ||
+        strstr(path_param, "..") ||
+        path_param[strlen(path_param) - 1] == '/') {
+        return send_error(req, "403 Forbidden", "path must start with sd/ and name a file");
+    }
+
+    if (req->content_len <= 0 || (long)req->content_len > MAX_FILE_BYTES) {
+        return send_error(req, "400 Bad Request", "file too large or empty");
+    }
+
+    if (ensure_sd_mounted() != ESP_OK) {
+        return send_error(req, "503 Service Unavailable", "sd_not_mounted");
+    }
+
+    // Build absolute destination: /sdcard/<…> (strip the "sd/" prefix).
+    char full[192];
+    snprintf(full, sizeof(full), "%s/%s", BSP_SD_MOUNT_POINT, path_param + 3);
+
+    // mkdir -p for all intermediate directories.
+    const size_t root_len = strlen(BSP_SD_MOUNT_POINT) + 1; // e.g. strlen("/sdcard/")
+    for (char *p = full + root_len; *p; p++) {
+        if (*p == '/') {
+            *p = '\0';
+            mkdir(full, 0775);   // EEXIST is fine
+            *p = '/';
+        }
+    }
+
+    FILE *f = fopen(full, "wb");
+    if (!f) {
+        ESP_LOGE(TAG, "file: fopen %s failed (errno=%d)", full, errno);
+        return send_error(req, "500 Internal Server Error", "open failed");
+    }
+
+    char buf[1024];
+    int remaining = (int)req->content_len;
+    while (remaining > 0) {
+        const int want = remaining < (int)sizeof(buf) ? remaining : (int)sizeof(buf);
+        int got = httpd_req_recv(req, buf, want);
+        if (got <= 0) {
+            if (got == HTTPD_SOCK_ERR_TIMEOUT) continue;
+            fclose(f);
+            unlink(full);
+            return send_error(req, "400 Bad Request", "recv failed");
+        }
+        if (fwrite(buf, 1, (size_t)got, f) != (size_t)got) {
+            fclose(f);
+            unlink(full);
+            return send_error(req, "500 Internal Server Error", "write failed");
+        }
+        remaining -= got;
+    }
+    fclose(f);
+
+    char resp[192];
+    snprintf(resp, sizeof(resp),
+             "{\"status\":\"ok\",\"path\":\"%s\",\"bytes\":%d}",
+             path_param, (int)req->content_len);
+    ESP_LOGI(TAG, "file: stored %s (%d B)", full, (int)req->content_len);
+    return send_json(req, "200 OK", resp);
+}
+
 // ---------- public init ----------
 
 httpd_handle_t scenario_server_handle(void) {
@@ -229,7 +359,7 @@ esp_err_t scenario_server_start(void) {
     }
     httpd_config_t cfg = HTTPD_DEFAULT_CONFIG();
     cfg.server_port = 80;
-    cfg.max_uri_handlers = 8;
+    cfg.max_uri_handlers = 12;
     cfg.stack_size = 8192;
 
     esp_err_t err = httpd_start(&s_server, &cfg);
@@ -246,9 +376,19 @@ esp_err_t scenario_server_start(void) {
         .uri = "/game/scenario", .method = HTTP_POST,
         .handler = handle_scenario_post, .user_ctx = NULL,
     };
+    static const httpd_uri_t uri_file = {
+        .uri = "/game/file", .method = HTTP_POST,
+        .handler = handle_file_post, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_cmd = {
+        .uri = "/game/cmd", .method = HTTP_POST,
+        .handler = handle_cmd_post, .user_ctx = NULL,
+    };
     httpd_register_uri_handler(s_server, &uri_healthz);
     httpd_register_uri_handler(s_server, &uri_scenario);
+    httpd_register_uri_handler(s_server, &uri_file);
+    httpd_register_uri_handler(s_server, &uri_cmd);
 
-    ESP_LOGI(TAG, "scenario server up on :80 (GET /healthz, POST /game/scenario)");
+    ESP_LOGI(TAG, "scenario server up on :80 (GET /healthz, POST /game/scenario, POST /game/file, POST /game/cmd)");
     return ESP_OK;
 }

box3_voice/sdkconfig.defaults

@@ -51,6 +51,12 @@ CONFIG_SPIRAM_MALLOC_RESERVE_INTERNAL=32768
 # CONFIG_ZACUS_WIFI_SSID="..."
 # CONFIG_ZACUS_WIFI_PASSWORD="..."
 
+# WiFi channel: must match the master ESP32's connected channel for ESP-NOW
+# co-channel operation. The master (idf_zacus) connects to channel 11 on the
+# lab multi-AP "Les cils" network. Set to 0 to scan all channels and connect
+# to whichever AP responds first (may differ from master → ESP-NOW will break).
+# CONFIG_ZACUS_WIFI_CHANNEL=11
+
 # esp-box BSP uses the legacy i2c driver; esp_codec_dev uses driver_ng.
 # Force-enable legacy mode so both coexist (and prevent the runtime abort
 # in bsp_display_start when both drivers race to register on the same bus).

idf_zacus/components/display_ui/display_ui.cpp

@@ -248,6 +248,30 @@ static bool     s_browser_isdir[BROWSER_MAX_ENTRIES];
 static uint8_t  s_browser_count = 0;
 static volatile uint8_t s_browser_sel = 0;
 static volatile bool s_browser_open = false;
+
+// ── Gamebook view ("livre dont vous êtes le héros") ─────────────────────────
+static lv_obj_t *s_scr_gamebook;
+static lv_obj_t *s_gb_title;          // page title, top
+static lv_obj_t *s_gb_body;           // full passage text, wrapped
+static lv_obj_t *s_gb_menu;           // choice lines, bottom
+static volatile bool s_gamebook_open = false;
+static lv_obj_t *s_gb_body_cont;      // scrollable container holding s_gb_body
+static char s_gb_title_buf[64];
+static char s_gb_body_buf[1100];
+static char s_gb_menu_buf[160];
+static volatile int  s_gb_scroll_req = 0;   // pending scroll: +pages down / -pages up
+static volatile bool s_gb_scroll_home = false;  // reset scroll to top on new passage
+
+// ── Gamebook library (tile grid) ────────────────────────────────────────────
+#define LIB_TILES 6
+static lv_obj_t *s_scr_library;
+static lv_obj_t *s_lib_tile[LIB_TILES];
+static lv_obj_t *s_lib_label[LIB_TILES];
+static volatile bool s_library_open = false;
+static char s_lib_title_buf[LIB_TILES][48];
+static int  s_lib_count_buf = 0;
+static int  s_lib_sel_buf   = 0;
+
 static volatile bool s_browser_reload = false;
 
 // ─── Intro — faithful port of the original cracktro ──────────────────────────
@@ -761,6 +785,93 @@ scroller:
     lv_obj_set_pos(s_intro_scroll, sx, sy);
 }
 
+static void build_library_screen(void) {
+    s_scr_library = lv_obj_create(NULL);
+    lv_obj_set_style_bg_color(s_scr_library, lv_color_hex(0x000000), 0);
+    lv_obj_set_style_bg_opa(s_scr_library, LV_OPA_COVER, 0);
+    lv_obj_clear_flag(s_scr_library, LV_OBJ_FLAG_SCROLLABLE);
+
+    lv_obj_t *hdr = lv_label_create(s_scr_library);
+    lv_obj_set_style_text_color(hdr, lv_color_hex(COL_CODE), 0);
+    lv_obj_set_style_text_font(hdr, &lv_font_montserrat_14, 0);
+    lv_label_set_text(hdr, "Choisis ton histoire");
+    lv_obj_align(hdr, LV_ALIGN_TOP_MID, 0, 4);
+
+    for (int i = 0; i < LIB_TILES; i++) {
+        int col = i % 2, row = i / 2;
+        lv_obj_t *t = lv_obj_create(s_scr_library);
+        lv_obj_set_size(t, 228, 84);
+        lv_obj_set_pos(t, 8 + col * 236, 26 + row * 94);
+        lv_obj_set_style_radius(t, 8, 0);
+        lv_obj_set_style_bg_color(t, lv_color_hex(0x101820), 0);
+        lv_obj_set_style_bg_opa(t, LV_OPA_COVER, 0);
+        lv_obj_set_style_border_width(t, 2, 0);
+        lv_obj_set_style_border_color(t, lv_color_hex(0x445566), 0);
+        lv_obj_set_style_pad_all(t, 6, 0);
+        lv_obj_clear_flag(t, LV_OBJ_FLAG_SCROLLABLE);
+        lv_obj_t *l = lv_label_create(t);
+        lv_obj_set_style_text_color(l, lv_color_hex(COL_VALUE), 0);
+        lv_obj_set_style_text_font(l, &lv_font_montserrat_14, 0);
+        lv_label_set_long_mode(l, LV_LABEL_LONG_WRAP);
+        lv_obj_set_width(l, 208);
+        lv_obj_set_style_text_align(l, LV_TEXT_ALIGN_CENTER, 0);
+        lv_label_set_text(l, "");
+        lv_obj_center(l);
+        s_lib_tile[i] = t;
+        s_lib_label[i] = l;
+    }
+}
+
+static void build_gamebook_screen(void) {
+    s_scr_gamebook = lv_obj_create(NULL);
+    lv_obj_set_style_bg_color(s_scr_gamebook, lv_color_hex(0x000000), 0);
+    lv_obj_set_style_bg_opa(s_scr_gamebook, LV_OPA_COVER, 0);
+
+    // Compact title line at the very top.
+    s_gb_title = lv_label_create(s_scr_gamebook);
+    lv_obj_set_style_text_color(s_gb_title, lv_color_hex(COL_CODE), 0);
+    lv_obj_set_style_text_font(s_gb_title, &lv_font_montserrat_14, 0);
+    lv_label_set_long_mode(s_gb_title, LV_LABEL_LONG_DOT);
+    lv_obj_set_width(s_gb_title, DUI_HOR_RES - 12);
+    lv_obj_set_style_text_align(s_gb_title, LV_TEXT_ALIGN_CENTER, 0);
+    lv_label_set_text(s_gb_title, "");
+    lv_obj_align(s_gb_title, LV_ALIGN_TOP_MID, 0, 4);
+
+    // Choice selector: a single line pinned at the very bottom. Left/Right cycle
+    // the selection, click confirms. Drawn last (foreground).
+    s_gb_menu = lv_label_create(s_scr_gamebook);
+    lv_obj_set_style_text_color(s_gb_menu, lv_color_hex(COL_CODE), 0);
+    lv_obj_set_style_bg_color(s_gb_menu, lv_color_hex(0x182230), 0);
+    lv_obj_set_style_bg_opa(s_gb_menu, LV_OPA_COVER, 0);
+    lv_obj_set_style_pad_all(s_gb_menu, 5, 0);
+    lv_obj_set_style_text_font(s_gb_menu, &lv_font_montserrat_14, 0);
+    lv_label_set_long_mode(s_gb_menu, LV_LABEL_LONG_DOT);
+    lv_obj_set_width(s_gb_menu, DUI_HOR_RES);
+    lv_obj_set_style_text_align(s_gb_menu, LV_TEXT_ALIGN_CENTER, 0);
+    lv_label_set_text(s_gb_menu, "");
+    lv_obj_align(s_gb_menu, LV_ALIGN_BOTTOM_MID, 0, 0);
+
+    // Reading area: a vertically-scrollable container filling the whole height
+    // between the title and the choice line. Up/Down scroll it by half-pages.
+    s_gb_body_cont = lv_obj_create(s_scr_gamebook);
+    lv_obj_set_style_bg_color(s_gb_body_cont, lv_color_hex(0x000000), 0);
+    lv_obj_set_style_bg_opa(s_gb_body_cont, LV_OPA_COVER, 0);
+    lv_obj_set_style_border_width(s_gb_body_cont, 0, 0);
+    lv_obj_set_style_pad_all(s_gb_body_cont, 6, 0);
+    lv_obj_set_size(s_gb_body_cont, DUI_HOR_RES, 264);   // 24 (title) .. 288 (menu)
+    lv_obj_set_pos(s_gb_body_cont, 0, 24);
+    lv_obj_set_scroll_dir(s_gb_body_cont, LV_DIR_VER);
+    lv_obj_set_scrollbar_mode(s_gb_body_cont, LV_SCROLLBAR_MODE_AUTO);
+
+    s_gb_body = lv_label_create(s_gb_body_cont);
+    lv_obj_set_style_text_color(s_gb_body, lv_color_hex(COL_VALUE), 0);
+    lv_obj_set_style_text_font(s_gb_body, &lv_font_montserrat_14, 0);
+    lv_label_set_long_mode(s_gb_body, LV_LABEL_LONG_WRAP);
+    lv_obj_set_width(s_gb_body, DUI_HOR_RES - 24);   // container width minus padding
+    lv_label_set_text(s_gb_body, "");
+    lv_obj_align(s_gb_body, LV_ALIGN_TOP_LEFT, 0, 0);
+}
+
 static void build_status_screen(void) {
     lv_obj_t *scr = s_scr_status;
     lv_obj_set_style_bg_color(scr, lv_color_hex(COL_BG), 0);
@@ -853,10 +964,49 @@ static void apply_status(const display_status_t *s) {
     // Active step → scene view; idle → status view. The 5-way buttons can
     // override (1=force status, 2=force scene; 0=auto). Fade 240 ms, the
     // original ui_manager default transition (SceneTransition::kFade, 240).
+    // Gamebook page: refresh its labels from the latest show() buffers, then
+    // apply any pending manual scroll (Up/Down) on the reading container.
+    if (s_gamebook_open) {
+        lv_label_set_text(s_gb_title, s_gb_title_buf);
+        lv_label_set_text(s_gb_body,  s_gb_body_buf);
+        lv_label_set_text(s_gb_menu,  s_gb_menu_buf);
+        if (s_gb_scroll_home) {
+            lv_obj_scroll_to_y(s_gb_body_cont, 0, LV_ANIM_OFF);
+            s_gb_scroll_home = false;
+        }
+        int req = s_gb_scroll_req;
+        if (req != 0) {
+            s_gb_scroll_req = 0;
+            // +1 page = read further down (content moves up): negative dy.
+            lv_obj_scroll_by(s_gb_body_cont, 0, -req * 130, LV_ANIM_ON);
+        }
+    }
+    // Library grid: refresh tile titles + highlight the selected one.
+    if (s_library_open) {
+        for (int i = 0; i < LIB_TILES; i++) {
+            if (i < s_lib_count_buf) {
+                lv_label_set_text(s_lib_label[i], s_lib_title_buf[i]);
+                lv_obj_clear_flag(s_lib_tile[i], LV_OBJ_FLAG_HIDDEN);
+            } else {
+                lv_obj_add_flag(s_lib_tile[i], LV_OBJ_FLAG_HIDDEN);
+            }
+            bool on = (i == s_lib_sel_buf);
+            lv_obj_set_style_border_color(s_lib_tile[i],
+                lv_color_hex(on ? COL_CODE : 0x445566), 0);
+            lv_obj_set_style_border_width(s_lib_tile[i], on ? 4 : 2, 0);
+            lv_obj_set_style_bg_color(s_lib_tile[i],
+                lv_color_hex(on ? 0x223040 : 0x101820), 0);
+        }
+    }
+
     extern volatile uint8_t g_dui_view_override;
     lv_obj_t *want;
     if (!s_intro_done) {
         want = s_scr_intro;            // boot intro plays out first
+    } else if (s_library_open) {
+        want = s_scr_library;          // story picker owns the screen
+    } else if (s_gamebook_open) {
+        want = s_scr_gamebook;         // gamebook owns the screen while running
     } else if (s_browser_open) {
         want = s_scr_browser;          // file browser on top of everything
     } else if (s_shell_open) {
@@ -964,6 +1114,8 @@ static void display_task(void *arg) {
     build_status_screen();
     build_scene_screen();
     build_shell_screen();
+    build_gamebook_screen();
+    build_library_screen();
     build_browser_screen();
     build_intro_screen();
 
@@ -1145,9 +1297,67 @@ extern "C" void display_ui_set_status(const display_status_t *s) {
     }
 }
 
+static display_ui_key_hook_t s_key_hook = nullptr;
+
+extern "C" void display_ui_set_key_hook(display_ui_key_hook_t hook) {
+    s_key_hook = hook;
+}
+
+extern "C" void display_ui_gamebook_show(const char *title, const char *body,
+                                         const char *menu, bool reset_scroll) {
+    if (s_mutex) xSemaphoreTake(s_mutex, portMAX_DELAY);
+    snprintf(s_gb_title_buf, sizeof(s_gb_title_buf), "%s", title ? title : "");
+    snprintf(s_gb_body_buf,  sizeof(s_gb_body_buf),  "%s", body  ? body  : "");
+    snprintf(s_gb_menu_buf,  sizeof(s_gb_menu_buf),  "%s", menu  ? menu  : "");
+    if (reset_scroll) s_gb_scroll_home = true;   // new passage → back to top
+    s_gamebook_open = true;
+    s_dirty = true;
+    if (s_mutex) xSemaphoreGive(s_mutex);
+}
+
+extern "C" void display_ui_gamebook_scroll(int dir) {
+    // dir > 0 : read further down ; dir < 0 : back up. Accumulated and applied
+    // on the display task (this runs on the buttons task).
+    s_gb_scroll_req += (dir > 0) ? 1 : -1;
+    s_dirty = true;
+}
+
+extern "C" void display_ui_gamebook_hide(void) {
+    s_gamebook_open = false;
+    s_gb_scroll_req = 0;
+    s_dirty = true;
+}
+
+extern "C" void display_ui_library_show(const char *const *titles, int count,
+                                        int sel) {
+    if (count > LIB_TILES) count = LIB_TILES;
+    if (count < 0) count = 0;
+    if (s_mutex) xSemaphoreTake(s_mutex, portMAX_DELAY);
+    s_lib_count_buf = count;
+    s_lib_sel_buf   = (count > 0) ? ((sel % count + count) % count) : 0;
+    for (int i = 0; i < LIB_TILES; i++) {
+        snprintf(s_lib_title_buf[i], sizeof(s_lib_title_buf[i]), "%s",
+                 (i < count && titles && titles[i]) ? titles[i] : "");
+    }
+    s_library_open  = true;
+    s_gamebook_open = false;   // library and a running story are exclusive
+    s_dirty = true;
+    if (s_mutex) xSemaphoreGive(s_mutex);
+}
+
+extern "C" void display_ui_library_hide(void) {
+    s_library_open = false;
+    s_dirty = true;
+}
+
 extern "C" void display_ui_handle_key(uint8_t key) {
     static uint8_t s_brightness = 100;
 
+    // Takeover mode (e.g. gamebook): if the hook consumes the key, stop here.
+    if (s_key_hook && s_key_hook(key)) {
+        return;
+    }
+
     // Any key skips the boot intro.
     if (!s_intro_done) {
         s_intro_done = true;

idf_zacus/components/display_ui/include/display_ui.h

@@ -78,6 +78,59 @@ void display_ui_camera_frame(const uint8_t *gray, int width, int height);
  */
 void display_ui_handle_key(uint8_t key);
 
+/**
+ * @brief Install an optional key interceptor.
+ *
+ * If a hook is set and returns true for a given key, the key is consumed and
+ * the normal shell/scene handling is skipped. Lets a takeover mode (e.g. the
+ * gamebook) own the 5-way pad without display_ui having to depend on it.
+ * Pass NULL to remove the hook. Thread-safe (single pointer write).
+ */
+typedef bool (*display_ui_key_hook_t)(uint8_t key);
+void display_ui_set_key_hook(display_ui_key_hook_t hook);
+
+/**
+ * @brief Show a full-screen gamebook page (takes over the display).
+ *
+ * Dedicated "livre dont vous êtes le héros" view: a centred title, the full
+ * passage text wrapped over multiple lines, and a choice menu at the bottom.
+ * Forces this view until display_ui_gamebook_hide(). Thread-safe (buffers
+ * copied under the mutex; the actual LVGL render runs on the display task).
+ * Strings should be ASCII (the embedded fonts have no accents).
+ *
+ * @param title  short page title (top)
+ * @param body   full passage text (wrapped); may be long
+ * @param menu   choice lines (bottom), e.g. "[OK] ...\n[<>] ..."
+ */
+void display_ui_gamebook_show(const char *title, const char *body,
+                              const char *menu, bool reset_scroll);
+
+/**
+ * @brief Scroll the gamebook reading area (Up/Down buttons).
+ * @param dir  >0 = read further down, <0 = back up. Half-page per call.
+ */
+void display_ui_gamebook_scroll(int dir);
+
+/** @brief Leave the gamebook view and return to the normal scene/status flow. */
+void display_ui_gamebook_hide(void);
+
+/**
+ * @brief Show the gamebook library as a grid of up to 6 tiles.
+ *
+ * Each tile shows a story title; the tile at index `sel` is highlighted. The
+ * gamebook owns the pad and calls this on every cursor move. Forces the
+ * library view until a story is loaded or display_ui_library_hide(). Titles
+ * should be ASCII.
+ *
+ * @param titles  array of `count` story titles
+ * @param count   number of stories (clamped to 6)
+ * @param sel     highlighted tile index
+ */
+void display_ui_library_show(const char *const *titles, int count, int sel);
+
+/** @brief Leave the library view. */
+void display_ui_library_hide(void);
+
 /**
  * @brief Pop the pending shell-app launch request, if any.
  *

idf_zacus/components/game_endpoint/CMakeLists.txt

@@ -17,6 +17,11 @@ idf_component_register(
         puzzle_state
         media_manager
         sd_storage
+        gamebook
     PRIV_REQUIRES
         local_puzzles
+        p7_coffre
+        p5_morse
+        p6_nfc
+        npc_engine
 )

idf_zacus/components/game_endpoint/game_endpoint.c

@@ -27,9 +27,14 @@
 #include "nvs_flash.h"
 
 #include "hints_client.h"
+#include "gamebook.h"
+#include "npc_engine.h"
 #include "scenario_mesh.h"
 #include "puzzle_binding.h"
 #include "local_puzzles.h"
+#include "p7_coffre.h"
+#include "p5_morse.h"
+#include "p6_nfc.h"
 
 static const char *TAG = "game_endpoint";
 
@@ -38,11 +43,40 @@ static const char *TAG = "game_endpoint";
 static puzzle_state_t *s_pstate = NULL;
 // Last step id armed via POST /game/step. Empty string = none.
 static char s_current_step_id[64] = {0};
-// Last armed puzzle type: "qr" | "sound" | "none" | "" (empty = nothing armed yet).
-static char s_current_armed[8] = {0};
+// Last armed puzzle type: "qr" | "sound" | "morse" | "nfc" | "none" | ""
+static char s_current_armed[12] = {0};   // "qr"|"sound"|"morse"|"nfc"|"none"
 // Display scene metadata of the current step (lenient parse; defaults safe).
 static scene_binding_t s_current_scene = {0};
 
+// ─── P5 / P6 solved-callback context ────────────────────────────────────────
+// Mirrors the on_qr_solved / on_sound_solved pattern in local_puzzles.c:
+// context is copied at arm time, callback is called on the puzzle task.
+
+typedef struct {
+    uint8_t id;
+    uint8_t frag[PB_MAX_FRAG];
+    uint8_t frag_len;
+} ep_puzzle_ctx_t;
+
+static ep_puzzle_ctx_t s_morse_ctx = {0};
+static ep_puzzle_ctx_t s_nfc_ctx   = {0};
+
+static void on_morse_solved(void)
+{
+    if (s_pstate) {
+        puzzle_state_report(s_pstate, s_morse_ctx.id,
+                            s_morse_ctx.frag, s_morse_ctx.frag_len);
+    }
+}
+
+static void on_nfc_solved(void)
+{
+    if (s_pstate) {
+        puzzle_state_report(s_pstate, s_nfc_ctx.id,
+                            s_nfc_ctx.frag, s_nfc_ctx.frag_len);
+    }
+}
+
 // Whitelist mirrored in the 4xx error message so the operator can
 // recover without grepping the source. Keep aligned with
 // hints_client_set_group_profile() validation.
@@ -332,8 +366,13 @@ static esp_err_t scenario_apply_buffer(const char *body, size_t len,
 
     // New scenario supersedes any live puzzle — disarm and clear remembered step.
     local_puzzles_disarm();
+    p5_morse_disarm();   // no-op when CONFIG_ZACUS_P5_MORSE_ENABLE=n
+    p6_nfc_disarm();     // no-op when CONFIG_ZACUS_P6_NFC_ENABLE=n
     s_current_step_id[0] = '\0';
     memset(&s_current_scene, 0, sizeof(s_current_scene));
+    // Re-arm the coffre for the new game session.
+    // p7_coffre_lock() is a no-op stub when CONFIG_ZACUS_P7_COFFRE_ENABLE=n.
+    p7_coffre_lock();
 
     // Hot-reload-via-reboot until scenario_engine_reload() lands (Phase 3).
     schedule_restart();
@@ -723,7 +762,7 @@ static esp_err_t handle_espnow_cmd_post(httpd_req_t *req) {
 
 // Core step-change logic, shared by the HTTP handler and internal callers
 // (e.g. shell app launch on the local display). Error contract:
-//   ESP_OK                 armed (armed_out = "qr"|"sound"|"none")
+//   ESP_OK                 armed (armed_out = "qr"|"sound"|"morse"|"nfc"|"none")
 //   ESP_ERR_INVALID_STATE  not ready (no puzzle_state wired)
 //   ESP_ERR_NOT_SUPPORTED  no scenario stored / storage unavailable
 //   ESP_ERR_NOT_FOUND      unknown step id
@@ -754,6 +793,9 @@ esp_err_t game_endpoint_apply_step(const char *step_id,
 
     // Disarm current puzzle before re-arming.
     local_puzzles_disarm();
+    // Disarm P5/P6 stubs (no-ops when their CONFIG flags are off).
+    p5_morse_disarm();
+    p6_nfc_disarm();
 
     // Parse binding for this step (+ display scene metadata, lenient).
     puzzle_binding_t binding;
@@ -767,6 +809,22 @@ esp_err_t game_endpoint_apply_step(const char *step_id,
     strncpy(s_current_step_id, step_id, sizeof(s_current_step_id) - 1);
     s_current_step_id[sizeof(s_current_step_id) - 1] = '\0';
 
+    // Sync the NPC/hints engine (and, through it, the voice gateway) with the
+    // scene this step enters. This is what makes the phone NPCs hint on — and
+    // scripted calls auto-ring for — the current scene, regardless of whether
+    // the step was driven by the local game loop or an external POST /game/step.
+    // Best-effort: notify is async + server-side idempotent and never blocks
+    // the step change. Steps with no scene_id (or an unknown one) are skipped.
+    if (s_current_scene.scene_id[0] != '\0') {
+        (void) npc_engine_set_scene_by_id(s_current_scene.scene_id, 0);
+    }
+
+    // P7 coffre: fire the actuator when the final win step is reached.
+    // p7_coffre_unlock() is a no-op stub when CONFIG_ZACUS_P7_COFFRE_ENABLE=n.
+    if (strcmp(step_id, "STEP_FINAL_WIN") == 0) {
+        p7_coffre_unlock();
+    }
+
     const char *armed = "none";
     if (binding.type == PB_QR) {
         const char *ptrs[PB_MAX_CODES];
@@ -794,6 +852,26 @@ esp_err_t game_endpoint_apply_step(const char *step_id,
         if (aerr == ESP_ERR_INVALID_STATE) return ESP_ERR_TIMEOUT;
         if (aerr != ESP_OK) return aerr;
         armed = "sound";
+    } else if (binding.type == PB_MORSE) {
+        // P5 morse: copy context then arm.
+        // p5_morse_arm() is a no-op stub when CONFIG_ZACUS_P5_MORSE_ENABLE=n.
+        s_morse_ctx.id       = binding.id;
+        s_morse_ctx.frag_len = binding.fragment_len;
+        if (binding.fragment_len > 0) {
+            memcpy(s_morse_ctx.frag, binding.fragment, binding.fragment_len);
+        }
+        p5_morse_arm(binding.morse_expected, on_morse_solved);
+        armed = "morse";
+    } else if (binding.type == PB_NFC) {
+        // P6 nfc: copy context then arm.
+        // p6_nfc_arm() is a no-op stub when CONFIG_ZACUS_P6_NFC_ENABLE=n.
+        s_nfc_ctx.id       = binding.id;
+        s_nfc_ctx.frag_len = binding.fragment_len;
+        if (binding.fragment_len > 0) {
+            memcpy(s_nfc_ctx.frag, binding.fragment, binding.fragment_len);
+        }
+        p6_nfc_arm(binding.nfc_uid, on_nfc_solved);
+        armed = "nfc";
     }
 
     strncpy(s_current_armed, armed, sizeof(s_current_armed) - 1);
@@ -917,6 +995,29 @@ static esp_err_t handle_media_play_post(httpd_req_t *req) {
     return send_error(req, "500 Internal Server Error", esp_err_to_name(err));
 }
 
+// ─── POST /game/gamebook[?action=stop] ───────────────────────────────────────
+//
+// Start the "livre dont vous êtes le héros" mode (loads /sdcard/gamebook/
+// gamebook.json, plays the start passage, takes over the 5-way pad). With
+// ?action=stop it leaves gamebook mode. No body required.
+static esp_err_t handle_gamebook_post(httpd_req_t *req) {
+    char query[32] = {0}, action[16] = {0};
+    if (httpd_req_get_url_query_str(req, query, sizeof(query)) == ESP_OK) {
+        httpd_query_key_value(query, "action", action, sizeof(action));
+    }
+    if (strcmp(action, "stop") == 0) {
+        gamebook_stop();
+        return send_json(req, "200 OK", "{\"gamebook\":\"stopped\"}");
+    }
+    esp_err_t err = gamebook_start();
+    if (err != ESP_OK) {
+        return send_error(req, "503 Service Unavailable",
+                          err == ESP_ERR_NOT_FOUND ? "gamebook_not_on_sd"
+                                                   : "gamebook_load_failed");
+    }
+    return send_json(req, "200 OK", "{\"gamebook\":\"started\"}");
+}
+
 // ─── GET /game/puzzle_state ──────────────────────────────────────────────────
 //
 // Returns {"step_id":"STEP_X"|null, "solved":[1,3], "code":"125"}.
@@ -1072,6 +1173,43 @@ static esp_err_t handle_file_post(httpd_req_t *req) {
     return send_json(req, "200 OK", resp);
 }
 
+// ─── DELETE /game/file?path=sd/<…>|apps/<…> — remove a staged file ───────────
+// Same path routing/whitelist as the POST handler. Lets the host clean stale
+// assets (e.g. orphan gamebook WAVs) off the SD without reflashing.
+static esp_err_t handle_file_delete(httpd_req_t *req) {
+    char query[160], path_param[96];
+    if (httpd_req_get_url_query_str(req, query, sizeof(query)) != ESP_OK ||
+        httpd_query_key_value(query, "path", path_param,
+                              sizeof(path_param)) != ESP_OK) {
+        return send_error(req, "400 Bad Request", "missing path param");
+    }
+    const bool to_sd   = (strncmp(path_param, "sd/", 3) == 0);
+    const bool to_apps = (strncmp(path_param, "apps/", 5) == 0);
+    if ((!to_sd && !to_apps) || strstr(path_param, "..") ||
+        path_param[strlen(path_param) - 1] == '/') {
+        return send_error(req, "403 Forbidden", "path must be under apps/ or sd/");
+    }
+    char full[192];
+    if (to_sd) {
+        if (!sd_storage_ready()) {
+            return send_error(req, "503 Service Unavailable", "sd_not_mounted");
+        }
+        snprintf(full, sizeof(full), "/sdcard/%s", path_param + 3);
+    } else {
+        if (mount_storage_lazy() != ESP_OK) {
+            return send_error(req, "503 Service Unavailable", "storage_unavailable");
+        }
+        snprintf(full, sizeof(full), "/littlefs/%s", path_param);
+    }
+    if (unlink(full) != 0) {
+        return send_error(req, "404 Not Found", "remove failed");
+    }
+    ESP_LOGI(TAG, "file removed: %s", full);
+    char resp[192];
+    snprintf(resp, sizeof(resp), "{\"status\":\"ok\",\"removed\":\"%s\"}", path_param);
+    return send_json(req, "200 OK", resp);
+}
+
 void game_endpoint_get_scene(scene_binding_t *out) {
     if (!out) return;
     // Lock-free snapshot, same class as get_puzzle_status: written on the
@@ -1143,6 +1281,12 @@ esp_err_t game_endpoint_init(httpd_handle_t server) {
         .handler  = handle_file_post,
         .user_ctx = NULL,
     };
+    static const httpd_uri_t uri_file_delete = {
+        .uri      = "/game/file",
+        .method   = HTTP_DELETE,
+        .handler  = handle_file_delete,
+        .user_ctx = NULL,
+    };
     static const httpd_uri_t uri_espnow_cmd_post = {
         .uri      = "/game/espnow/cmd",
         .method   = HTTP_POST,
@@ -1155,6 +1299,12 @@ esp_err_t game_endpoint_init(httpd_handle_t server) {
         .handler  = handle_media_play_post,
         .user_ctx = NULL,
     };
+    static const httpd_uri_t uri_gamebook_post = {
+        .uri      = "/game/gamebook",
+        .method   = HTTP_POST,
+        .handler  = handle_gamebook_post,
+        .user_ctx = NULL,
+    };
 
     // Bring up the ESP-NOW mesh transport. The master is primarily a sender
     // (the relay handler) but we also register the apply adapter so a peer
@@ -1219,6 +1369,10 @@ esp_err_t game_endpoint_init(httpd_handle_t server) {
     if (err != ESP_OK) {
         ESP_LOGW(TAG, "register POST /game/step: %s", esp_err_to_name(err));
     }
+    err = httpd_register_uri_handler(server, &uri_file_delete);
+    if (err != ESP_OK) {
+        ESP_LOGW(TAG, "register DELETE /game/file: %s", esp_err_to_name(err));
+    }
     err = httpd_register_uri_handler(server, &uri_file_post);
     if (err != ESP_OK) {
         ESP_LOGW(TAG, "register /game/file: %s", esp_err_to_name(err));
@@ -1231,6 +1385,10 @@ esp_err_t game_endpoint_init(httpd_handle_t server) {
     if (err != ESP_OK) {
         ESP_LOGW(TAG, "register POST /game/media/play: %s", esp_err_to_name(err));
     }
+    err = httpd_register_uri_handler(server, &uri_gamebook_post);
+    if (err != ESP_OK) {
+        ESP_LOGW(TAG, "register POST /game/gamebook: %s", esp_err_to_name(err));
+    }
 
     ESP_LOGI(TAG, "game endpoint registered "
                   "(GET+POST /game/group_profile, POST /game/scenario%s, "

idf_zacus/components/game_endpoint/include/puzzle_binding.h

@@ -11,7 +11,7 @@
 #define PB_MAX_FRAG     4   // = PUZZLE_MAX_FRAG
 #define PB_MAX_PUZZLE_ID 8  // = PUZZLE_MAX_ID
 
-typedef enum { PB_NONE = 0, PB_QR, PB_SOUND } puzzle_binding_type_t;
+typedef enum { PB_NONE = 0, PB_QR, PB_SOUND, PB_MORSE, PB_NFC } puzzle_binding_type_t;
 
 typedef struct {
     puzzle_binding_type_t type;     // PB_NONE if the step has no puzzle
@@ -23,6 +23,10 @@ typedef struct {
     int     melody[PB_MAX_NOTES];
     size_t  note_count;
     int     tolerance;              // default 1
+    // morse (P5): expected word in uppercase, e.g. "ZACUS"
+    char    morse_expected[32];
+    // nfc (P6): expected tag UID as colon-hex string, e.g. "A1:B2:C3:D4"
+    char    nfc_uid[20];
     // common
     uint8_t fragment[PB_MAX_FRAG];
     uint8_t fragment_len;
@@ -51,6 +55,7 @@ esp_err_t puzzle_binding_from_ir(const char *ir_json, const char *step_id,
 #define SB_MAX_TITLE    48
 #define SB_MAX_SUBTITLE 64
 #define SB_MAX_SYMBOL   16
+#define SB_MAX_SCENE_ID 40
 
 typedef enum {
     SB_FX_PULSE = 0,   // default (original SceneEffect::kPulse)
@@ -61,6 +66,8 @@ typedef enum {
 
 typedef struct {
     bool present;                     // step has a scene object
+    char scene_id[SB_MAX_SCENE_ID];   // canonical SCENE_* id (sibling of step id;
+                                      // set even when no display scene object)
     char title[SB_MAX_TITLE];
     char subtitle[SB_MAX_SUBTITLE];
     char symbol[SB_MAX_SYMBOL];

idf_zacus/components/game_endpoint/puzzle_binding.c

@@ -60,6 +60,10 @@ esp_err_t puzzle_binding_from_ir(const char *ir_json, const char *step_id,
         ptype = PB_QR;
     } else if (strcmp(jtype->valuestring, "sound") == 0) {
         ptype = PB_SOUND;
+    } else if (strcmp(jtype->valuestring, "morse") == 0) {
+        ptype = PB_MORSE;
+    } else if (strcmp(jtype->valuestring, "nfc") == 0) {
+        ptype = PB_NFC;
     } else {
         cJSON_Delete(root);
         memset(out, 0, sizeof(*out));
@@ -120,7 +124,7 @@ esp_err_t puzzle_binding_from_ir(const char *ir_json, const char *step_id,
             out->codes[i][PB_MAX_LABEL - 1] = '\0';
         }
         out->code_count = (size_t)n;
-    } else { // PB_SOUND
+    } else if (ptype == PB_SOUND) {
         const cJSON *jmelody = cJSON_GetObjectItemCaseSensitive(puzzle, "melody");
         if (!cJSON_IsArray(jmelody)) {
             cJSON_Delete(root);
@@ -156,6 +160,37 @@ esp_err_t puzzle_binding_from_ir(const char *ir_json, const char *step_id,
         } else {
             out->tolerance = 1;
         }
+    } else if (ptype == PB_MORSE) {
+        // puzzle.expected: required string, uppercase word, e.g. "ZACUS"
+        const cJSON *jexp = cJSON_GetObjectItemCaseSensitive(puzzle, "expected");
+        if (!cJSON_IsString(jexp) || !jexp->valuestring || !jexp->valuestring[0]) {
+            cJSON_Delete(root);
+            memset(out, 0, sizeof(*out));
+            return ESP_ERR_INVALID_ARG;
+        }
+        if (strlen(jexp->valuestring) >= sizeof(out->morse_expected)) {
+            cJSON_Delete(root);
+            memset(out, 0, sizeof(*out));
+            return ESP_ERR_INVALID_ARG;
+        }
+        strncpy(out->morse_expected, jexp->valuestring,
+                sizeof(out->morse_expected) - 1);
+        out->morse_expected[sizeof(out->morse_expected) - 1] = '\0';
+    } else if (ptype == PB_NFC) {
+        // puzzle.uid: required string, colon-hex UID, e.g. "A1:B2:C3:D4"
+        const cJSON *juid = cJSON_GetObjectItemCaseSensitive(puzzle, "uid");
+        if (!cJSON_IsString(juid) || !juid->valuestring || !juid->valuestring[0]) {
+            cJSON_Delete(root);
+            memset(out, 0, sizeof(*out));
+            return ESP_ERR_INVALID_ARG;
+        }
+        if (strlen(juid->valuestring) >= sizeof(out->nfc_uid)) {
+            cJSON_Delete(root);
+            memset(out, 0, sizeof(*out));
+            return ESP_ERR_INVALID_ARG;
+        }
+        strncpy(out->nfc_uid, juid->valuestring, sizeof(out->nfc_uid) - 1);
+        out->nfc_uid[sizeof(out->nfc_uid) - 1] = '\0';
     }
 
     // id/type written last — all error exits above leave *out zeroed
@@ -204,9 +239,15 @@ esp_err_t scene_binding_from_ir(const char *ir_json, const char *step_id,
     }
     if (!step) { cJSON_Delete(root); return ESP_ERR_NOT_FOUND; }
 
+    // Canonical scene id (SCENE_*), sibling of the step "id". Captured
+    // unconditionally so npc_engine / the voice gateway can be synced even
+    // on steps that carry no display "scene" object.
+    copy_scene_str(out->scene_id, sizeof(out->scene_id),
+                   cJSON_GetObjectItemCaseSensitive(step, "scene_id"));
+
     const cJSON *scene = cJSON_GetObjectItemCaseSensitive(step, "scene");
     if (!scene || !cJSON_IsObject(scene)) {
-        // No scene — present stays false, ESP_OK.
+        // No scene object — present stays false, ESP_OK (scene_id may be set).
         cJSON_Delete(root);
         return ESP_OK;
     }

idf_zacus/components/gamebook/CMakeLists.txt

@@ -0,0 +1,11 @@
+idf_component_register(
+    SRCS
+        "gamebook.c"
+    INCLUDE_DIRS
+        "include"
+    REQUIRES
+        json
+        display_ui
+        media_manager
+        log
+)

idf_zacus/components/gamebook/gamebook.c

@@ -0,0 +1,306 @@
+// gamebook.c — see gamebook.h. Standalone gamebook player + library for the
+// Freenove master. On boot it shows the library (a grid of story tiles);
+// picking one loads and plays that book; finishing returns to the library.
+#include "gamebook.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "cJSON.h"
+#include "esp_log.h"
+#include "esp_heap_caps.h"
+
+#include "display_ui.h"
+#include "media_manager.h"
+
+static const char *TAG = "gamebook";
+
+#define GAMEBOOK_DIR   "/sdcard/gamebook"
+#define LIBRARY_JSON   GAMEBOOK_DIR "/library.json"
+#define JSON_MAX       (256 * 1024)  /* expanded books are ~80 KB of JSON */
+#define LIB_MAX        6             /* matches the display tile grid */
+
+typedef enum { GB_OFF, GB_LIBRARY, GB_STORY } gb_mode_t;
+
+// Library state
+static cJSON   *s_lib_root  = NULL;  /* owns library.json while loaded   */
+static cJSON   *s_lib       = NULL;  /* borrowed: root->"library" array  */
+static int      s_lib_n     = 0;
+static int      s_lib_sel   = 0;
+
+// Story state
+static cJSON   *s_book      = NULL;  /* owns the current <book>.json      */
+static cJSON   *s_passages  = NULL;  /* borrowed: book->"passages"        */
+static char     s_title[48] = {0};
+static char     s_current[48] = {0};
+static int      s_sel       = 0;     /* highlighted choice index          */
+
+static volatile gb_mode_t s_mode = GB_OFF;
+
+bool gamebook_active(void) { return s_mode != GB_OFF; }
+
+/* PSRAM allocators for cJSON — book trees are large (see gamebook_init). */
+static void *gb_malloc(size_t sz) { return heap_caps_malloc(sz, MALLOC_CAP_SPIRAM); }
+static void  gb_free(void *p)      { heap_caps_free(p); }
+
+/* Read a whole JSON file from the SD into a cJSON tree (caller frees). */
+static cJSON *load_json(const char *path)
+{
+    FILE *f = fopen(path, "rb");
+    if (!f) { ESP_LOGW(TAG, "open %s failed", path); return NULL; }
+    fseek(f, 0, SEEK_END);
+    long sz = ftell(f);
+    rewind(f);
+    if (sz <= 0 || sz > JSON_MAX) { fclose(f); ESP_LOGW(TAG, "%s size %ld", path, sz); return NULL; }
+    char *buf = heap_caps_malloc((size_t)sz + 1, MALLOC_CAP_SPIRAM);
+    if (!buf) { fclose(f); return NULL; }
+    size_t rd = fread(buf, 1, (size_t)sz, f);
+    fclose(f);
+    buf[rd] = '\0';
+    cJSON *root = cJSON_Parse(buf);
+    heap_caps_free(buf);
+    if (!root) ESP_LOGW(TAG, "malformed JSON: %s", path);
+    return root;
+}
+
+// ── Library ──────────────────────────────────────────────────────────────────
+
+static void render_library(void)
+{
+    const char *titles[LIB_MAX] = {0};
+    int n = (s_lib_n < LIB_MAX) ? s_lib_n : LIB_MAX;
+    for (int i = 0; i < n; i++) {
+        const cJSON *t = cJSON_GetObjectItem(cJSON_GetArrayItem(s_lib, i), "title");
+        titles[i] = cJSON_IsString(t) ? t->valuestring : "?";
+    }
+    display_ui_library_show(titles, n, s_lib_sel);
+}
+
+static void free_story(void)
+{
+    media_manager_stop();
+    if (s_book) { cJSON_Delete(s_book); s_book = NULL; }
+    s_passages = NULL;
+    s_current[0] = '\0';
+}
+
+static esp_err_t open_library(void)
+{
+    free_story();
+    if (s_lib_root) { cJSON_Delete(s_lib_root); s_lib_root = NULL; s_lib = NULL; }
+
+    s_lib_root = load_json(LIBRARY_JSON);
+    if (!s_lib_root) return ESP_ERR_NOT_FOUND;
+    s_lib = cJSON_GetObjectItem(s_lib_root, "library");
+    if (!cJSON_IsArray(s_lib) || cJSON_GetArraySize(s_lib) == 0) {
+        cJSON_Delete(s_lib_root); s_lib_root = NULL; s_lib = NULL;
+        ESP_LOGW(TAG, "library.json has no stories");
+        return ESP_ERR_INVALID_ARG;
+    }
+    s_lib_n = cJSON_GetArraySize(s_lib);
+    s_lib_sel = 0;
+    s_mode = GB_LIBRARY;
+    display_ui_gamebook_hide();
+    render_library();
+    ESP_LOGI(TAG, "library open (%d stories)", s_lib_n);
+    return ESP_OK;
+}
+
+// ── Story ─────────────────────────────────────────────────────────────────────
+
+static const cJSON *cur_passage(void) { return cJSON_GetObjectItem(s_passages, s_current); }
+
+static int cur_choice_count(void)
+{
+    const cJSON *ch = cJSON_GetObjectItem(cur_passage(), "choices");
+    return cJSON_IsArray(ch) ? cJSON_GetArraySize(ch) : 0;
+}
+
+/* (Re)draw the current page. `home` true = new passage (reset scroll to top);
+ * false = same passage, only the choice cursor moved (keep scroll position).
+ * The bottom line shows ONE choice with < > arrows (Left/Right cycles). */
+static void render_page(bool home)
+{
+    const cJSON *p = cur_passage();
+    if (!cJSON_IsObject(p)) return;
+    const cJSON *screen  = cJSON_GetObjectItem(p, "screen");
+    const cJSON *text    = cJSON_GetObjectItem(p, "text");
+    const cJSON *choices = cJSON_GetObjectItem(p, "choices");
+    int n = cJSON_IsArray(choices) ? cJSON_GetArraySize(choices) : 0;
+
+    char menu[160];
+    if (n == 0) {
+        snprintf(menu, sizeof(menu), "~ Fin ~   (clic = bibliotheque)");
+    } else {
+        const cJSON *lbl = cJSON_GetObjectItem(cJSON_GetArrayItem(choices, s_sel),
+                                               "label");
+        const char *txt = cJSON_IsString(lbl) ? lbl->valuestring : "?";
+        if (n > 1) {
+            /* Down cycles the answer, Click validates. */
+            snprintf(menu, sizeof(menu), "%d/%d  %s   (bas / clic)",
+                     s_sel + 1, n, txt);
+        } else {
+            snprintf(menu, sizeof(menu), "%s   (clic)", txt);
+        }
+    }
+    display_ui_gamebook_show(
+        cJSON_IsString(screen) ? screen->valuestring : s_title,
+        cJSON_IsString(text)   ? text->valuestring   : "",
+        menu, home);
+}
+
+static void enter_passage(const char *pid)
+{
+    const cJSON *p = cJSON_GetObjectItem(s_passages, pid);
+    if (!cJSON_IsObject(p)) { ESP_LOGW(TAG, "passage '%s' not found", pid); return; }
+    snprintf(s_current, sizeof(s_current), "%s", pid);
+    s_sel = 0;
+    render_page(true);
+
+    const cJSON *wav = cJSON_GetObjectItem(p, "wav");
+    if (cJSON_IsString(wav) && wav->valuestring[0]) {
+        char path[96];
+        snprintf(path, sizeof(path), "%s/%s", GAMEBOOK_DIR, wav->valuestring);
+        media_manager_stop();   /* a choice skips the narration */
+        media_manager_play(path);
+    }
+    ESP_LOGI(TAG, "passage '%s' (%d choices)", pid, cur_choice_count());
+}
+
+/* Load story #idx from the library and play it. */
+static void load_book(int idx)
+{
+    const cJSON *e = cJSON_GetArrayItem(s_lib, idx);
+    const cJSON *file = cJSON_GetObjectItem(e, "book");
+    if (!cJSON_IsString(file)) { ESP_LOGW(TAG, "library[%d] has no book", idx); return; }
+
+    char path[128];
+    snprintf(path, sizeof(path), "%s/%s", GAMEBOOK_DIR, file->valuestring);
+    cJSON *book = load_json(path);
+    if (!book) return;
+    const cJSON *passages = cJSON_GetObjectItem(book, "passages");
+    const cJSON *start = cJSON_GetObjectItem(book, "start");
+    const cJSON *title = cJSON_GetObjectItem(book, "title");
+    if (!cJSON_IsObject(passages) || !cJSON_IsString(start)) {
+        ESP_LOGW(TAG, "%s missing passages/start", path);
+        cJSON_Delete(book);
+        return;
+    }
+    free_story();
+    s_book = book;
+    s_passages = (cJSON *)passages;
+    snprintf(s_title, sizeof(s_title), "%s",
+             cJSON_IsString(title) ? title->valuestring : "");
+    s_mode = GB_STORY;
+    display_ui_library_hide();
+    ESP_LOGI(TAG, "load book \"%s\" @ '%s'", s_title, start->valuestring);
+    enter_passage(start->valuestring);
+}
+
+// ── Input ─────────────────────────────────────────────────────────────────────
+
+/* MEASURED pad mapping on this hardware (ADC ladder, see calibration):
+ *   key1 = CLICK (centre, ~0 mV)
+ *   key2 = LEFT  (gauche, ~700 mV)
+ *   key4 = DOWN  (bas,    ~1350 mV)
+ *   key5 = RIGHT (droite, ~1994 mV)
+ *   key3 is never produced; the physical UP button is electrically dead
+ *   (held = no voltage change), so nothing can depend on it. */
+#define PAD_CLICK 1
+#define PAD_LEFT  2
+#define PAD_DOWN  4
+#define PAD_RIGHT 5
+
+/* Library nav: Down/Right → next tile, Left → previous, Click → open.
+ * Selection wraps, so the working buttons reach every story without Up. */
+static bool library_key(uint8_t key)
+{
+    if (s_lib_n <= 0) return true;
+    switch (key) {
+    case PAD_DOWN: case PAD_RIGHT:
+        s_lib_sel = (s_lib_sel + 1) % s_lib_n; render_library(); break;
+    case PAD_LEFT:
+        s_lib_sel = (s_lib_sel - 1 + s_lib_n) % s_lib_n; render_library(); break;
+    case PAD_CLICK:
+        load_book(s_lib_sel); break;
+    default: break;
+    }
+    return true;
+}
+
+/* Story nav (Up is dead, so it's never used):
+ *   Left  → scroll the reading text UP
+ *   Right → scroll the reading text DOWN
+ *   Down  → next answer (cycles through all choices, wraps around)
+ *   Click → validate the highlighted answer
+ * On an ending page (no choices): Left/Right still scroll, Click → library. */
+static bool story_key(uint8_t key)
+{
+    if (key == PAD_LEFT)  { display_ui_gamebook_scroll(-1); return true; }  /* scroll up   */
+    if (key == PAD_RIGHT) { display_ui_gamebook_scroll(+1); return true; }  /* scroll down */
+
+    int n = cur_choice_count();
+    if (n == 0) {                       /* ending → click returns to library */
+        if (key == PAD_CLICK) open_library();
+        return true;
+    }
+    switch (key) {
+    case PAD_DOWN: s_sel = (s_sel + 1) % n; render_page(false); break;  /* next answer */
+    case PAD_CLICK: {                                                   /* validate */
+        const cJSON *choices = cJSON_GetObjectItem(cur_passage(), "choices");
+        const cJSON *g = cJSON_GetObjectItem(cJSON_GetArrayItem(choices, s_sel), "goto");
+        if (cJSON_IsString(g)) {
+            ESP_LOGI(TAG, "validate #%d -> '%s'", s_sel, g->valuestring);
+            enter_passage(g->valuestring);
+        }
+        break;
+    }
+    default: break;
+    }
+    return true;
+}
+
+static bool gamebook_key_hook(uint8_t key)
+{
+    switch (s_mode) {
+    case GB_LIBRARY: return library_key(key);
+    case GB_STORY:   return story_key(key);
+    default:         return false;   /* not active → let the shell have it */
+    }
+}
+
+// ── Public API ────────────────────────────────────────────────────────────────
+
+esp_err_t gamebook_start(void)
+{
+    return open_library();
+}
+
+void gamebook_stop(void)
+{
+    if (s_mode == GB_OFF) return;
+    s_mode = GB_OFF;
+    free_story();
+    if (s_lib_root) { cJSON_Delete(s_lib_root); s_lib_root = NULL; s_lib = NULL; }
+    display_ui_gamebook_hide();
+    display_ui_library_hide();
+    ESP_LOGI(TAG, "stopped");
+}
+
+void gamebook_init(void)
+{
+    /* Keep the large book trees in PSRAM — an expanded book is ~80 KB of JSON,
+     * whose parsed node tree would exhaust the internal heap on this firmware. */
+    static cJSON_Hooks hooks = {
+        .malloc_fn = gb_malloc,
+        .free_fn   = gb_free,
+    };
+    cJSON_InitHooks(&hooks);
+
+    /* Only register the pad hook here — the SD and media_manager aren't up yet
+     * at this point in boot. main.c calls gamebook_start() once they are, to
+     * boot into the library. */
+    display_ui_set_key_hook(gamebook_key_hook);
+    ESP_LOGI(TAG, "key hook installed");
+}

idf_zacus/components/gamebook/include/gamebook.h

@@ -0,0 +1,42 @@
+#pragma once
+// gamebook — "livre dont vous êtes le héros" mode for the Freenove master.
+//
+// Reads /sdcard/gamebook/gamebook.json (built by tools/gamebook/build_gamebook.py),
+// plays each passage's WAV narration from the SD via media_manager, shows the
+// passage title + choice menu on the display, and navigates with the 5-way pad
+// (it installs a display_ui key hook so it owns the buttons while active).
+// Fully local: no model, no gateway.
+//
+// JSON shape:
+//   { "title": "...", "start": "intro",
+//     "passages": { "intro": { "wav": "intro.wav", "screen": "...",
+//                              "menu": "[OK] ... | [<>] ...",
+//                              "choices": [ {"key": 1, "goto": "machine"}, ... ] },
+//                   ... } }
+// key codes match the 5-way pad: 1=SELECT 2=DOWN 3=MENU 4=LEFT/RIGHT 5=UP.
+// MENU (3) always quits the gamebook.
+
+#include "esp_err.h"
+#include <stdbool.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Install the display_ui key hook. Call once at boot, after display_ui_init().
+void gamebook_init(void);
+
+// Load /sdcard/gamebook/gamebook.json and enter the start passage (show text +
+// play its WAV). Returns ESP_ERR_* if the SD/JSON is missing or malformed.
+esp_err_t gamebook_start(void);
+
+// Leave gamebook mode: stop playback, release the JSON, hand the pad back to
+// the shell. Idempotent.
+void gamebook_stop(void);
+
+// True while a gamebook is running (used by the key hook to claim the pad).
+bool gamebook_active(void);
+
+#ifdef __cplusplus
+}
+#endif

idf_zacus/components/npc_engine/CMakeLists.txt

@@ -22,6 +22,7 @@ idf_component_register(
     REQUIRES
         media_manager
         hints_client
+        esp_http_client
         esp_timer
         esp_system
         nvs_flash

idf_zacus/components/npc_engine/include/npc_engine.h

@@ -169,6 +169,16 @@ esp_err_t npc_engine_trigger_cue(const char *cue_id);
 // and primes the stuck timer.
 esp_err_t npc_engine_set_step(uint8_t step_id, uint32_t expected_duration_ms);
 
+// Same bridge as npc_engine_set_step(), but keyed by the canonical scene
+// string id (e.g. "SCENE_WARNING") rather than its numeric index. Resolves
+// the id against the engine's internal scene table and forwards to
+// npc_engine_set_step(). Lets the game endpoint sync the NPC/hints engine
+// (and the voice gateway) straight from the scenario IR's `scene_id`.
+// Returns ESP_ERR_NOT_FOUND when the id matches no known scene (no notify
+// is sent in that case), ESP_ERR_INVALID_ARG on a NULL/empty id.
+esp_err_t npc_engine_set_scene_by_id(const char *scene_id,
+                                     uint32_t expected_duration_ms);
+
 // Request a hint for `puzzle_id` at escalation `level` (0..3). The engine
 // invokes `cb` with the resulting text. The current implementation is a
 // LOCAL STUB: it returns a hardcoded French placeholder synchronously.
@@ -184,6 +194,12 @@ const npc_state_t *npc_engine_state(void);
 // Thin wrapper kept here to give callers a single npc_engine_* surface.
 esp_err_t npc_engine_set_group_profile(const char *profile);
 
+// Point the engine at the voice gateway (tools/zacus-gateway). On each scene
+// change the engine POSTs the active SCENE_* to {base_url}/game/step?scene=… so
+// the phone NPCs disguise that scene's hint. Best-effort; pass NULL/"" to
+// disable. `token` is sent as a Bearer header (gateway require_token).
+void npc_engine_set_gateway(const char *base_url, const char *token);
+
 // Write the active puzzle id (e.g. "SCENE_LA_DETECTOR") into `out`,
 // truncated to `cap` bytes (NUL-terminated). Falls back to "SCENE_NPC"
 // when no scene is active or the engine is not initialised so callers

idf_zacus/components/npc_engine/npc_engine.c

@@ -17,16 +17,78 @@
 #include "npc_engine.h"
 
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
 
 #include "esp_err.h"
+#include "esp_http_client.h"
 #include "esp_log.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
 
 #include "media_manager.h"
 #include "hints_client.h"
 
 static const char *TAG = "npc_engine";
 
+// Voice gateway (tools/zacus-gateway) — we push the active SCENE_* to it at each
+// scene change so the phone NPCs disguise THIS scene's hint. Best-effort, set by
+// npc_engine_set_gateway(); empty = disabled (no-op, e.g. CI / dry runs).
+static char s_gw_url[128]   = {0};
+static char s_gw_token[80]  = {0};
+
+void npc_engine_set_gateway(const char *base_url, const char *token) {
+    if (base_url) snprintf(s_gw_url, sizeof(s_gw_url), "%s", base_url);
+    if (token)    snprintf(s_gw_token, sizeof(s_gw_token), "%s", token);
+}
+
+// One-shot worker: POST {gateway}/game/step?scene=<scene> then self-delete.
+// Runs on its own task so a slow/unreachable gateway NEVER stalls the scene
+// transition (the game must not lag 2 s on every scene change if the gateway is
+// offline). `arg` is a heap copy of the scene, freed here. Best-effort.
+static void gw_notify_task(void *arg) {
+    char *scene = (char *) arg;
+    char url[256];
+    snprintf(url, sizeof(url), "%s/game/step?scene=%s", s_gw_url, scene);
+    esp_http_client_config_t cfg = {
+        .url = url,
+        .method = HTTP_METHOD_POST,
+        .timeout_ms = 2000,
+    };
+    esp_http_client_handle_t cli = esp_http_client_init(&cfg);
+    if (cli) {
+        if (s_gw_token[0]) {
+            char auth[112];
+            snprintf(auth, sizeof(auth), "Bearer %s", s_gw_token);
+            esp_http_client_set_header(cli, "Authorization", auth);
+        }
+        esp_err_t err = esp_http_client_perform(cli);
+        if (err != ESP_OK) {
+            ESP_LOGW(TAG, "gateway /game/step notify failed: %s", esp_err_to_name(err));
+        } else {
+            ESP_LOGI(TAG, "gateway scene -> %s (HTTP %d)", scene,
+                     esp_http_client_get_status_code(cli));
+        }
+        esp_http_client_cleanup(cli);
+    }
+    free(scene);
+    vTaskDelete(NULL);
+}
+
+// Fire-and-forget the gateway scene notification on a detached task so a scene
+// change is never blocked by the network. No-op if no gateway configured.
+static void notify_gateway_scene(const char *scene) {
+    if (!s_gw_url[0] || !scene || !scene[0]) return;
+    size_t n = strlen(scene) + 1;
+    char *copy = malloc(n);
+    if (!copy) return;
+    memcpy(copy, scene, n);
+    if (xTaskCreate(gw_notify_task, "gw_notify", 4096, copy, 3, NULL) != pdPASS) {
+        ESP_LOGW(TAG, "gw_notify task spawn failed");
+        free(copy);
+    }
+}
+
 // ─── Core: scene/trigger/mood lookup tables (verbatim Arduino) ──────────────
 
 static const char *const kSceneIds[] = {
@@ -372,10 +434,25 @@ esp_err_t npc_engine_set_step(uint8_t step_id, uint32_t expected_duration_ms) {
                      (unsigned) prev_scene, (unsigned) step_id);
         }
         (void) hints_client_puzzle_start(puzzle_id);
+        // Keep the voice gateway in sync so the phone NPCs hint on THIS scene.
+        notify_gateway_scene(puzzle_id);
     }
     return ESP_OK;
 }
 
+esp_err_t npc_engine_set_scene_by_id(const char *scene_id,
+                                     uint32_t expected_duration_ms) {
+    if (scene_id == NULL || scene_id[0] == '\0') return ESP_ERR_INVALID_ARG;
+    for (uint8_t i = 0; i < kSceneCount; i++) {
+        if (strcmp(scene_id, kSceneIds[i]) == 0) {
+            return npc_engine_set_step(i, expected_duration_ms);
+        }
+    }
+    ESP_LOGW(TAG, "set_scene_by_id: unknown scene \"%s\" (no kSceneIds match)",
+             scene_id);
+    return ESP_ERR_NOT_FOUND;
+}
+
 esp_err_t npc_engine_set_group_profile(const char *profile) {
     // Thin pass-through. hints_client validates the value and logs the
     // outcome. Kept on npc_engine so the rest of the firmware doesn't

idf_zacus/components/ota_server/ota_server.c

@@ -310,8 +310,9 @@ esp_err_t ota_server_init(void) {
     // HTTP server config
     httpd_config_t config = HTTPD_DEFAULT_CONFIG();
     config.server_port     = OTA_SERVER_PORT;
-    config.max_uri_handlers = 20;   // ota + voice_hook + game (incl. /game/step,
-                                    // /game/puzzle_state, /game/file, relay)
+    config.max_uri_handlers = 24;   // ota + voice_hook + game (incl. /game/step,
+                                    // /game/puzzle_state, /game/file POST+DELETE,
+                                    // relay, /game/gamebook)
                                     // + headroom
     config.uri_match_fn    = httpd_uri_match_wildcard;
     config.stack_size      = 8192;

idf_zacus/components/p5_morse/CMakeLists.txt

@@ -0,0 +1,10 @@
+idf_component_register(
+    SRCS
+        "p5_morse.c"
+    INCLUDE_DIRS
+        "include"
+    REQUIRES
+        esp_driver_gpio
+        log
+        freertos
+)

idf_zacus/components/p5_morse/include/p5_morse.h

@@ -0,0 +1,68 @@
+// p5_morse.h — Morse / telegraph-key puzzle driver for Zacus master (P5).
+//
+// When CONFIG_ZACUS_P5_MORSE_ENABLE=n (default) all functions are empty stubs
+// (init returns ESP_OK, arm/disarm are no-ops) so callers need no #ifdef guards.
+//
+// When enabled:
+//   1. Call p5_morse_init() once at boot (after gpio driver is available).
+//   2. Call p5_morse_arm(expected, puzzle_id, fragment, frag_len) when the
+//      gateway POSTs /game/step with a "morse" puzzle.
+//   3. A FreeRTOS task decodes dot/dash timing on CONFIG_ZACUS_P5_GPIO and
+//      calls the solved_cb supplied to p5_morse_arm() when the decoded word
+//      matches `expected`.
+//   4. Call p5_morse_disarm() on step change or game reset.
+//
+// Timing constants are set via Kconfig (DOT_MS, DASH_MS, GAP_MS).
+// The key is active-LOW by default (CONFIG_ZACUS_P5_ACTIVE_LOW=y).
+
+#pragma once
+
+#include "esp_err.h"
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Callback type invoked on the p5_morse task when the sequence is solved.
+// Must be ISR-safe in the sense that it runs inside the morse task context;
+// do NOT call p5_morse_arm/disarm from it — defer via a task notification.
+typedef void (*p5_morse_solved_cb_t)(void);
+
+/**
+ * @brief Initialise the P5 Morse puzzle driver.
+ *
+ * Configures CONFIG_ZACUS_P5_GPIO as input with pull-up and spawns the
+ * morse decoding task (blocked until p5_morse_arm() is called).
+ * Idempotent — safe to call more than once.
+ *
+ * When CONFIG_ZACUS_P5_MORSE_ENABLE=n returns ESP_OK immediately.
+ *
+ * @return ESP_OK on success, or a driver error code.
+ */
+esp_err_t p5_morse_init(void);
+
+/**
+ * @brief Arm the morse puzzle for a specific expected sequence.
+ *
+ * @param expected   NUL-terminated Morse code string to match,
+ *                   e.g. "...-" (upper-case letters via kMorseTable)
+ *                   OR a plain word like "ZACUS" (decoded internally).
+ *                   The string is copied — caller buffer may be transient.
+ * @param solved_cb  Callback invoked exactly once when the sequence matches.
+ *                   Pass NULL to skip the notification (e.g. during tests).
+ *
+ * When CONFIG_ZACUS_P5_MORSE_ENABLE=n this is a no-op.
+ */
+void p5_morse_arm(const char *expected, p5_morse_solved_cb_t solved_cb);
+
+/**
+ * @brief Disarm the puzzle (stop decoding, reset state).
+ *
+ * When CONFIG_ZACUS_P5_MORSE_ENABLE=n this is a no-op.
+ */
+void p5_morse_disarm(void);
+
+#ifdef __cplusplus
+}
+#endif

idf_zacus/components/p5_morse/p5_morse.c

@@ -0,0 +1,265 @@
+// p5_morse.c — P5 Morse / telegraph-key puzzle driver.
+//
+// All real implementation is inside #if CONFIG_ZACUS_P5_MORSE_ENABLE so the
+// object compiles to pure stubs when the flag is off (default). No #ifdef
+// leaks into callers — they include p5_morse.h and call unconditionally.
+//
+// Logic adapted from puzzles/p5_morse/main/main.c (standalone ESP-NOW
+// firmware), repackaged as a library with an arm/disarm API and a solved
+// callback instead of the espnow_slave_notify_solved path.
+
+#include "p5_morse.h"
+
+#include "sdkconfig.h"
+#include "esp_log.h"
+#include "esp_err.h"
+
+static const char *TAG = "p5_morse";
+
+#if CONFIG_ZACUS_P5_MORSE_ENABLE
+
+#include "driver/gpio.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include <string.h>
+#include <stdbool.h>
+
+// ─── Morse table A-Z ────────────────────────────────────────────────────────
+
+static const char *kMorseTable[26] = {
+    ".-",    "-...",  "-.-.",  "-..",   ".",     "..-.",  "--.",
+    "....",  "..",    ".---",  "-.-",   ".-..",  "--",    "-.",
+    "---",   ".--.",  "--.-",  ".-.",   "...",   "-",     "..-",
+    "...-",  ".--",   "-..-",  "-.--",  "--.."
+};
+
+// ─── State ──────────────────────────────────────────────────────────────────
+
+// Expected word (decoded as letters, uppercase, e.g. "ZACUS").
+static char s_expected[32]       = {0};
+// Accumulator for letters decoded so far.
+static char s_received[32]       = {0};
+static uint8_t s_recv_pos        = 0;
+// Current symbol dot/dash buffer.
+static char s_symbol[16]         = {0};
+static uint8_t s_sym_pos         = 0;
+
+static volatile bool s_armed     = false;
+static volatile bool s_solved    = false;
+
+static p5_morse_solved_cb_t s_solved_cb = NULL;
+
+static bool s_initialised = false;
+
+static TaskHandle_t s_task_handle = NULL;
+
+// ─── Helpers ────────────────────────────────────────────────────────────────
+
+static uint32_t now_ms(void)
+{
+    return (uint32_t)(xTaskGetTickCount() * portTICK_PERIOD_MS);
+}
+
+static bool key_pressed(void)
+{
+#if CONFIG_ZACUS_P5_ACTIVE_LOW
+    return (gpio_get_level(CONFIG_ZACUS_P5_GPIO) == 0);
+#else
+    return (gpio_get_level(CONFIG_ZACUS_P5_GPIO) != 0);
+#endif
+}
+
+// Decode accumulated symbol (dot/dash string) to a letter and append it.
+static void decode_symbol(void)
+{
+    if (s_sym_pos == 0) return;
+    s_symbol[s_sym_pos] = '\0';
+
+    for (int i = 0; i < 26; i++) {
+        if (strcmp(kMorseTable[i], s_symbol) == 0) {
+            char letter = (char)('A' + i);
+            if (s_recv_pos < (uint8_t)(sizeof(s_received) - 1)) {
+                s_received[s_recv_pos++] = letter;
+                s_received[s_recv_pos]   = '\0';
+            }
+            ESP_LOGD(TAG, "decoded: %s → %c (so far: %s)",
+                     s_symbol, letter, s_received);
+            break;
+        }
+    }
+    s_sym_pos = 0;
+    memset(s_symbol, 0, sizeof(s_symbol));
+}
+
+// Check whether the received word matches; invoke callback on match.
+static void check_word(void)
+{
+    if (strcmp(s_received, s_expected) == 0) {
+        s_solved = true;
+        ESP_LOGI(TAG, "SOLVED: \"%s\" decoded correctly", s_expected);
+        if (s_solved_cb) {
+            s_solved_cb();
+        }
+    } else {
+        ESP_LOGW(TAG, "wrong word: got \"%s\", expected \"%s\" — resetting",
+                 s_received, s_expected);
+        s_recv_pos = 0;
+        memset(s_received, 0, sizeof(s_received));
+    }
+}
+
+// ─── Morse decoding task ─────────────────────────────────────────────────────
+
+static void morse_task(void *arg)
+{
+    (void)arg;
+
+    uint32_t press_start  = 0;
+    uint32_t last_release = 0;
+    bool     key_down     = false;
+
+    for (;;) {
+        if (!s_armed || s_solved) {
+            // Idle: yield and wait for the next arm cycle.
+            vTaskDelay(pdMS_TO_TICKS(50));
+            continue;
+        }
+
+        uint32_t t       = now_ms();
+        bool     pressed = key_pressed();
+
+        if (pressed && !key_down) {
+            // Key press start.
+            key_down    = true;
+            press_start = t;
+
+        } else if (!pressed && key_down) {
+            // Key release — classify as dot or dash.
+            key_down = false;
+            uint32_t duration = t - press_start;
+            last_release = t;
+
+            if (duration <= (uint32_t)CONFIG_ZACUS_P5_DOT_MS) {
+                if (s_sym_pos < (uint8_t)(sizeof(s_symbol) - 1))
+                    s_symbol[s_sym_pos++] = '.';
+                ESP_LOGD(TAG, "dot (%lu ms)", (unsigned long)duration);
+            } else if (duration >= (uint32_t)CONFIG_ZACUS_P5_DASH_MS) {
+                if (s_sym_pos < (uint8_t)(sizeof(s_symbol) - 1))
+                    s_symbol[s_sym_pos++] = '-';
+                ESP_LOGD(TAG, "dash (%lu ms)", (unsigned long)duration);
+            } else {
+                ESP_LOGD(TAG, "ignored press %lu ms (between dot and dash thresholds)",
+                         (unsigned long)duration);
+            }
+
+        } else if (!key_down && last_release > 0) {
+            uint32_t silence = t - last_release;
+
+            // Letter gap: decode accumulated symbol.
+            if (silence > (uint32_t)CONFIG_ZACUS_P5_GAP_MS && s_sym_pos > 0) {
+                decode_symbol();
+                last_release = t;
+            }
+            // Word gap (2x letter gap): check assembled word.
+            if (silence > (uint32_t)(CONFIG_ZACUS_P5_GAP_MS * 2u) && s_recv_pos > 0) {
+                check_word();
+                last_release = 0;
+            }
+        }
+
+        vTaskDelay(pdMS_TO_TICKS(10));   // 100 Hz sampling
+    }
+}
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+esp_err_t p5_morse_init(void)
+{
+    if (s_initialised) return ESP_OK;
+
+    gpio_config_t cfg = {
+        .pin_bit_mask = (1ULL << CONFIG_ZACUS_P5_GPIO),
+        .mode         = GPIO_MODE_INPUT,
+        .pull_up_en   = GPIO_PULLUP_ENABLE,
+        .pull_down_en = GPIO_PULLDOWN_DISABLE,
+        .intr_type    = GPIO_INTR_DISABLE,
+    };
+    esp_err_t err = gpio_config(&cfg);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "gpio_config(GPIO %d): %s",
+                 CONFIG_ZACUS_P5_GPIO, esp_err_to_name(err));
+        return err;
+    }
+
+    BaseType_t rc = xTaskCreate(morse_task, "p5_morse", 3072, NULL, 5,
+                                &s_task_handle);
+    if (rc != pdPASS) {
+        ESP_LOGE(TAG, "xTaskCreate(p5_morse) failed");
+        return ESP_ERR_NO_MEM;
+    }
+
+    s_initialised = true;
+    ESP_LOGI(TAG, "init OK (GPIO %d, dot<=%d ms, dash>=%d ms, gap>%d ms)",
+             CONFIG_ZACUS_P5_GPIO,
+             CONFIG_ZACUS_P5_DOT_MS,
+             CONFIG_ZACUS_P5_DASH_MS,
+             CONFIG_ZACUS_P5_GAP_MS);
+    return ESP_OK;
+}
+
+void p5_morse_arm(const char *expected, p5_morse_solved_cb_t solved_cb)
+{
+    if (!s_initialised) {
+        ESP_LOGW(TAG, "arm called before init — ignored");
+        return;
+    }
+    // Reset state.
+    s_armed    = false;   // pause task briefly while we reset
+    s_solved   = false;
+    s_recv_pos = 0;
+    s_sym_pos  = 0;
+    memset(s_received, 0, sizeof(s_received));
+    memset(s_symbol,   0, sizeof(s_symbol));
+
+    if (expected && expected[0]) {
+        strncpy(s_expected, expected, sizeof(s_expected) - 1);
+        s_expected[sizeof(s_expected) - 1] = '\0';
+    } else {
+        s_expected[0] = '\0';
+    }
+    s_solved_cb = solved_cb;
+    s_armed     = true;
+    ESP_LOGI(TAG, "armed, expected \"%s\"", s_expected);
+}
+
+void p5_morse_disarm(void)
+{
+    s_armed  = false;
+    s_solved = false;
+    s_recv_pos = 0;
+    s_sym_pos  = 0;
+    memset(s_received, 0, sizeof(s_received));
+    memset(s_symbol,   0, sizeof(s_symbol));
+    ESP_LOGI(TAG, "disarmed");
+}
+
+#else // CONFIG_ZACUS_P5_MORSE_ENABLE not set — emit stubs only
+
+esp_err_t p5_morse_init(void)
+{
+    return ESP_OK;
+}
+
+void p5_morse_arm(const char *expected, p5_morse_solved_cb_t solved_cb)
+{
+    (void)expected;
+    (void)solved_cb;
+    /* stub — CONFIG_ZACUS_P5_MORSE_ENABLE=n */
+}
+
+void p5_morse_disarm(void)
+{
+    /* stub — CONFIG_ZACUS_P5_MORSE_ENABLE=n */
+}
+
+#endif // CONFIG_ZACUS_P5_MORSE_ENABLE

idf_zacus/components/p6_nfc/CMakeLists.txt

@@ -0,0 +1,12 @@
+idf_component_register(
+    SRCS
+        "p6_nfc.c"
+    INCLUDE_DIRS
+        "include"
+    REQUIRES
+        esp_driver_gpio
+        esp_driver_i2c
+        esp_driver_spi
+        log
+        freertos
+)

idf_zacus/components/p6_nfc/include/p6_nfc.h

@@ -0,0 +1,72 @@
+// p6_nfc.h — NFC / alchemical-symbols puzzle driver for Zacus master (P6).
+//
+// When CONFIG_ZACUS_P6_NFC_ENABLE=n (default) all functions are empty stubs
+// (init returns ESP_OK, arm/disarm are no-ops) so callers need no #ifdef guards.
+//
+// When enabled:
+//   1. Call p6_nfc_init() once at boot.
+//   2. Call p6_nfc_arm(expected_uid, solved_cb) when the gateway POSTs
+//      /game/step with a "nfc" puzzle. `expected_uid` is a hex UID string
+//      like "A1:B2:C3:D4" (4-byte NTAG213) or a space-separated sequence
+//      of UIDs for ordered multi-tag placement.
+//   3. A FreeRTOS task polls the NFC reader at ~5 Hz and calls solved_cb
+//      when the presented tag UID matches (or the full sequence is placed).
+//   4. Call p6_nfc_disarm() on step change or game reset.
+//
+// Hardware backend is selected via Kconfig:
+//   PN532 over I2C or SPI  → CONFIG_ZACUS_P6_READER_PN532
+//   PN7150 over I2C        → CONFIG_ZACUS_P6_READER_PN7150
+//
+// The hardware HAL is STUBBED in this scaffold revision: the read function
+// compiles, links, and logs "NFC read (stub)" but returns no UIDs until the
+// real HAL is wired in. This is the same pattern as media_manager stubs.
+
+#pragma once
+
+#include "esp_err.h"
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Callback type invoked on the p6_nfc task when the tag / sequence is solved.
+// Must not call p6_nfc_arm/disarm from within — defer via task notification.
+typedef void (*p6_nfc_solved_cb_t)(void);
+
+/**
+ * @brief Initialise the P6 NFC puzzle driver.
+ *
+ * Initialises the configured bus (I2C or SPI) and the NFC reader IC.
+ * Spawns the NFC polling task (blocked until p6_nfc_arm() is called).
+ * Idempotent — safe to call more than once.
+ *
+ * When CONFIG_ZACUS_P6_NFC_ENABLE=n returns ESP_OK immediately.
+ *
+ * @return ESP_OK on success, or a driver error code.
+ */
+esp_err_t p6_nfc_init(void);
+
+/**
+ * @brief Arm the NFC puzzle for a specific expected tag UID or sequence.
+ *
+ * @param expected_uid  NUL-terminated UID string to match. Format:
+ *                      single tag  → "A1:B2:C3:D4"
+ *                      The string is copied — caller buffer may be transient.
+ * @param solved_cb     Callback invoked exactly once on a match.
+ *                      Pass NULL to skip (e.g. during tests).
+ *
+ * When CONFIG_ZACUS_P6_NFC_ENABLE=n this is a no-op.
+ */
+void p6_nfc_arm(const char *expected_uid, p6_nfc_solved_cb_t solved_cb);
+
+/**
+ * @brief Disarm the puzzle (stop polling, reset state).
+ *
+ * When CONFIG_ZACUS_P6_NFC_ENABLE=n this is a no-op.
+ */
+void p6_nfc_disarm(void);
+
+#ifdef __cplusplus
+}
+#endif

idf_zacus/components/p6_nfc/p6_nfc.c

@@ -0,0 +1,290 @@
+// p6_nfc.c — P6 NFC / alchemical-symbols puzzle driver.
+//
+// All real implementation is inside #if CONFIG_ZACUS_P6_NFC_ENABLE so the
+// object compiles to pure stubs when the flag is off (default). No #ifdef
+// leaks into callers.
+//
+// Hardware HAL for the NFC reader is STUBBED: nfc_hal_read_uid() returns
+// false (no card present) and logs once per arm cycle. The real HAL will be
+// implemented when the PN532/PN7150 module is wired (replace the stub block
+// marked with "STUB" below).
+//
+// Logic adapted from puzzles/p6_symboles_nfc/main/main.c, repackaged as a
+// library with arm/disarm API and a solved callback.
+
+#include "p6_nfc.h"
+
+#include "sdkconfig.h"
+#include "esp_log.h"
+#include "esp_err.h"
+
+static const char *TAG = "p6_nfc";
+
+#if CONFIG_ZACUS_P6_NFC_ENABLE
+
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include <string.h>
+#include <stdbool.h>
+#include <stdio.h>
+
+// ─── I2C / SPI bus init (HAL layer) ─────────────────────────────────────────
+//
+// The bus init and NFC read are separated into nfc_hal_init() /
+// nfc_hal_read_uid() so the stub is easy to swap for real silicon code.
+
+#if CONFIG_ZACUS_P6_BUS_I2C
+#include "driver/i2c_master.h"
+
+static i2c_master_bus_handle_t s_i2c_bus   = NULL;
+static i2c_master_dev_handle_t s_i2c_dev   = NULL;
+
+static esp_err_t nfc_hal_init(void)
+{
+    i2c_master_bus_config_t bus_cfg = {
+        .i2c_port      = I2C_NUM_0,
+        .sda_io_num    = CONFIG_ZACUS_P6_I2C_SDA,
+        .scl_io_num    = CONFIG_ZACUS_P6_I2C_SCL,
+        .clk_source    = I2C_CLK_SRC_DEFAULT,
+        .glitch_ignore_cnt = 7,
+        .flags.enable_internal_pullup = true,
+    };
+    esp_err_t err = i2c_new_master_bus(&bus_cfg, &s_i2c_bus);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "i2c_new_master_bus: %s", esp_err_to_name(err));
+        return err;
+    }
+
+    i2c_device_config_t dev_cfg = {
+        .dev_addr_length = I2C_ADDR_BIT_LEN_7,
+        .device_address  = (uint16_t)CONFIG_ZACUS_P6_I2C_ADDR,
+        .scl_speed_hz    = 100000,
+    };
+    err = i2c_master_bus_add_device(s_i2c_bus, &dev_cfg, &s_i2c_dev);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "i2c_master_bus_add_device(addr=0x%02X): %s",
+                 CONFIG_ZACUS_P6_I2C_ADDR, esp_err_to_name(err));
+        return err;
+    }
+
+    ESP_LOGI(TAG, "I2C bus init OK (SDA=%d SCL=%d addr=0x%02X)",
+             CONFIG_ZACUS_P6_I2C_SDA, CONFIG_ZACUS_P6_I2C_SCL,
+             CONFIG_ZACUS_P6_I2C_ADDR);
+    return ESP_OK;
+}
+
+// STUB: real I2C read of PN532/PN7150 comes here.
+// Returns true and fills uid_out[4] when a card is present.
+static bool nfc_hal_read_uid(uint8_t uid_out[4])
+{
+    (void)uid_out;
+    // HAL stub — hardware not yet wired.  Replace with PN532/PN7150 I2C
+    // read sequence once the module is connected.
+    ESP_LOGD(TAG, "NFC read (stub, I2C 0x%02X)", CONFIG_ZACUS_P6_I2C_ADDR);
+    return false;
+}
+
+#elif CONFIG_ZACUS_P6_BUS_SPI
+#include "driver/spi_master.h"
+
+static spi_device_handle_t s_spi = NULL;
+
+static esp_err_t nfc_hal_init(void)
+{
+    spi_bus_config_t buscfg = {
+        .miso_io_num     = 13,   // default SPI2 pins — override via sdkconfig if needed
+        .mosi_io_num     = 11,
+        .sclk_io_num     = 12,
+        .quadwp_io_num   = -1,
+        .quadhd_io_num   = -1,
+        .max_transfer_sz = 64,
+    };
+    // Use SPI3_HOST to avoid conflict with possible SD card on SPI2.
+    esp_err_t err = spi_bus_initialize(SPI3_HOST, &buscfg, SPI_DMA_CH_AUTO);
+    if (err != ESP_OK && err != ESP_ERR_INVALID_STATE) {
+        ESP_LOGE(TAG, "spi_bus_initialize(SPI3): %s", esp_err_to_name(err));
+        return err;
+    }
+
+    spi_device_interface_config_t devcfg = {
+        .clock_speed_hz = 1000000,   // 1 MHz (PN532 max SPI is 5 MHz)
+        .mode           = 0,
+        .spics_io_num   = 10,        // CS pin — adjust via Kconfig when available
+        .queue_size     = 4,
+    };
+    err = spi_bus_add_device(SPI3_HOST, &devcfg, &s_spi);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "spi_bus_add_device: %s", esp_err_to_name(err));
+        return err;
+    }
+
+    ESP_LOGI(TAG, "SPI bus init OK (SPI3, CS=10)");
+    return ESP_OK;
+}
+
+// STUB: real SPI read of PN532 comes here.
+static bool nfc_hal_read_uid(uint8_t uid_out[4])
+{
+    (void)uid_out;
+    ESP_LOGD(TAG, "NFC read (stub, SPI)");
+    return false;
+}
+
+#endif // bus type
+
+// ─── UID comparison helper ───────────────────────────────────────────────────
+
+// Parse a colon-separated hex UID string "A1:B2:C3:D4" into uid[4].
+// Returns true on success.
+static bool parse_uid(const char *str, uint8_t uid[4])
+{
+    unsigned int b[4];
+    if (sscanf(str, "%x:%x:%x:%x", &b[0], &b[1], &b[2], &b[3]) != 4)
+        return false;
+    for (int i = 0; i < 4; i++) uid[i] = (uint8_t)b[i];
+    return true;
+}
+
+// ─── State ──────────────────────────────────────────────────────────────────
+
+static char s_expected_uid[20]  = {0};   // "A1:B2:C3:D4\0"
+static uint8_t s_expected_raw[4] = {0};
+static bool s_expected_valid    = false;
+
+static volatile bool s_armed    = false;
+static volatile bool s_solved   = false;
+
+static p6_nfc_solved_cb_t s_solved_cb = NULL;
+
+static bool s_initialised = false;
+
+// ─── NFC polling task ────────────────────────────────────────────────────────
+
+static void nfc_task(void *arg)
+{
+    (void)arg;
+
+    uint8_t uid[4]      = {0};
+    uint8_t last_uid[4] = {0};
+    uint32_t last_read_ms = 0;
+
+    for (;;) {
+        if (!s_armed || s_solved) {
+            vTaskDelay(pdMS_TO_TICKS(200));
+            continue;
+        }
+
+        uint32_t t = (uint32_t)(xTaskGetTickCount() * portTICK_PERIOD_MS);
+
+        if (nfc_hal_read_uid(uid)) {
+            // Debounce: ignore same tag within 1 s.
+            bool same = (memcmp(uid, last_uid, 4) == 0);
+            if (!same || (t - last_read_ms) > 1000) {
+                memcpy(last_uid, uid, 4);
+                last_read_ms = t;
+
+                ESP_LOGI(TAG, "tag read: %02X:%02X:%02X:%02X",
+                         uid[0], uid[1], uid[2], uid[3]);
+
+                if (s_expected_valid &&
+                    memcmp(uid, s_expected_raw, 4) == 0) {
+                    s_solved = true;
+                    ESP_LOGI(TAG, "SOLVED: tag %s matched", s_expected_uid);
+                    if (s_solved_cb) {
+                        s_solved_cb();
+                    }
+                } else {
+                    ESP_LOGW(TAG, "tag %02X:%02X:%02X:%02X not the expected %s",
+                             uid[0], uid[1], uid[2], uid[3], s_expected_uid);
+                }
+            }
+        }
+
+        vTaskDelay(pdMS_TO_TICKS(200));   // 5 Hz polling
+    }
+}
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+esp_err_t p6_nfc_init(void)
+{
+    if (s_initialised) return ESP_OK;
+
+    esp_err_t err = nfc_hal_init();
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "nfc_hal_init failed: %s", esp_err_to_name(err));
+        return err;
+    }
+
+    BaseType_t rc = xTaskCreate(nfc_task, "p6_nfc", 3072, NULL, 5, NULL);
+    if (rc != pdPASS) {
+        ESP_LOGE(TAG, "xTaskCreate(p6_nfc) failed");
+        return ESP_ERR_NO_MEM;
+    }
+
+    s_initialised = true;
+#if CONFIG_ZACUS_P6_READER_PN532 && CONFIG_ZACUS_P6_BUS_I2C
+    ESP_LOGI(TAG, "init OK (PN532, I2C, stub HAL)");
+#elif CONFIG_ZACUS_P6_READER_PN532
+    ESP_LOGI(TAG, "init OK (PN532, SPI, stub HAL)");
+#else
+    ESP_LOGI(TAG, "init OK (PN7150, I2C, stub HAL)");
+#endif
+    return ESP_OK;
+}
+
+void p6_nfc_arm(const char *expected_uid, p6_nfc_solved_cb_t solved_cb)
+{
+    if (!s_initialised) {
+        ESP_LOGW(TAG, "arm called before init — ignored");
+        return;
+    }
+
+    s_armed    = false;
+    s_solved   = false;
+    s_expected_valid = false;
+    memset(s_expected_uid, 0, sizeof(s_expected_uid));
+    memset(s_expected_raw, 0, sizeof(s_expected_raw));
+
+    if (expected_uid && expected_uid[0]) {
+        strncpy(s_expected_uid, expected_uid, sizeof(s_expected_uid) - 1);
+        s_expected_uid[sizeof(s_expected_uid) - 1] = '\0';
+        s_expected_valid = parse_uid(s_expected_uid, s_expected_raw);
+        if (!s_expected_valid) {
+            ESP_LOGW(TAG, "arm: could not parse UID \"%s\" — tag match disabled",
+                     expected_uid);
+        }
+    }
+
+    s_solved_cb = solved_cb;
+    s_armed     = true;
+    ESP_LOGI(TAG, "armed, expected UID \"%s\"", s_expected_uid);
+}
+
+void p6_nfc_disarm(void)
+{
+    s_armed  = false;
+    s_solved = false;
+    ESP_LOGI(TAG, "disarmed");
+}
+
+#else // CONFIG_ZACUS_P6_NFC_ENABLE not set — emit stubs only
+
+esp_err_t p6_nfc_init(void)
+{
+    return ESP_OK;
+}
+
+void p6_nfc_arm(const char *expected_uid, p6_nfc_solved_cb_t solved_cb)
+{
+    (void)expected_uid;
+    (void)solved_cb;
+    /* stub — CONFIG_ZACUS_P6_NFC_ENABLE=n */
+}
+
+void p6_nfc_disarm(void)
+{
+    /* stub — CONFIG_ZACUS_P6_NFC_ENABLE=n */
+}
+
+#endif // CONFIG_ZACUS_P6_NFC_ENABLE

idf_zacus/components/p7_coffre/CMakeLists.txt

@@ -0,0 +1,11 @@
+idf_component_register(
+    SRCS
+        "p7_coffre.c"
+    INCLUDE_DIRS
+        "include"
+    REQUIRES
+        esp_driver_ledc
+        esp_driver_gpio
+        log
+        freertos
+)

idf_zacus/components/p7_coffre/include/p7_coffre.h

@@ -0,0 +1,60 @@
+// p7_coffre — unlocking actuator for the Zacus final chest (P7).
+//
+// When CONFIG_ZACUS_P7_COFFRE_ENABLE=n (default) all three functions are
+// empty stubs (init returns ESP_OK, unlock/lock are no-ops) so callers do
+// not need #ifdef guards.
+//
+// When enabled, call p7_coffre_init() once at boot, then p7_coffre_unlock()
+// when STEP_FINAL_WIN is reached.  p7_coffre_lock() re-arms between games.
+//
+// Actuator selection and GPIO are configured via Kconfig (menuconfig or
+// sdkconfig):
+//   SERVO  – LEDC TIMER_2 / CHANNEL_2, 50 Hz, 13-bit duty cycle.
+//   RELAY  – plain GPIO output with optional one-shot pulse.
+
+#pragma once
+
+#include "esp_err.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @brief Initialise the P7 coffre actuator.
+ *
+ * Servo: configures LEDC timer + channel and moves to the locked position.
+ * Relay: configures the GPIO as output and asserts the rest (inactive) state.
+ * Idempotent — safe to call more than once.
+ *
+ * When CONFIG_ZACUS_P7_COFFRE_ENABLE=n this is a stub that returns ESP_OK.
+ *
+ * @return ESP_OK on success, or a driver error code.
+ */
+esp_err_t p7_coffre_init(void);
+
+/**
+ * @brief Unlock the coffre (STEP_FINAL_WIN reached).
+ *
+ * Servo: moves to CONFIG_ZACUS_P7_SERVO_UNLOCK_DEG.
+ * Relay: energises the coil; if CONFIG_ZACUS_P7_RELAY_PULSE_MS > 0 the coil
+ *        is de-energised after that many milliseconds (blocking call on the
+ *        caller's task — keep short; 800 ms default).
+ *
+ * When CONFIG_ZACUS_P7_COFFRE_ENABLE=n this is a no-op.
+ */
+void p7_coffre_unlock(void);
+
+/**
+ * @brief Re-arm the coffre to the locked position (between games).
+ *
+ * Servo: moves back to CONFIG_ZACUS_P7_SERVO_LOCK_DEG.
+ * Relay: de-energises the coil (if not already de-energised by the pulse).
+ *
+ * When CONFIG_ZACUS_P7_COFFRE_ENABLE=n this is a no-op.
+ */
+void p7_coffre_lock(void);
+
+#ifdef __cplusplus
+}
+#endif

idf_zacus/components/p7_coffre/p7_coffre.c

@@ -0,0 +1,232 @@
+// p7_coffre — P7 coffre unlocking actuator driver.
+//
+// All real implementation is inside #if CONFIG_ZACUS_P7_COFFRE_ENABLE so the
+// object compiles to pure stubs when the flag is off (default).  No #ifdef
+// leaks into callers — they include p7_coffre.h and call the three functions
+// unconditionally.
+
+#include "p7_coffre.h"
+
+#include "sdkconfig.h"
+#include "esp_log.h"
+#include "esp_err.h"
+
+static const char *TAG = "p7_coffre";
+
+// ─── Servo duty helper ──────────────────────────────────────────────────────
+//
+// Maps an angle in degrees [0, 180] to a LEDC duty value.
+//
+// Timer parameters (compile-time constants):
+//   Period       : 20 ms  (50 Hz)
+//   Pulse range  : 500 µs (0°) … 2500 µs (180°)  — standard SG90 / MG90S
+//   Resolution   : 13 bits → 8192 counts per 20 ms
+//
+//   counts_per_us = 8192 / 20000 = 0.4096
+//   duty(0°)   = 500  * 0.4096 =  205
+//   duty(90°)  = 1500 * 0.4096 =  614
+//   duty(180°) = 2500 * 0.4096 = 1024
+
+#if CONFIG_ZACUS_P7_COFFRE_ENABLE
+
+#include "driver/ledc.h"
+#include "driver/gpio.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+
+// LEDC resources — chosen to avoid conflicts with existing assignments:
+//   TIMER_0 / CHANNEL_0  =  camera XCLK  (qr_puzzle)
+//   TIMER_1 / CHANNEL_1  =  display backlight  (display_ui)
+//   TIMER_2 / CHANNEL_2  =  P7 servo (this file)
+#define P7_LEDC_TIMER    LEDC_TIMER_2
+#define P7_LEDC_CHANNEL  LEDC_CHANNEL_2
+#define P7_LEDC_MODE     LEDC_LOW_SPEED_MODE
+#define P7_LEDC_RES      LEDC_TIMER_13_BIT   // 8192 counts per period
+#define P7_LEDC_FREQ_HZ  50u                 // 20 ms period
+
+// counts = pulse_us * (2^13 / 20000)
+// Use integer arithmetic: counts = pulse_us * 8192 / 20000
+#define P7_SERVO_COUNTS(pulse_us)  ((uint32_t)(pulse_us) * 8192u / 20000u)
+
+// Pulse width at each extreme (SG90-compatible).
+#define P7_SERVO_PULSE_MIN_US  500u    // 0°
+#define P7_SERVO_PULSE_MAX_US  2500u   // 180°
+
+// Convert angle [0, 180] to duty count.
+static uint32_t angle_to_duty(int deg) {
+    if (deg < 0)   deg = 0;
+    if (deg > 180) deg = 180;
+    uint32_t pulse_us = P7_SERVO_PULSE_MIN_US +
+                        (uint32_t) deg *
+                        (P7_SERVO_PULSE_MAX_US - P7_SERVO_PULSE_MIN_US) / 180u;
+    return P7_SERVO_COUNTS(pulse_us);
+}
+
+static bool s_initialised = false;
+
+// ─── Servo path ─────────────────────────────────────────────────────────────
+
+#if CONFIG_ZACUS_P7_ACTUATOR_SERVO
+
+static esp_err_t servo_set_angle(int deg) {
+    uint32_t duty = angle_to_duty(deg);
+    esp_err_t err = ledc_set_duty(P7_LEDC_MODE, P7_LEDC_CHANNEL, duty);
+    if (err != ESP_OK) return err;
+    return ledc_update_duty(P7_LEDC_MODE, P7_LEDC_CHANNEL);
+}
+
+esp_err_t p7_coffre_init(void) {
+    if (s_initialised) return ESP_OK;
+
+    // Configure the 50 Hz timer.
+    ledc_timer_config_t timer_cfg = {
+        .speed_mode      = P7_LEDC_MODE,
+        .duty_resolution = P7_LEDC_RES,
+        .timer_num       = P7_LEDC_TIMER,
+        .freq_hz         = P7_LEDC_FREQ_HZ,
+        .clk_cfg         = LEDC_AUTO_CLK,
+    };
+    esp_err_t err = ledc_timer_config(&timer_cfg);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "ledc_timer_config: %s", esp_err_to_name(err));
+        return err;
+    }
+
+    // Bind the channel to the GPIO at the lock angle.
+    uint32_t lock_duty = angle_to_duty(CONFIG_ZACUS_P7_SERVO_LOCK_DEG);
+    ledc_channel_config_t ch_cfg = {
+        .gpio_num   = CONFIG_ZACUS_P7_GPIO,
+        .speed_mode = P7_LEDC_MODE,
+        .channel    = P7_LEDC_CHANNEL,
+        .intr_type  = LEDC_INTR_DISABLE,
+        .timer_sel  = P7_LEDC_TIMER,
+        .duty       = lock_duty,
+        .hpoint     = 0,
+    };
+    err = ledc_channel_config(&ch_cfg);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "ledc_channel_config: %s", esp_err_to_name(err));
+        return err;
+    }
+
+    s_initialised = true;
+    ESP_LOGI(TAG, "servo init OK (GPIO %d, lock=%d°, unlock=%d°, "
+                  "LEDC TIMER_%d/CH_%d 50 Hz 13-bit)",
+             CONFIG_ZACUS_P7_GPIO,
+             CONFIG_ZACUS_P7_SERVO_LOCK_DEG,
+             CONFIG_ZACUS_P7_SERVO_UNLOCK_DEG,
+             (int) P7_LEDC_TIMER, (int) P7_LEDC_CHANNEL);
+    return ESP_OK;
+}
+
+void p7_coffre_unlock(void) {
+    if (!s_initialised) {
+        ESP_LOGW(TAG, "unlock called before init — ignored");
+        return;
+    }
+    esp_err_t err = servo_set_angle(CONFIG_ZACUS_P7_SERVO_UNLOCK_DEG);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "servo_set_angle(%d): %s",
+                 CONFIG_ZACUS_P7_SERVO_UNLOCK_DEG, esp_err_to_name(err));
+    } else {
+        ESP_LOGI(TAG, "coffre UNLOCKED (servo -> %d deg)",
+                 CONFIG_ZACUS_P7_SERVO_UNLOCK_DEG);
+    }
+}
+
+void p7_coffre_lock(void) {
+    if (!s_initialised) return;
+    esp_err_t err = servo_set_angle(CONFIG_ZACUS_P7_SERVO_LOCK_DEG);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "servo_set_angle(%d): %s",
+                 CONFIG_ZACUS_P7_SERVO_LOCK_DEG, esp_err_to_name(err));
+    } else {
+        ESP_LOGI(TAG, "coffre re-LOCKED (servo -> %d deg)",
+                 CONFIG_ZACUS_P7_SERVO_LOCK_DEG);
+    }
+}
+
+#endif // CONFIG_ZACUS_P7_ACTUATOR_SERVO
+
+// ─── Relay path ─────────────────────────────────────────────────────────────
+
+#if CONFIG_ZACUS_P7_ACTUATOR_RELAY
+
+// Helper: set the relay coil to active (energised) or rest.
+// Handles active-high vs active-low inversion.
+static void relay_set(bool active) {
+    int level;
+#if CONFIG_ZACUS_P7_RELAY_ACTIVE_HIGH
+    level = active ? 1 : 0;
+#else
+    level = active ? 0 : 1;
+#endif
+    gpio_set_level(CONFIG_ZACUS_P7_GPIO, level);
+}
+
+esp_err_t p7_coffre_init(void) {
+    if (s_initialised) return ESP_OK;
+
+    gpio_config_t cfg = {
+        .pin_bit_mask = (1ULL << CONFIG_ZACUS_P7_GPIO),
+        .mode         = GPIO_MODE_OUTPUT,
+        .pull_up_en   = GPIO_PULLUP_DISABLE,
+        .pull_down_en = GPIO_PULLDOWN_DISABLE,
+        .intr_type    = GPIO_INTR_DISABLE,
+    };
+    esp_err_t err = gpio_config(&cfg);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "gpio_config(GPIO %d): %s",
+                 CONFIG_ZACUS_P7_GPIO, esp_err_to_name(err));
+        return err;
+    }
+
+    // Start in rest (de-energised) state.
+    relay_set(false);
+    s_initialised = true;
+    ESP_LOGI(TAG, "relay init OK (GPIO %d, active-%s, pulse=%d ms)",
+             CONFIG_ZACUS_P7_GPIO,
+             CONFIG_ZACUS_P7_RELAY_ACTIVE_HIGH ? "HIGH" : "LOW",
+             CONFIG_ZACUS_P7_RELAY_PULSE_MS);
+    return ESP_OK;
+}
+
+void p7_coffre_unlock(void) {
+    if (!s_initialised) {
+        ESP_LOGW(TAG, "unlock called before init — ignored");
+        return;
+    }
+    relay_set(true);
+    ESP_LOGI(TAG, "coffre UNLOCKED (relay energised)");
+
+#if CONFIG_ZACUS_P7_RELAY_PULSE_MS > 0
+    vTaskDelay(pdMS_TO_TICKS(CONFIG_ZACUS_P7_RELAY_PULSE_MS));
+    relay_set(false);
+    ESP_LOGI(TAG, "coffre relay de-energised after %d ms pulse",
+             CONFIG_ZACUS_P7_RELAY_PULSE_MS);
+#endif
+}
+
+void p7_coffre_lock(void) {
+    if (!s_initialised) return;
+    relay_set(false);
+    ESP_LOGI(TAG, "coffre relay de-energised (lock/re-arm)");
+}
+
+#endif // CONFIG_ZACUS_P7_ACTUATOR_RELAY
+
+#else // CONFIG_ZACUS_P7_COFFRE_ENABLE not set — emit stubs only
+
+esp_err_t p7_coffre_init(void) {
+    return ESP_OK;
+}
+
+void p7_coffre_unlock(void) {
+    /* stub — CONFIG_ZACUS_P7_COFFRE_ENABLE=n */
+}
+
+void p7_coffre_lock(void) {
+    /* stub — CONFIG_ZACUS_P7_COFFRE_ENABLE=n */
+}
+
+#endif // CONFIG_ZACUS_P7_COFFRE_ENABLE

idf_zacus/main/CMakeLists.txt

@@ -23,5 +23,9 @@ idf_component_register(
         esp_event
         espressif__mdns
         display_ui
+        gamebook
         puzzle_state
+        p7_coffre
+        p5_morse
+        p6_nfc
 )

idf_zacus/main/Kconfig.projbuild

@@ -0,0 +1,212 @@
+menu "Zacus Game Hardware"
+
+    config ZACUS_P7_COFFRE_ENABLE
+        bool "Enable P7 coffre actuator (servo or relay)"
+        default n
+        help
+            Drives the unlocking actuator for the final chest (P7) when
+            STEP_FINAL_WIN is reached. When disabled all p7_coffre_*()
+            functions are empty stubs so callers compile without #ifdef.
+            Enable once the actuator (9g servo or 5V relay) is wired up.
+
+    if ZACUS_P7_COFFRE_ENABLE
+
+        choice ZACUS_P7_ACTUATOR_TYPE
+            prompt "Actuator type"
+            default ZACUS_P7_ACTUATOR_SERVO
+            help
+                Select the hardware wired to CONFIG_ZACUS_P7_GPIO.
+                SERVO  – 50 Hz PWM via LEDC (TIMER_2 / CHANNEL_2).
+                RELAY  – logic-level GPIO output (active-high or active-low).
+
+            config ZACUS_P7_ACTUATOR_SERVO
+                bool "Servo PWM (9g, 50 Hz LEDC)"
+            config ZACUS_P7_ACTUATOR_RELAY
+                bool "Relay / solid-state switch (GPIO)"
+        endchoice
+
+        config ZACUS_P7_GPIO
+            int "GPIO pin for the actuator"
+            default 13
+            range 0 48
+            help
+                GPIO number for the servo signal wire or the relay control
+                input. Default 13 is free on the Freenove Media Kit board.
+
+        if ZACUS_P7_ACTUATOR_SERVO
+            config ZACUS_P7_SERVO_LOCK_DEG
+                int "Servo lock angle (degrees, 0-180)"
+                default 0
+                range 0 180
+                help
+                    Angle at which the servo holds the coffre locked.
+                    Corresponds to a 500-2500 µs pulse on a 20 ms period.
+
+            config ZACUS_P7_SERVO_UNLOCK_DEG
+                int "Servo unlock angle (degrees, 0-180)"
+                default 90
+                range 0 180
+                help
+                    Angle driven on STEP_FINAL_WIN to open the coffre.
+        endif
+
+        if ZACUS_P7_ACTUATOR_RELAY
+            config ZACUS_P7_RELAY_ACTIVE_HIGH
+                bool "Relay active-HIGH (coil energised when GPIO=1)"
+                default y
+                help
+                    Set y for most 5V relay modules with active-high inputs.
+                    Set n for active-low modules (e.g. opto-coupled boards
+                    where the relay fires when the pin is pulled LOW).
+
+            config ZACUS_P7_RELAY_PULSE_MS
+                int "Relay pulse duration (ms, 0 = hold permanently)"
+                default 800
+                range 0 10000
+                help
+                    If > 0 the relay is energised for this many milliseconds
+                    then returned to rest (one-shot electronic latch).
+                    Set 0 to keep the relay energised until p7_coffre_lock()
+                    is called (e.g. when a magnetic solenoid holds the bolt).
+        endif
+
+    endif # ZACUS_P7_COFFRE_ENABLE
+
+    # ─── P5 Morse / telegraph key ───────────────────────────────────────────────
+
+    config ZACUS_P5_MORSE_ENABLE
+        bool "Enable P5 Morse puzzle (telegraph key input)"
+        default n
+        help
+            Drives the P5 code-Morse puzzle: player taps a telegraph key and
+            the firmware decodes dot/dash sequences. When disabled all
+            p5_morse_*() functions are empty stubs so callers compile without
+            #ifdef guards.
+            Enable once the telegraph key (or push button) is wired up.
+
+    if ZACUS_P5_MORSE_ENABLE
+
+        config ZACUS_P5_GPIO
+            int "GPIO pin for the telegraph key / button"
+            default 14
+            range 0 48
+            help
+                Input GPIO. The key connects this pin to GND when pressed.
+                An internal pull-up is enabled by the driver.
+
+        config ZACUS_P5_ACTIVE_LOW
+            bool "Key active-LOW (pressed = GPIO reads 0)"
+            default y
+            help
+                Set y when using a standard normally-open button to GND
+                (internal pull-up enabled). Set n for active-high sensors.
+
+        config ZACUS_P5_DOT_MS
+            int "Maximum dot press duration (ms)"
+            default 200
+            range 10 2000
+            help
+                A key press shorter than or equal to this value is decoded
+                as a dot. Presses above this threshold become dashes.
+
+        config ZACUS_P5_DASH_MS
+            int "Minimum dash press duration (ms)"
+            default 600
+            range 50 5000
+            help
+                A key press must last at least this long to be decoded as a
+                dash. Presses between DOT_MS+1 and DASH_MS-1 are discarded
+                as noise or accidental touches.
+
+        config ZACUS_P5_GAP_MS
+            int "Letter-gap silence duration (ms)"
+            default 800
+            range 100 5000
+            help
+                Silence between key presses longer than this value triggers
+                decoding of the accumulated symbol (letter boundary).
+
+    endif # ZACUS_P5_MORSE_ENABLE
+
+    # ─── P6 NFC / alchemical symbols ────────────────────────────────────────────
+
+    config ZACUS_P6_NFC_ENABLE
+        bool "Enable P6 NFC puzzle (alchemical symbol tags)"
+        default n
+        help
+            Drives the P6 alchemical-symbols puzzle: the player places NFC
+            tags (NTAG213) on a wooden tablet; the firmware reads the UIDs
+            and checks the placement order. When disabled all p6_nfc_*()
+            functions are empty stubs so callers compile without #ifdef guards.
+            Enable once the NFC reader module is wired up.
+
+    if ZACUS_P6_NFC_ENABLE
+
+        choice ZACUS_P6_READER_TYPE
+            prompt "NFC reader IC"
+            default ZACUS_P6_READER_PN532
+            help
+                Select the NFC reader module on the board.
+                PN532  – NXP PN532 (common breakout, SPI or I2C).
+                PN7150 – NXP PN7150 (NCI-based, I2C only).
+
+            config ZACUS_P6_READER_PN532
+                bool "PN532 (NXP)"
+            config ZACUS_P6_READER_PN7150
+                bool "PN7150 (NXP NCI)"
+        endchoice
+
+        choice ZACUS_P6_BUS_TYPE
+            prompt "Communication bus"
+            default ZACUS_P6_BUS_I2C
+            help
+                Select the bus connecting the ESP32 to the NFC reader.
+                I2C – two-wire, uses SDA/SCL. PN7150 only supports I2C.
+                SPI – four-wire, faster. PN532 only.
+
+            config ZACUS_P6_BUS_I2C
+                bool "I2C"
+            config ZACUS_P6_BUS_SPI
+                bool "SPI (PN532 only)"
+        endchoice
+
+        config ZACUS_P6_I2C_SDA
+            int "I2C SDA GPIO"
+            default 8
+            range 0 48
+            depends on ZACUS_P6_BUS_I2C
+            help
+                GPIO number for the I2C data line (SDA).
+                Default 8 is free on the Freenove Media Kit board.
+
+        config ZACUS_P6_I2C_SCL
+            int "I2C SCL GPIO"
+            default 9
+            range 0 48
+            depends on ZACUS_P6_BUS_I2C
+            help
+                GPIO number for the I2C clock line (SCL).
+                Default 9 is free on the Freenove Media Kit board.
+
+        config ZACUS_P6_I2C_ADDR
+            hex "NFC reader I2C address"
+            default 0x24
+            range 0x08 0x77
+            depends on ZACUS_P6_BUS_I2C
+            help
+                7-bit I2C address of the NFC reader.
+                PN532 default: 0x24. PN7150 default: 0x28.
+
+        config ZACUS_P6_IRQ_GPIO
+            int "IRQ / data-ready GPIO (-1 = polling mode)"
+            default -1
+            range -1 48
+            help
+                Optional interrupt / data-ready pin from the NFC reader.
+                Set -1 to use polling mode (5 Hz by default).
+                When wired, the driver waits for a falling edge before
+                reading, which reduces SPI/I2C bus traffic.
+
+    endif # ZACUS_P6_NFC_ENABLE
+
+endmenu

idf_zacus/main/main.c

@@ -55,11 +55,21 @@
 #include "mic_broker.h"
 #include "board_pins_mediakit.h"
 #include "display_ui.h"
+#include "gamebook.h"
+#include "p7_coffre.h"
+#include "p5_morse.h"
+#include "p6_nfc.h"
 
 // Hints engine endpoint (slice 5). Hardcoded for now — slice 7 will move
 // this to NVS so the field operator can repoint the firmware without a flash.
 #define ZACUS_HINTS_BASE_URL  "http://192.168.0.150:8302"
 
+// Voice gateway (tools/zacus-gateway) — the engine POSTs each scene change to
+// {URL}/game/step?scene=SCENE_* so the phone NPCs disguise that scene's hint.
+// Hardcoded for now like the URLs above — moves to NVS in the same follow-up.
+#define ZACUS_GATEWAY_BASE_URL  "http://192.168.0.175:8401"
+#define ZACUS_GATEWAY_TOKEN     "testtoken"
+
 // Slice 7: voice-bridge WebSocket on the MacStudio (Tailscale address).
 // Hardcoded here for the same reason as ZACUS_HINTS_BASE_URL — moves to
 // NVS in a follow-up slice. The bridge endpoint is documented in
@@ -427,6 +437,8 @@ void app_main(void) {
         if (btn_err != ESP_OK) {
             ESP_LOGW(TAG, "buttons_input_init: %s", esp_err_to_name(btn_err));
         }
+        // Gamebook mode owns the 5-way pad when started (POST /game/gamebook).
+        gamebook_init();
     }
 
     bool sta_ok = wifi_bring_up();
@@ -511,6 +523,13 @@ void app_main(void) {
             ESP_LOGI(TAG, "media smoke play -> %s",
                      esp_err_to_name(play_err));
 
+            // Boot into the gamebook library (the 6-story tile picker) when the
+            // SD pack is present. Best-effort: no /sdcard/gamebook/library.json
+            // -> the normal shell stays in charge.
+            if (gamebook_start() == ESP_OK) {
+                ESP_LOGI(TAG, "gamebook library up (boot picker)");
+            }
+
             // Slice 4: bring up the ported npc_engine. Cue table is empty
             // at this stage — wiring the scenario IR-driven cue catalog is
             // a follow-up slice. The engine still boots, accepts ticks
@@ -527,6 +546,9 @@ void app_main(void) {
                 ESP_LOGE(TAG, "npc_engine_init failed: %s",
                          esp_err_to_name(npc_err));
             }
+            // Push each scene change to the voice gateway so the phone NPCs
+            // disguise the CURRENT scene's hint (best-effort, see main.c URL).
+            npc_engine_set_gateway(ZACUS_GATEWAY_BASE_URL, ZACUS_GATEWAY_TOKEN);
 
             // Slice 5: bring up the hints HTTP client (so npc_engine can
             // route hint requests through the real backend) and the voice
@@ -587,6 +609,29 @@ void app_main(void) {
             local_puzzles_init(&s_pstate);
             ESP_LOGI(TAG, "puzzle_state + local_puzzles initialised");
 
+            // P7 coffre actuator: configure GPIO/LEDC at boot so the
+            // first call to p7_coffre_unlock() (on STEP_FINAL_WIN) has
+            // zero warm-up latency.  A no-op stub when
+            // CONFIG_ZACUS_P7_COFFRE_ENABLE=n (default).
+            esp_err_t coffre_err = p7_coffre_init();
+            if (coffre_err != ESP_OK) {
+                ESP_LOGW(TAG, "p7_coffre_init: %s", esp_err_to_name(coffre_err));
+            }
+
+            // P5 Morse puzzle: configure telegraph-key GPIO and start the
+            // decoding task. No-op stub when CONFIG_ZACUS_P5_MORSE_ENABLE=n.
+            esp_err_t morse_err = p5_morse_init();
+            if (morse_err != ESP_OK) {
+                ESP_LOGW(TAG, "p5_morse_init: %s", esp_err_to_name(morse_err));
+            }
+
+            // P6 NFC puzzle: initialise bus and NFC polling task.
+            // No-op stub when CONFIG_ZACUS_P6_NFC_ENABLE=n.
+            esp_err_t nfc_err = p6_nfc_init();
+            if (nfc_err != ESP_OK) {
+                ESP_LOGW(TAG, "p6_nfc_init: %s", esp_err_to_name(nfc_err));
+            }
+
             // Task 7: mic_broker takes ownership of the Media Kit I2S IN
             // pins (3/14/46 per board_pins_mediakit.h) BEFORE
             // voice_pipeline_init — the pipeline's own init call then

plip_voice/CMakeLists.txt

@@ -0,0 +1,8 @@
+cmake_minimum_required(VERSION 3.16)
+
+# scenario_mesh is shared with box3_voice — single copy in lib/ so the
+# ESP-NOW protocol stays in sync with the master and other annexes.
+set(EXTRA_COMPONENT_DIRS ../lib/scenario_mesh)
+
+include($ENV{IDF_PATH}/tools/cmake/project.cmake)
+project(zacus-plip-voice)

plip_voice/main/CMakeLists.txt

@@ -0,0 +1,33 @@
+idf_component_register(
+    SRCS
+        "main.c"
+        "es8388.c"
+        "audio.c"
+        "net.c"
+        "cmd_exec.c"
+        "hook_client.c"
+        "phone.c"
+        "tones.c"
+        "dialer.c"
+        "conversation.c"
+        "dtmf.c"
+        "turn_client.c"
+        "slic.c"
+        "plip_gamebook.c"
+    INCLUDE_DIRS "."
+    PRIV_REQUIRES
+        driver
+        esp_driver_gpio
+        esp_event
+        esp_http_client
+        esp_http_server
+        esp_netif
+        esp_timer
+        esp_wifi
+        json
+        nvs_flash
+        spiffs
+        fatfs
+        sdmmc
+        scenario_mesh
+)

plip_voice/main/Kconfig.projbuild

@@ -0,0 +1,131 @@
+menu "PLIP Voice Configuration"
+
+    config PLIP_WIFI_SSID
+        string "WiFi SSID"
+        default "zacus-net"
+        help
+            WiFi network SSID. Credentials can also be stored in NVS
+            (namespace "wifi", keys "ssid"/"pwd") — NVS takes precedence.
+
+    config PLIP_WIFI_PASSWORD
+        string "WiFi Password"
+        default ""
+        help
+            WiFi password. Leave empty for open networks.
+
+    config PLIP_WIFI_CHANNEL
+        int "WiFi channel hint (0 = auto)"
+        default 11
+        range 0 13
+        help
+            Bias the STA scan to start on this channel. Must match the
+            master ESP32's connected channel for ESP-NOW co-channel
+            operation. Lab network uses channel 11.
+
+    config PLIP_MASTER_URL
+        string "Zacus Master Base URL"
+        default "http://192.168.0.188"
+        help
+            Base URL of the Zacus master ESP32 (Freenove board). The PLIP
+            reports hook transitions to <url>/voice/hook.
+
+    config PLIP_SPEAKER_VOLUME
+        int "Default Speaker Volume (0-100)"
+        default 98
+        range 0 100
+        help
+            Default speaker output volume at boot.
+
+    config PLIP_HOOK_GPIO
+        int "Off-hook GPIO number"
+        default 23
+        help
+            GPIO that signals handset off-hook.
+            SLIC SHK line is GPIO23 (A1S board KEY4, re-assigned to SLIC).
+            Legacy dev kit stub used GPIO4 (BOOT button, active-LOW).
+
+    config PLIP_HOOK_ACTIVE_HIGH
+        bool "Hook GPIO active-HIGH means off-hook"
+        default y
+        help
+            When enabled, a HIGH level on PLIP_HOOK_GPIO means the handset is
+            off-hook (SLIC SHK polarity).  When disabled, LOW means off-hook
+            (original dev-kit pull-up + BOOT button polarity).
+
+    config PLIP_DIAL_PULSE
+        bool "Enable rotary dial pulse decoding on SHK GPIO"
+        default y
+        help
+            When enabled, brief open/close pulses on the hook GPIO (from a
+            rotary dial) are decoded into digits and pushed to the dialer.
+            Each pulse train: ~60-100 ms open + ~40 ms closed; 10 pulses = digit 0.
+            A gap > PLIP_DIAL_PULSE_MAX_GAP_MS terminates the digit.
+
+    config PLIP_DIAL_PULSE_MAX_GAP_MS
+        int "Rotary pulse inter-digit gap (ms)"
+        default 200
+        depends on PLIP_DIAL_PULSE
+        range 100 500
+        help
+            If the hook GPIO stays closed for more than this duration after
+            a pulse train, the train is considered complete and the digit is
+            emitted.  200 ms is standard for French rotary dials.
+
+    config PLIP_DIAL_DTMF
+        bool "Enable DTMF (touch-tone) dialing via Goertzel"
+        default n
+        help
+            When enabled, a background task reads 20 ms microphone frames and
+            runs a Goertzel-based DTMF detector (8 frequencies: 697-1633 Hz).
+            Confirmed digits (≥ 40 ms tone, with twist and dominance guards)
+            are pushed to the dialer just like rotary pulses.
+            The detector is active only between off-hook and the start of the
+            NPC greeting; it is disarmed during voice capture (CONNECTED state).
+            Can be combined with PLIP_DIAL_PULSE: whichever source detects a
+            digit first wins.  Default off — enable for touch-tone handsets.
+
+    config PLIP_AUTO_RING
+        bool "Auto story-ring (rings by itself every 15-30 min)"
+        default n
+        help
+            When enabled, the phone rings on its own at a random 15-30 minute
+            interval; picking up launches a random audio story straight away
+            (no menu). If nobody answers within 1 minute the bell stops and the
+            next ring is rescheduled. Default off — enable for the escape-room
+            ambience where the phone calls the players.
+
+    config PLIP_GATEWAY_URL
+        string "NPC Gateway Base URL"
+        default "http://192.168.0.50:8401"
+        help
+            Base URL of the PLIP voice gateway (zacus-gateway FastAPI), as seen
+            from the PLIP on the local LAN.  Override at build time or via
+            sdkconfig.defaults.  The turn_client appends /v1/voice/turn.
+            Example: http://192.168.0.10:8401 (IP of the Mac running the gateway).
+
+    config PLIP_GATEWAY_TOKEN
+        string "NPC Gateway Bearer Token"
+        default ""
+        help
+            Bearer token sent as "Authorization: Bearer <token>" on every
+            /v1/voice/turn request.  Leave empty to skip the header.
+
+    config PLIP_VOICE_REPLY
+        bool "Enable Stage-3 conversational LISTEN loop (capture -> /v1/voice/reply -> play)"
+        default n
+        help
+            When enabled, after the NPC greeting is played (STATE_CONNECTED),
+            the firmware enters a continuous listen loop:
+              1. Capture mic audio (up to 8 s, VAD-gated) via audio_capture_wav().
+              2. POST the captured WAV as multipart/form-data to
+                 CONFIG_PLIP_GATEWAY_URL/v1/voice/reply (STT + NPC reply via Kyutai).
+              3. Play the NPC response WAV from /spiffs/reply.wav.
+              4. Repeat until the handset is hung up.
+            Requires the gateway (zacus-gateway FastAPI) to be reachable and
+            the /v1/voice/reply endpoint to be operational.
+            Capture buffer (~256 KB for 8 s) is allocated from PSRAM when
+            available; falls back to internal heap with reduced duration (4 s).
+            Leave OFF (default) to keep STATE_CONNECTED as a terminal state
+            (Stage 2 behaviour — greeting only, no further interaction).
+
+endmenu

plip_voice/main/audio.c

@@ -0,0 +1,896 @@
+/*
+ * audio.c — I2S + ES8388 audio driver for the PLIP voice annex (Phase A/D).
+ *
+ * Init order:
+ *   1. es8388_init()       — I2C master + codec register sequence
+ *   2. i2s_new_channel()   — allocate TX (speaker) + RX (mic) channels
+ *   3. i2s_channel_init_std_mode() — configure Philips 16-bit stereo @ 16kHz
+ *   4. i2s_channel_enable()
+ *   5. gpio PA_ENABLE = 1  (already done by es8388_init)
+ *
+ * Threading:
+ *   - audio_play_tone() is blocking — call from a task, not the I2S callback.
+ *   - audio_play_async() posts to a FreeRTOS queue; a worker task drains it.
+ *   - The I2S handle is published via audio_spk_handle() for cmd_exec.
+ */
+
+#include "audio.h"
+#include "board_config.h"
+#include "es8388.h"
+#include "phone.h"
+#include "slic.h"
+
+#include <inttypes.h>
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "driver/gpio.h"
+#include "esp_heap_caps.h"
+#include "esp_log.h"
+#include "esp_spiffs.h"
+#include "esp_vfs_fat.h"
+#include "driver/i2s_std.h"
+#include "driver/sdspi_host.h"
+#include "driver/spi_common.h"
+#include "sdmmc_cmd.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/queue.h"
+#include "freertos/task.h"
+
+#define TAG "audio"
+
+/* ── Audio parameters ────────────────────────────────────────────────────── */
+#define SAMPLE_RATE     PLIP_SAMPLE_RATE     /* 16000 */
+#define BITS            PLIP_BITS_PER_SAMPLE /* 16 */
+#define CHANNELS        PLIP_CHANNELS        /* 2 = stereo (ES8388 I2S) */
+
+/* ── WAV header ──────────────────────────────────────────────────────────── */
+#define WAV_MIN_HEADER  44
+
+typedef struct {
+    uint16_t audio_format;
+    uint16_t channels;
+    uint32_t sample_rate;
+    uint16_t bits_per_sample;
+    uint32_t data_offset;
+    uint32_t data_size;
+} wav_info_t;
+
+/* ── Async play queue ────────────────────────────────────────────────────── */
+typedef enum { CMD_PLAY, CMD_STOP, CMD_RING } audio_cmd_kind_t;
+
+typedef struct {
+    audio_cmd_kind_t kind;
+    char path[160];
+} audio_cmd_t;
+
+static QueueHandle_t      s_queue;
+static i2s_chan_handle_t  s_spk_handle = NULL;
+static i2s_chan_handle_t  s_mic_handle = NULL;
+static volatile bool      s_stop_req   = false;
+static volatile bool      s_playing    = false;  /* true while the worker plays a clip */
+static bool               s_sd_mounted = false;
+static bool               s_spiffs_mounted = false;
+
+/* ── SPIFFS lazy mount ───────────────────────────────────────────────────── */
+
+static void ensure_spiffs_audio(void)
+{
+    if (s_spiffs_mounted) return;
+    esp_vfs_spiffs_conf_t conf = {
+        .base_path              = "/spiffs",
+        .partition_label        = "storage",
+        .max_files              = 8,
+        .format_if_mount_failed = true,  /* format a blank/corrupt partition at boot */
+    };
+    esp_err_t ret = esp_vfs_spiffs_register(&conf);
+    if (ret == ESP_OK || ret == ESP_ERR_INVALID_STATE /* already mounted */) {
+        s_spiffs_mounted = true;
+        ESP_LOGI(TAG, "SPIFFS mounted at /spiffs");
+    } else {
+        ESP_LOGW(TAG, "SPIFFS mount failed: %s", esp_err_to_name(ret));
+    }
+}
+
+/* ── SD card init (best-effort) ──────────────────────────────────────────── */
+
+static void ensure_sd_mounted(void)
+{
+    if (s_sd_mounted) return;
+
+    sdmmc_host_t host = SDSPI_HOST_DEFAULT();
+    host.slot = SPI3_HOST;
+
+    spi_bus_config_t bus_cfg = {
+        .mosi_io_num     = PLIP_SD_MOSI,
+        .miso_io_num     = PLIP_SD_MISO,
+        .sclk_io_num     = PLIP_SD_SCK,
+        .quadwp_io_num   = -1,
+        .quadhd_io_num   = -1,
+        .max_transfer_sz = 4000,
+    };
+    esp_err_t ret = spi_bus_initialize(SPI3_HOST, &bus_cfg, SDSPI_DEFAULT_DMA);
+    if (ret != ESP_OK && ret != ESP_ERR_INVALID_STATE) {
+        ESP_LOGW(TAG, "spi_bus_initialize: %s", esp_err_to_name(ret));
+        return;
+    }
+
+    sdspi_device_config_t slot = SDSPI_DEVICE_CONFIG_DEFAULT();
+    slot.gpio_cs = PLIP_SD_CS;
+    slot.host_id = SPI3_HOST;
+
+    esp_vfs_fat_sdmmc_mount_config_t mount_cfg = {
+        .format_if_mount_failed = false,
+        .max_files = 4,
+        .allocation_unit_size = 16 * 1024,
+    };
+    sdmmc_card_t *card = NULL;
+    ret = esp_vfs_fat_sdspi_mount(PLIP_SD_MOUNT, &host, &slot, &mount_cfg, &card);
+    if (ret == ESP_OK) {
+        s_sd_mounted = true;
+        ESP_LOGI(TAG, "SD card mounted at %s", PLIP_SD_MOUNT);
+        sdmmc_card_print_info(stdout, card);
+    } else {
+        ESP_LOGW(TAG, "SD card mount failed: %s", esp_err_to_name(ret));
+    }
+}
+
+bool audio_ensure_sd(void)
+{
+    ensure_sd_mounted();
+    return s_sd_mounted;
+}
+
+/* ── WAV parser ──────────────────────────────────────────────────────────── */
+
+static esp_err_t parse_wav_header(const uint8_t *buf, size_t len, wav_info_t *out)
+{
+    if (len < WAV_MIN_HEADER) return ESP_ERR_INVALID_SIZE;
+    if (memcmp(buf, "RIFF", 4) != 0 || memcmp(buf + 8, "WAVE", 4) != 0)
+        return ESP_ERR_INVALID_ARG;
+
+    size_t pos = 12;
+    bool got_fmt = false, got_data = false;
+    while (pos + 8 <= len && !(got_fmt && got_data)) {
+        uint32_t chunk_size;
+        memcpy(&chunk_size, buf + pos + 4, 4);
+        if (memcmp(buf + pos, "fmt ", 4) == 0 && chunk_size >= 16) {
+            memcpy(&out->audio_format,   buf + pos + 8,  2);
+            memcpy(&out->channels,       buf + pos + 10, 2);
+            memcpy(&out->sample_rate,    buf + pos + 12, 4);
+            memcpy(&out->bits_per_sample,buf + pos + 22, 2);
+            got_fmt = true;
+        } else if (memcmp(buf + pos, "data", 4) == 0) {
+            out->data_offset = (uint32_t)(pos + 8);
+            out->data_size   = chunk_size;
+            got_data = true;
+        }
+        pos += 8 + chunk_size;
+        if (chunk_size == 0) break;
+    }
+    if (!got_fmt || !got_data) return ESP_ERR_NOT_FOUND;
+    if (out->audio_format != 1) return ESP_ERR_NOT_SUPPORTED; /* not PCM */
+    return ESP_OK;
+}
+
+/* WAV streaming chunk size — keeps heap usage well under 8 KB. */
+#define PLAY_CHUNK_BYTES 4096
+
+/* Play WAV from filesystem (SPIFFS or SD) using streaming I/O.
+ * Only the WAV header (≤512 bytes) is buffered; PCM is piped directly
+ * to I2S in PLAY_CHUNK_BYTES slices, avoiding any large heap allocation.
+ * Paths starting with /spiffs/ are read from SPIFFS; all others use SD. */
+static void play_wav_file(const char *path)
+{
+    bool is_spiffs = (strncmp(path, "/spiffs/", 8) == 0 || strncmp(path, "/spiffs", 7) == 0);
+    if (is_spiffs) {
+        ensure_spiffs_audio();
+        if (!s_spiffs_mounted) {
+            ESP_LOGW(TAG, "SPIFFS not mounted — beep fallback");
+            audio_play_tone(880.0f, 200);
+            return;
+        }
+    } else {
+        ensure_sd_mounted();
+        if (!s_sd_mounted) {
+            ESP_LOGW(TAG, "SD not mounted — beep fallback");
+            audio_play_tone(880.0f, 200);
+            return;
+        }
+    }
+
+    char full[192];
+    if (path[0] == '/') {
+        snprintf(full, sizeof(full), "%s", path);
+    } else {
+        snprintf(full, sizeof(full), "%s/%s", PLIP_SD_MOUNT, path);
+    }
+
+    FILE *f = fopen(full, "rb");
+    if (!f) {
+        ESP_LOGW(TAG, "file not found: %s", full);
+        audio_play_tone(880.0f, 200);
+        return;
+    }
+
+    /* Read enough of the file to parse the WAV header (up to 512 bytes). */
+    uint8_t hdr_buf[512];
+    size_t hdr_read = fread(hdr_buf, 1, sizeof(hdr_buf), f);
+    if (hdr_read < WAV_MIN_HEADER) {
+        ESP_LOGW(TAG, "WAV too short: %zu bytes", hdr_read);
+        fclose(f);
+        audio_play_tone(880.0f, 200);
+        return;
+    }
+
+    wav_info_t wi = {0};
+    if (parse_wav_header(hdr_buf, hdr_read, &wi) != ESP_OK) {
+        ESP_LOGW(TAG, "WAV parse failed: %s", full);
+        fclose(f);
+        audio_play_tone(880.0f, 200);
+        return;
+    }
+    if (wi.bits_per_sample != 16) {
+        ESP_LOGW(TAG, "WAV: %d-bit not supported (need 16-bit)", wi.bits_per_sample);
+        fclose(f);
+        return;
+    }
+
+    ESP_LOGI(TAG, "WAV: %"PRIu32" Hz %d-bit %d ch, %"PRIu32" bytes PCM",
+             wi.sample_rate, wi.bits_per_sample, wi.channels, wi.data_size);
+
+    /* Seek to the PCM data section. */
+    if (fseek(f, (long)wi.data_offset, SEEK_SET) != 0) {
+        ESP_LOGW(TAG, "fseek to data failed");
+        fclose(f);
+        return;
+    }
+
+    /* The I2S TX slot is STEREO @ SAMPLE_RATE. A mono WAV must be expanded to
+     * L+R or it is consumed at 2x rate (the "chipmunk" fast/high-pitch bug).
+     * A WAV whose rate differs from SAMPLE_RATE would also play at the wrong
+     * speed — warn (the gateway TTS always returns 16 kHz, matching). */
+    if (wi.sample_rate != SAMPLE_RATE) {
+        ESP_LOGW(TAG, "WAV rate %"PRIu32" Hz != I2S %d Hz — playback speed will be off",
+                 wi.sample_rate, SAMPLE_RATE);
+    }
+    const bool mono = (wi.channels == 1);
+
+    /* Load the ENTIRE PCM into PSRAM BEFORE playing. Streaming fread() from
+     * SPIFFS *between* I2S writes stalls the DMA → underrun → audible
+     * distortion ("saturation"). Diagnostic confirmed: the stored WAV is clean
+     * (0% clip) and RAM-generated tones play clean, but SPIFFS-streamed WAVs
+     * distorted. Reading it all up-front = zero file I/O during playback. */
+    uint8_t *pcm = heap_caps_malloc(wi.data_size, MALLOC_CAP_SPIRAM);
+    if (!pcm) pcm = malloc(wi.data_size);
+    if (!pcm) {
+        ESP_LOGE(TAG, "play: OOM for %"PRIu32"-byte PCM buffer", wi.data_size);
+        fclose(f);
+        return;
+    }
+    size_t pcm_len = fread(pcm, 1, wi.data_size, f);
+    fclose(f);
+
+    static int16_t s_stereo_chunk[PLAY_CHUNK_BYTES];   /* mono→stereo scratch */
+    size_t off = 0;
+    size_t total_written = 0;
+    const size_t step = mono ? (PLAY_CHUNK_BYTES / 2) : PLAY_CHUNK_BYTES;
+
+    while (!s_stop_req && off < pcm_len) {
+        size_t bytes = (pcm_len - off < step) ? (pcm_len - off) : step;
+        const uint8_t *out = pcm + off;
+        size_t out_len = bytes;
+        if (mono) {
+            /* Duplicate each 16-bit mono sample into L and R. */
+            size_t samples = bytes / 2;
+            const int16_t *src = (const int16_t *)(pcm + off);
+            for (size_t k = 0; k < samples; k++) {
+                s_stereo_chunk[2 * k]     = src[k];
+                s_stereo_chunk[2 * k + 1] = src[k];
+            }
+            out = (const uint8_t *)s_stereo_chunk;
+            out_len = samples * 4;  /* 2 channels × 2 bytes */
+        }
+        size_t w = 0;
+        esp_err_t ret = i2s_channel_write(s_spk_handle, out, out_len, &w, pdMS_TO_TICKS(500));
+        if (ret != ESP_OK) {
+            ESP_LOGW(TAG, "I2S write error: %s", esp_err_to_name(ret));
+            break;
+        }
+        total_written += w;
+        off += bytes;
+    }
+    free(pcm);
+    float dur = (float)total_written / (float)(SAMPLE_RATE * 2 * 2);  /* 16k stereo 16-bit */
+    ESP_LOGI(TAG, "play done: %zu bytes written, %.2fs", total_written, dur);
+}
+
+/* Generate a simple embedded 3-tone cue C5-E5-G5 as proof-of-life. */
+static void play_embedded_cue(void)
+{
+    /* Three 200ms tones: C5 523Hz, E5 659Hz, G5 784Hz */
+    const float freqs[] = {523.0f, 659.0f, 784.0f};
+    for (int i = 0; i < 3 && !s_stop_req; i++) {
+        audio_play_tone(freqs[i], 200);
+        vTaskDelay(pdMS_TO_TICKS(50));
+    }
+}
+
+/* ── Ring tone task ──────────────────────────────────────────────────────── */
+
+static void ring_task(void *arg)
+{
+    (void)arg;
+    /* French ring: 1.5s ON at ~440Hz, 3.5s OFF, repeat until stop */
+    while (!s_stop_req) {
+        /* ON: 1.5s at 440 Hz */
+        const int on_ms = 1500;
+        const int total_samples = SAMPLE_RATE * on_ms / 1000;
+        const float freq = 440.0f;
+        const float amp  = 12000.0f;
+#define RING_CHUNK 128
+        int16_t buf[RING_CHUNK * 2];  /* stereo buffer */
+        int sample_idx = 0;
+        while (!s_stop_req && sample_idx < total_samples) {
+            int chunk = (total_samples - sample_idx < RING_CHUNK)
+                            ? (total_samples - sample_idx) : RING_CHUNK;
+            for (int i = 0; i < chunk; i++) {
+                float t = (float)(sample_idx + i) / (float)SAMPLE_RATE;
+                int16_t v = (int16_t)(amp * sinf(2.0f * (float)M_PI * freq * t));
+                buf[i * 2]     = v; /* L */
+                buf[i * 2 + 1] = v; /* R */
+            }
+            size_t written = 0;
+            i2s_channel_write(s_spk_handle, buf, (size_t)chunk * 2 * sizeof(int16_t),
+                              &written, pdMS_TO_TICKS(500));
+            sample_idx += chunk;
+        }
+#undef RING_CHUNK
+        if (s_stop_req) break;
+        /* OFF: 3.5s silence */
+        const int off_ms = 3500;
+        const int steps = off_ms / 50;
+        for (int i = 0; i < steps && !s_stop_req; i++) {
+            vTaskDelay(pdMS_TO_TICKS(50));
+        }
+    }
+    vTaskDelete(NULL);
+}
+
+/* ── Worker task ─────────────────────────────────────────────────────────── */
+
+static void audio_worker_task(void *arg)
+{
+    (void)arg;
+    ESP_LOGI(TAG, "audio worker ready");
+    audio_cmd_t cmd;
+    for (;;) {
+        if (xQueueReceive(s_queue, &cmd, portMAX_DELAY) != pdTRUE) continue;
+        s_stop_req = false;
+
+        switch (cmd.kind) {
+        case CMD_STOP:
+            s_stop_req = true;
+            ESP_LOGI(TAG, "stop");
+            break;
+        case CMD_RING:
+            /* Ring must sound on-hook (caller hears it before picking up).
+             * Force PA on for the duration of the ring regardless of hook state. */
+            ESP_LOGI(TAG, "ring start — forcing PA on");
+            audio_pa_set(true);
+            xTaskCreate(ring_task, "ring", 4096, NULL, 4, NULL);
+            break;
+        case CMD_PLAY: {
+            /* Gate playback on the REAL hook line (raw SLIC SHK), not phone.c's
+             * debounced flag. The debounced flag lags or misses the pickup while
+             * the bell is ringing, so an incoming-call greeting enqueued right
+             * after pickup would be dropped here ("on-hook") even though the
+             * handset is physically up. The raw SHK reads the pickup cleanly, and
+             * is the single source of truth for hook state. */
+            if (!slic_is_offhook()) {
+                ESP_LOGI(TAG, "play ignored: on-hook (handset down)");
+                break;
+            }
+            s_playing = true;
+            const char *p = cmd.path;
+            if (!p || !*p || strncmp(p, "embedded:", 9) == 0) {
+                ESP_LOGI(TAG, "play: embedded cue");
+                play_embedded_cue();
+            } else {
+                ESP_LOGI(TAG, "play: %s", p);
+                play_wav_file(p);
+            }
+            s_playing = false;
+            break;
+        }
+        default:
+            break;
+        }
+    }
+}
+
+/* ── Public API ──────────────────────────────────────────────────────────── */
+
+void audio_pa_set(bool enable)
+{
+    gpio_set_level(PLIP_PA_ENABLE, enable ? 1 : 0);
+    ESP_LOGI(TAG, "PA %s", enable ? "ON" : "OFF");
+}
+
+bool audio_is_playing(void)
+{
+    return s_playing;
+}
+
+esp_err_t audio_init(void)
+{
+    /* 1. ES8388 I2C init + register sequence. */
+    esp_err_t ret = es8388_init();
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "es8388_init failed: %s", esp_err_to_name(ret));
+        return ret;
+    }
+
+    /* Apply the configured output volume. es8388_init() leaves OUT2 at 0 dB
+     * (max) — too loud for a handset earpiece — so set it explicitly here. */
+    es8388_set_volume(CONFIG_PLIP_SPEAKER_VOLUME);
+
+    /* 2. Create I2S channels (TX = speaker, RX = mic — full-duplex pair). */
+    i2s_chan_config_t chan_cfg = I2S_CHANNEL_DEFAULT_CONFIG(PLIP_I2S_NUM,
+                                                             I2S_ROLE_MASTER);
+    chan_cfg.auto_clear = true;
+    ret = i2s_new_channel(&chan_cfg, &s_spk_handle, &s_mic_handle);
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "i2s_new_channel: %s", esp_err_to_name(ret));
+        return ret;
+    }
+
+    /* 3a+3b. Full-duplex I2S config: SAME gpio_cfg for TX and RX.
+     * IDF5 i2s_std constitutes full-duplex ONLY when TX and RX configs are
+     * identical (memcmp). If they differ, on ESP32 HW v1 it tries to move
+     * RX to I2S_NUM_1, breaking clock routing and yielding permanent zeros.
+     * Fix: set both dout=PLIP_I2S_DOUT and din=PLIP_I2S_DIN in the same
+     * config and apply it to both handles. The IDF GPIO driver handles
+     * direction internally (dout→output, din→input). */
+    i2s_std_config_t std_cfg = {
+        .clk_cfg  = I2S_STD_CLK_DEFAULT_CONFIG(SAMPLE_RATE),
+        .slot_cfg = I2S_STD_PHILIPS_SLOT_DEFAULT_CONFIG(
+                        I2S_DATA_BIT_WIDTH_16BIT, I2S_SLOT_MODE_STEREO),
+        .gpio_cfg = {
+            .mclk = PLIP_I2S_MCLK,
+            .bclk = PLIP_I2S_BCLK,
+            .ws   = PLIP_I2S_WS,
+            .dout = PLIP_I2S_DOUT,
+            .din  = PLIP_I2S_DIN,
+            .invert_flags = {
+                .mclk_inv = false,
+                .bclk_inv = false,
+                .ws_inv   = false,
+            },
+        },
+    };
+    ret = i2s_channel_init_std_mode(s_spk_handle, &std_cfg);
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "i2s_channel_init_std_mode (TX): %s", esp_err_to_name(ret));
+        return ret;
+    }
+    ret = i2s_channel_init_std_mode(s_mic_handle, &std_cfg);
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "i2s_channel_init_std_mode (RX): %s", esp_err_to_name(ret));
+        return ret;
+    }
+
+    /* Enable both TX and RX at boot for full-duplex operation.
+     * Both must be enabled for the I2S hardware to route ASDOUT→DIN correctly.
+     * RX data will be read only during capture; TX stays active for clock/DAC. */
+    ret = i2s_channel_enable(s_spk_handle);
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "i2s_channel_enable (TX): %s", esp_err_to_name(ret));
+        return ret;
+    }
+    ret = i2s_channel_enable(s_mic_handle);
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "i2s_channel_enable (RX): %s", esp_err_to_name(ret));
+        /* Non-fatal: TX still works for playback; capture will fail gracefully. */
+        ESP_LOGW(TAG, "RX enable failed — capture disabled, playback unaffected");
+        s_mic_handle = NULL;
+    }
+
+    /* 3c. Quick RX sanity: read multiple frames to check DMA and signal quality.
+     * Do this before the 440Hz boot tone to see baseline ADC output. */
+    if (s_mic_handle) {
+        int16_t test_buf[320 * 2];
+        size_t test_read = 0;
+        /* Drain any startup garbage (first few frames may be codec settling). */
+        for (int flush = 0; flush < 5; flush++) {
+            i2s_channel_read(s_mic_handle, test_buf, sizeof(test_buf), &test_read, pdMS_TO_TICKS(50));
+        }
+        /* Read a real frame and report. */
+        esp_err_t rx_ret = i2s_channel_read(s_mic_handle, test_buf,
+                                             sizeof(test_buf), &test_read,
+                                             pdMS_TO_TICKS(200));
+        int32_t max_val = 0, rms_sq = 0;
+        int n = (int)(test_read / 2);
+        for (int i = 0; i < n; i++) {
+            int32_t v = test_buf[i] < 0 ? -test_buf[i] : test_buf[i];
+            if (v > max_val) max_val = v;
+            rms_sq += (int32_t)test_buf[i] * test_buf[i];
+        }
+        rms_sq /= (n > 0 ? n : 1);
+        ESP_LOGI(TAG, "RX sanity: ret=%s read=%zu max=%"PRId32" rms²=%"PRId32" s[0..3]=(%d,%d,%d,%d)",
+                 esp_err_to_name(rx_ret), test_read, max_val, rms_sq,
+                 n>0?test_buf[0]:0, n>0?test_buf[1]:0, n>1?test_buf[2]:0, n>1?test_buf[3]:0);
+    }
+
+    /* 4. Create async command queue + worker. */
+    s_queue = xQueueCreate(8, sizeof(audio_cmd_t));
+    if (!s_queue) return ESP_ERR_NO_MEM;
+
+    BaseType_t ok = xTaskCreatePinnedToCore(audio_worker_task, "audio_work",
+                                             8192, NULL, 5, NULL, 0);
+    if (ok != pdPASS) return ESP_ERR_NO_MEM;
+
+    /* Mount SPIFFS at init (not lazily) so turn_client can WRITE the NPC WAV to
+     * /spiffs before any playback has triggered the lazy mount. */
+    ensure_spiffs_audio();
+
+    ESP_LOGI(TAG, "audio init OK (I2S TX ready, RX handle allocated, ES8388 live)");
+    return ESP_OK;
+}
+
+/* ── Mic capture ─────────────────────────────────────────────────────────── */
+
+/*
+ * audio_capture_wav() — capture microphone audio and return a WAV buffer.
+ *
+ * Strategy: half-duplex. Disable TX, enable RX, read until VAD silence timeout
+ * or hard cap, then disable RX, re-enable TX.
+ *
+ * ES8388 delivers stereo I2S frames (L+R int16) even though the mic is mono.
+ * We average L+R into a single mono sample for the output WAV.
+ *
+ * Returns number of bytes written to out (header + PCM), or -1 on error.
+ */
+int audio_capture_wav(uint8_t *out, size_t out_max, int max_ms, int silence_ms)
+{
+    if (!s_mic_handle || !s_spk_handle) {
+        ESP_LOGE(TAG, "capture: I2S handles not ready");
+        return -1;
+    }
+    if (!out || out_max < 44 + 320) {
+        ESP_LOGE(TAG, "capture: output buffer too small");
+        return -1;
+    }
+
+    /* WAV parameters for output: 16 kHz, mono, S16-LE. */
+    const uint32_t cap_rate   = SAMPLE_RATE;   /* 16000 */
+    const uint16_t cap_ch     = 1;             /* mono output */
+    const uint16_t cap_bits   = 16;
+    const size_t   hdr_bytes  = 44;
+
+    /* Per-frame: 20 ms at 16kHz = 320 mono samples = 640 bytes PCM out.
+     * ES8388 stereo frame = 640 bytes input (320 samples × 2 ch × 2 bytes). */
+    const int      frame_samples = 320;        /* mono samples per frame */
+    const size_t   frame_in_bytes  = (size_t)frame_samples * 2 * sizeof(int16_t); /* stereo */
+    const size_t   frame_out_bytes = (size_t)frame_samples * sizeof(int16_t);     /* mono */
+
+    /* Allocate scratch buffer in SPIRAM if available, else internal heap. */
+    int16_t *rx_buf = (int16_t *)heap_caps_malloc(frame_in_bytes,
+                                                    MALLOC_CAP_SPIRAM);
+    if (!rx_buf) {
+        rx_buf = (int16_t *)malloc(frame_in_bytes);
+    }
+    if (!rx_buf) {
+        ESP_LOGE(TAG, "capture: OOM for rx_buf");
+        return -1;
+    }
+
+    /* RX is already enabled at boot (full-duplex). Calling enable again returns
+     * ESP_ERR_INVALID_STATE — that's fine, it just means RX is already running.
+     * Only a genuinely different error is fatal. */
+    esp_err_t ret = i2s_channel_enable(s_mic_handle);
+    if (ret != ESP_OK && ret != ESP_ERR_INVALID_STATE) {
+        ESP_LOGE(TAG, "capture: i2s_channel_enable(RX): %s", esp_err_to_name(ret));
+        free(rx_buf);
+        return -1;
+    }
+
+    /* Reserve space for WAV header at start of out buffer. */
+    uint8_t  *pcm_out    = out + hdr_bytes;
+    size_t    pcm_max    = out_max - hdr_bytes;
+    size_t    pcm_written = 0;
+
+    const int max_frames     = (max_ms     / 20);
+    const int silence_frames = (silence_ms / 20);
+
+    /* Calibrated to the QUIET handset voice (DC-blocked: body 0.16-0.8 % FS RMS,
+     * loud syllables ~2-4 %, noise floor ~0.3 %). The old onset (1.4 %) NEVER
+     * triggered → "no sustained voice"; the old silence (0.6 %) sat ABOVE the
+     * body → cut mid-sentence. Now: onset catches a real syllable just above the
+     * floor (3-frame confirm guards clicks); silence sits just above the floor so
+     * the turn ends only on a true pause, holding the whole sentence. */
+    const int32_t  vad_onset_rms_sq   = 230 * 230;   /* ~0.70% FS² — quiet-voice onset */
+    const int32_t  vad_silence_rms_sq = 110 * 110;   /* ~0.34% FS² — just above noise floor */
+
+    int  silent_frames = 0;
+    int  total_frames  = 0;
+
+    /* One-pole DC blocker (high-pass ~80 Hz) applied to the captured mono.
+     * The SLIC/handset path carries a huge DC/sub-audio offset (~80 % FS,
+     * measured). Left in, it (a) swamps the faint voice and (b) keeps the VAD
+     * RMS permanently above the silence threshold, so the capture never ends
+     * and always runs the full max_ms — and the gateway then sees only a DC
+     * transient, transcribing empty. Removing it lets the VAD track the real
+     * speech envelope (onset/silence work) and hands the gateway a clean,
+     * voice-dominated signal. Telephone band (300-3400 Hz) is untouched.
+     * y[n] = x[n] - x[n-1] + R*y[n-1]; R=0.97 → cutoff ~80 Hz at 16 kHz. */
+    float       dc_x1 = 0.0f, dc_y1 = 0.0f;
+    const float dc_R  = 0.97f;
+
+    ESP_LOGI(TAG, "capture: RX enabled, max_ms=%d silence_ms=%d", max_ms, silence_ms);
+
+    /* ── Phase A: wait for SUSTAINED voice before committing to a capture ─────
+     * Only start recording when the caller actually speaks. We require the
+     * DC-blocked RMS to stay above the onset threshold for ONSET_CONFIRM
+     * consecutive frames (~60 ms) — a single loud frame is rejected as a
+     * transient (e.g. the click when the PA mutes just before capture). Audio
+     * is discarded during this phase. If no sustained voice arrives within
+     * max_ms we return an empty WAV so the caller posts nothing (the NPC must
+     * not "reply" to silence). The listen loop simply calls us again. */
+    const int ONSET_CONFIRM = 3;
+    int  onset_run  = 0;
+    bool got_onset  = false;
+    for (int f = 0; f < max_frames; f++) {
+        size_t bytes_read = 0;
+        ret = i2s_channel_read(s_mic_handle, rx_buf, frame_in_bytes,
+                               &bytes_read, pdMS_TO_TICKS(100));
+        if (ret != ESP_OK || bytes_read == 0) continue;
+        int64_t rms_sq = 0;
+        int     n_stereo = (int)(bytes_read / (2 * sizeof(int16_t)));
+        for (int i = 0; i < n_stereo; i++) {
+            float xf = (float)((rx_buf[i * 2] + rx_buf[i * 2 + 1]) / 2);
+            float yf = xf - dc_x1 + dc_R * dc_y1;     /* DC-blocking high-pass */
+            dc_x1 = xf; dc_y1 = yf;
+            int32_t s = (int32_t)yf;
+            rms_sq += (int64_t)s * s;
+        }
+        rms_sq /= (n_stereo > 0 ? n_stereo : 1);
+        if (rms_sq >= vad_onset_rms_sq) {
+            if (++onset_run >= ONSET_CONFIRM) {
+                got_onset = true;
+                ESP_LOGI(TAG, "capture: sustained voice onset at frame %d (rms²=%"PRId64")",
+                         f, rms_sq);
+                break;
+            }
+        } else {
+            onset_run = 0;
+        }
+    }
+
+    if (got_onset) {
+        /* ── Phase B: record the utterance until silence_ms of silence ──────── */
+        for (int f = 0; f < max_frames && pcm_written + frame_out_bytes <= pcm_max; f++) {
+            size_t bytes_read = 0;
+            ret = i2s_channel_read(s_mic_handle, rx_buf, frame_in_bytes,
+                                   &bytes_read, pdMS_TO_TICKS(100));
+            if (ret != ESP_OK || bytes_read == 0) {
+                ESP_LOGW(TAG, "capture: read error f=%d: %s", f, esp_err_to_name(ret));
+                continue;
+            }
+
+            /* Downmix stereo → mono (DC-blocked), compute RMS². */
+            int16_t *out_frame = (int16_t *)(pcm_out + pcm_written);
+            int64_t  rms_sq    = 0;
+            int      n_stereo  = (int)(bytes_read / (2 * sizeof(int16_t)));
+            for (int i = 0; i < n_stereo; i++) {
+                float   xf = (float)((rx_buf[i * 2] + rx_buf[i * 2 + 1]) / 2);
+                float   yf = xf - dc_x1 + dc_R * dc_y1;
+                dc_x1 = xf; dc_y1 = yf;
+                if (yf > 32767.0f)       yf = 32767.0f;
+                else if (yf < -32768.0f) yf = -32768.0f;
+                int16_t mono = (int16_t)yf;
+                out_frame[i] = mono;
+                rms_sq += (int64_t)mono * mono;
+            }
+            rms_sq /= (n_stereo > 0 ? n_stereo : 1);
+
+            pcm_written += (size_t)n_stereo * sizeof(int16_t);
+            total_frames = f + 1;
+
+            if (rms_sq < vad_silence_rms_sq) {
+                if (++silent_frames >= silence_frames) {
+                    ESP_LOGI(TAG, "capture: VAD end (silence %d frames at f=%d)",
+                             silent_frames, f);
+                    break;
+                }
+            } else {
+                silent_frames = 0;
+            }
+        }
+    } else {
+        ESP_LOGI(TAG, "capture: no sustained voice in %d ms — nothing to send", max_ms);
+        /* pcm_written stays 0 → empty WAV → caller skips posting. */
+    }
+
+    /* Leave RX enabled (full-duplex, as at boot). Disabling it here would make
+     * the next capture's enable a no-op INVALID_STATE AND break other RX users
+     * (streaming capture / DTMF) that assume RX stays running. */
+    free(rx_buf);
+
+    if (pcm_written == 0) {
+        ESP_LOGW(TAG, "capture: no audio captured (silence or timeout)");
+        /* Return a valid but minimal WAV with 0 PCM bytes so caller can inspect. */
+    }
+
+    uint32_t data_size     = (uint32_t)pcm_written;
+    uint32_t byte_rate     = cap_rate * cap_ch * (cap_bits / 8);
+    uint16_t block_align   = (uint16_t)(cap_ch * (cap_bits / 8));
+    uint32_t riff_size     = 36 + data_size;
+
+    /* Write WAV header (44 bytes, little-endian). */
+    memcpy(out +  0, "RIFF", 4);
+    memcpy(out +  4, &riff_size, 4);
+    memcpy(out +  8, "WAVE", 4);
+    memcpy(out + 12, "fmt ", 4);
+    uint32_t fmt_size = 16;
+    uint16_t fmt_pcm  = 1;
+    memcpy(out + 16, &fmt_size,    4);
+    memcpy(out + 20, &fmt_pcm,     2);
+    memcpy(out + 22, &cap_ch,      2);
+    memcpy(out + 24, &cap_rate,    4);
+    memcpy(out + 28, &byte_rate,   4);
+    memcpy(out + 32, &block_align, 2);
+    memcpy(out + 34, &cap_bits,    2);
+    memcpy(out + 36, "data", 4);
+    memcpy(out + 40, &data_size,   4);
+
+    int total_bytes = (int)(hdr_bytes + pcm_written);
+    float duration_s = (float)pcm_written / (float)(cap_rate * (cap_bits / 8) * cap_ch);
+    ESP_LOGI(TAG, "capture done: %d frames, %.2fs, %zu PCM bytes, %d total bytes",
+             total_frames, duration_s, pcm_written, total_bytes);
+    return total_bytes;
+}
+
+/* ── Streaming capture API ───────────────────────────────────────────────── */
+
+/* Scratch buffer reused across frames (lives for the duration of a capture). */
+static int16_t *s_rx_scratch = NULL;  /* stereo frame: 320*2 int16_t = 1280 bytes */
+static int      s_frame_debug_count = 0;
+#define CAP_FRAME_SAMPLES 320         /* 20 ms @ 16kHz */
+#define CAP_FRAME_IN_BYTES (CAP_FRAME_SAMPLES * 2 * sizeof(int16_t))
+
+int audio_capture_begin(int max_ms, int silence_ms)
+{
+    (void)max_ms; (void)silence_ms;  /* caller drives VAD/timeout */
+    if (!s_mic_handle) {
+        ESP_LOGE(TAG, "capture_begin: RX handle not available");
+        return -1;
+    }
+    s_rx_scratch = (int16_t *)heap_caps_malloc(CAP_FRAME_IN_BYTES, MALLOC_CAP_SPIRAM);
+    if (!s_rx_scratch) s_rx_scratch = (int16_t *)malloc(CAP_FRAME_IN_BYTES);
+    if (!s_rx_scratch) {
+        ESP_LOGE(TAG, "capture_begin: OOM for scratch buf");
+        return -1;
+    }
+    s_frame_debug_count = 0;
+
+    /* Full-duplex: keep TX running (provides MCLK/BCLK/WS clocks to codec).
+     * RX is already enabled at boot. */
+    ESP_LOGI(TAG, "capture_begin: ready (full-duplex, TX running)");
+    return 0;
+}
+
+/*
+ * Read one 20 ms frame from the mic, downmix stereo→mono.
+ * mono_out must hold n_samples (320) int16_t values.
+ * Sets *rms_sq_out to mean-squared energy of the frame.
+ * Returns number of mono samples written, 0 on read timeout, -1 on error.
+ */
+int audio_capture_read_frame(int16_t *mono_out, int n_samples, int64_t *rms_sq_out)
+{
+    if (!s_rx_scratch) return -1;
+    size_t bytes_read = 0;
+    esp_err_t ret = i2s_channel_read(s_mic_handle, s_rx_scratch,
+                                      CAP_FRAME_IN_BYTES, &bytes_read,
+                                      pdMS_TO_TICKS(100));
+    if (ret != ESP_OK || bytes_read == 0) return 0;
+
+    int n = (int)(bytes_read / (2 * sizeof(int16_t)));
+    if (n > n_samples) n = n_samples;
+
+    /* Debug: log first 3 frames raw stereo values to diagnose zero data. */
+    if (s_frame_debug_count < 3) {
+        int16_t r0 = (n > 0) ? s_rx_scratch[0] : 0;
+        int16_t r1 = (n > 0) ? s_rx_scratch[1] : 0;
+        int16_t r2 = (n > 1) ? s_rx_scratch[2] : 0;
+        int16_t r3 = (n > 1) ? s_rx_scratch[3] : 0;
+        ESP_LOGI(TAG, "RX frame %d: bytes_read=%zu n=%d raw[0..3]=(%d,%d,%d,%d)",
+                 s_frame_debug_count, bytes_read, n, r0, r1, r2, r3);
+        s_frame_debug_count++;
+    }
+
+    int64_t rms_sq = 0;
+    for (int i = 0; i < n; i++) {
+        int32_t l = s_rx_scratch[i * 2];
+        int32_t r = s_rx_scratch[i * 2 + 1];
+        int16_t mono = (int16_t)((l + r) / 2);
+        mono_out[i] = mono;
+        rms_sq += (int64_t)mono * mono;
+    }
+    if (rms_sq_out) *rms_sq_out = (n > 0) ? rms_sq / n : 0;
+    return n;
+}
+
+void audio_capture_end(void)
+{
+    if (s_rx_scratch) { free(s_rx_scratch); s_rx_scratch = NULL; }
+    ESP_LOGI(TAG, "capture_end: done");
+}
+
+i2s_chan_handle_t audio_spk_handle(void)
+{
+    return s_spk_handle;
+}
+
+void audio_play_tone(float frequency_hz, int duration_ms)
+{
+    if (!s_spk_handle || duration_ms <= 0) return;
+    const int total_samples = SAMPLE_RATE * duration_ms / 1000;
+    const float amp = 12000.0f;
+    /* Stereo: each sample = L + R = 2 int16_t values. */
+    /* Process 128 mono samples per iteration = 256 stereo int16_t = 512 bytes. */
+#define TONE_CHUNK 128
+    int16_t buf[TONE_CHUNK * 2];   /* stereo buffer: 2 samples per frame */
+    size_t written = 0;
+    int idx = 0;
+    while (idx < total_samples) {
+        int chunk = (total_samples - idx < TONE_CHUNK)
+                        ? (total_samples - idx) : TONE_CHUNK;
+        for (int i = 0; i < chunk; i++) {
+            float t = (float)(idx + i) / (float)SAMPLE_RATE;
+            int16_t v = (int16_t)(amp * sinf(2.0f * (float)M_PI * frequency_hz * t));
+            buf[i * 2]     = v;   /* L */
+            buf[i * 2 + 1] = v;  /* R */
+        }
+        i2s_channel_write(s_spk_handle, buf, (size_t)chunk * 2 * sizeof(int16_t),
+                          &written, pdMS_TO_TICKS(500));
+        idx += chunk;
+    }
+#undef TONE_CHUNK
+}
+
+esp_err_t audio_play_async(const char *path)
+{
+    if (!s_queue) return ESP_ERR_INVALID_STATE;
+    audio_cmd_t cmd = { .kind = CMD_PLAY };
+    if (path && *path) {
+        strncpy(cmd.path, path, sizeof(cmd.path) - 1);
+    }
+    if (xQueueSend(s_queue, &cmd, 0) != pdTRUE) {
+        ESP_LOGW(TAG, "play queue full");
+        return ESP_FAIL;
+    }
+    return ESP_OK;
+}
+
+esp_err_t audio_stop(void)
+{
+    s_stop_req = true;
+    if (!s_queue) return ESP_ERR_INVALID_STATE;
+    audio_cmd_t cmd = { .kind = CMD_STOP };
+    xQueueSend(s_queue, &cmd, 0);
+    return ESP_OK;
+}
+
+esp_err_t audio_ring_start(void)
+{
+    s_stop_req = false;
+    if (!s_queue) return ESP_ERR_INVALID_STATE;
+    audio_cmd_t cmd = { .kind = CMD_RING };
+    if (xQueueSend(s_queue, &cmd, 0) != pdTRUE) {
+        return ESP_FAIL;
+    }
+    return ESP_OK;
+}

plip_voice/main/audio.h

@@ -0,0 +1,99 @@
+#pragma once
+/*
+ * audio.h — I2S + ES8388 audio interface for the PLIP voice annex.
+ *
+ * Provides:
+ *  - es8388 + I2S initialisation (called once from app_main)
+ *  - Blocking 440 Hz tone (Phase A proof)
+ *  - WAV file playback from SD or embedded asset
+ *  - Ring tone (400/450 Hz cadence) for Phase D
+ *  - Async playback command queue (used by cmd_exec / ESP-NOW)
+ */
+
+#include <stdint.h>
+#include <stddef.h>
+#include "esp_err.h"
+#include "driver/i2s_std.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Initialise ES8388 and I2S channels.  Must be called before any other
+ * audio_* function.  Returns ESP_OK on success. */
+esp_err_t audio_init(void);
+
+/* Return the speaker I2S TX channel handle (used by cmd_exec for WAV play). */
+i2s_chan_handle_t audio_spk_handle(void);
+
+/* Play a pure sine tone for duration_ms (blocking). */
+void audio_play_tone(float frequency_hz, int duration_ms);
+
+/* Enqueue an async play command.  path may be:
+ *   - "/sdcard/<file>.wav"  — read from SD
+ *   - "embedded://"        — built-in C5-E5-G5 cue
+ *   - ""  or NULL          — same as embedded://
+ * Returns ESP_OK if the command was enqueued (non-blocking from any task). */
+esp_err_t audio_play_async(const char *path);
+
+/* Stop current playback immediately. */
+esp_err_t audio_stop(void);
+
+/* Mount the microSD card at /sdcard on demand (best-effort, idempotent) and
+ * report whether it is mounted. Unlike playback, this does NOT depend on the
+ * hook state — it lets the REST layer stage a voice pack onto the SD while the
+ * handset is down. Returns true if /sdcard is usable. */
+bool audio_ensure_sd(void);
+
+/* True while the audio worker is playing a clip (tone/WAV). The conversation
+ * LISTEN loop polls this to stay half-duplex: never capture while playing
+ * (avoids the earpiece→mic feedback that saturated the line). */
+bool audio_is_playing(void);
+
+/* Start ring tone cadence (ON 1s / OFF 2s) at ~440 Hz.  Continues until
+ * audio_stop() is called.  Non-blocking — spawns an internal task. */
+esp_err_t audio_ring_start(void);
+
+/* Enable or disable the power amplifier (GPIO21 PA_ENABLE).
+ * Call audio_pa_set(true) on off-hook, audio_pa_set(false) on on-hook.
+ * ring_start forces PA on internally; ring_stop calls audio_pa_set(false)
+ * only if the handset is on-hook. */
+void audio_pa_set(bool enable);
+
+/*
+ * Capture microphone audio and encode as WAV (16 kHz, mono, S16-LE).
+ *
+ *   out       : caller-provided buffer (must be >= 44 + PCM bytes)
+ *   out_max   : size of out in bytes
+ *   max_ms    : hard cap on capture duration in milliseconds
+ *   silence_ms: milliseconds of silence after voice onset to trigger stop
+ *
+ * Returns total bytes written (44-byte WAV header + PCM), or -1 on error.
+ * Uses half-duplex: disables TX speaker during capture and re-enables it
+ * on return.  Safe to call from any task; not reentrant.
+ */
+int audio_capture_wav(uint8_t *out, size_t out_max, int max_ms, int silence_ms);
+
+/*
+ * Streaming capture API — avoids a large output buffer by letting the caller
+ * write chunks incrementally (e.g. directly into a TCP socket).
+ *
+ * Usage:
+ *   audio_capture_begin(max_ms, silence_ms)   — enable RX, returns 0 on OK
+ *   audio_capture_read_frame(mono_out, n_samples, rms_sq_out)
+ *                                             — read one 20 ms frame (mono S16)
+ *                                               fills n_samples mono samples,
+ *                                               sets *rms_sq_out (energy metric)
+ *                                               returns actual samples read, 0=timeout, -1=error
+ *   audio_capture_end()                       — disable RX, re-enable TX
+ *
+ * n_samples must be 320 (= SAMPLE_RATE/1000*20, one 20 ms frame).
+ * The caller is responsible for VAD and stop logic.
+ */
+int  audio_capture_begin(int max_ms, int silence_ms);
+int  audio_capture_read_frame(int16_t *mono_out, int n_samples, int64_t *rms_sq_out);
+void audio_capture_end(void);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/main/board_config.h

@@ -0,0 +1,49 @@
+#pragma once
+
+/*
+ * AI-Thinker ESP32-A1S Audio Kit V2.2 — Hardware Pin Configuration
+ * ES8388 codec (I2C control + I2S data), SD card (SPI), power amp.
+ * DIP switch: 1=OFF, 2=ON, 3=ON, 4=OFF, 5=OFF  (SD active, KEY2 busy)
+ */
+
+/* ---------- I2C Bus (ES8388 control) ---------- */
+#define PLIP_I2C_PORT       I2C_NUM_0
+#define PLIP_I2C_SCL        32
+#define PLIP_I2C_SDA        33
+#define PLIP_I2C_FREQ_HZ    400000
+
+/* ---------- ES8388 Audio Codec (I2C address) ---------- */
+#define PLIP_ES8388_ADDR    0x10   /* 7-bit: 0x10 with ADDR pin low */
+
+/* ---------- I2S Audio Data ---------- */
+#define PLIP_I2S_NUM        I2S_NUM_0
+#define PLIP_I2S_MCLK       0      /* GPIO0 — must be 0/1/3 on original ESP32 */
+#define PLIP_I2S_BCLK       27
+#define PLIP_I2S_WS         25     /* LRCK */
+#define PLIP_I2S_DOUT       26     /* DAC → speaker */
+#define PLIP_I2S_DIN        35     /* ADC ← mic (input-only, ES8388 ASDOUT) */
+
+/* ---------- Power Amplifier ---------- */
+#define PLIP_PA_ENABLE      21     /* Active HIGH, drives NS4150 amp */
+
+/* ---------- SD Card (SPI / HSPI) ---------- */
+#define PLIP_SD_CS          13
+#define PLIP_SD_MOSI        15
+#define PLIP_SD_MISO        2
+#define PLIP_SD_SCK         14
+#define PLIP_SD_MOUNT       "/sdcard"
+
+/* ---------- Audio Parameters ---------- */
+#define PLIP_SAMPLE_RATE    16000
+#define PLIP_BITS_PER_SAMPLE 16
+#define PLIP_CHANNELS       2      /* ES8388 I2S always stereo; output averaged */
+
+/* ---------- Off-hook GPIO (dev kit uses BOOT/KEY1 GPIO4 as stand-in) ---------- */
+/* Actual value comes from CONFIG_PLIP_HOOK_GPIO (Kconfig) */
+
+/* ---------- SLIC K50835F / AG1171-class front-end (A1S board wiring) ---------- */
+/* KEY3=GPIO19, KEY4=GPIO23, KEY5=GPIO18, KEY6=GPIO5 share these pins — reassigned to SLIC */
+#define PLIP_SLIC_RM    18   /* Ring Mode output — HIGH = ring burst active */
+#define PLIP_SLIC_FR    5    /* Forward/Reverse output — toggled at 25 Hz for bell */
+#define PLIP_SLIC_SHK   23   /* Switch Hook input  — active-LOW (K50835F open-collector) */
+#define PLIP_SLIC_PD    19   /* Power Down (open-drain) — HIGH = SLIC active */

plip_voice/main/cmd_exec.c

@@ -0,0 +1,87 @@
+/*
+ * cmd_exec.c — PLIP voice CMD executor (D5 contract, adapted from box3_voice).
+ *
+ * The PLIP has no display; screen/evt/led ops are logged and ignored.
+ * play → audio_play_async(path).
+ */
+
+#include "cmd_exec.h"
+#include "audio.h"
+
+#include <string.h>
+
+#include "cJSON.h"
+#include "esp_log.h"
+
+#define TAG "cmd_exec"
+
+static esp_err_t exec_play(const cJSON *a)
+{
+    const cJSON *path_j = a ? cJSON_GetObjectItemCaseSensitive(a, "p") : NULL;
+    const char *path = cJSON_IsString(path_j) ? path_j->valuestring : "embedded://";
+
+    ESP_LOGI(TAG, "CMD op=play path=\"%s\"", path);
+    return audio_play_async(path);
+}
+
+static esp_err_t exec_screen(const cJSON *a)
+{
+    const cJSON *t_j = a ? cJSON_GetObjectItemCaseSensitive(a, "t") : NULL;
+    ESP_LOGI(TAG, "CMD op=screen t=\"%s\" (no display on PLIP — ignored)",
+             cJSON_IsString(t_j) ? t_j->valuestring : "");
+    return ESP_OK;
+}
+
+static esp_err_t exec_evt(const cJSON *a)
+{
+    const cJSON *n_j = a ? cJSON_GetObjectItemCaseSensitive(a, "n") : NULL;
+    ESP_LOGI(TAG, "CMD op=evt n=\"%s\"",
+             cJSON_IsString(n_j) ? n_j->valuestring : "<none>");
+    return ESP_OK;
+}
+
+static esp_err_t exec_led(const cJSON *a)
+{
+    const cJSON *p_j = a ? cJSON_GetObjectItemCaseSensitive(a, "p") : NULL;
+    ESP_LOGI(TAG, "CMD op=led pattern=\"%s\" (no LED — ignored)",
+             cJSON_IsString(p_j) ? p_j->valuestring : "<none>");
+    return ESP_OK;
+}
+
+esp_err_t cmd_exec_handle(const char *data, size_t len)
+{
+    if (!data || len == 0) return ESP_ERR_INVALID_ARG;
+
+    cJSON *root = cJSON_ParseWithLength(data, len);
+    if (!root) {
+        ESP_LOGW(TAG, "CMD parse error: not valid JSON (%.40s)", data);
+        return ESP_ERR_INVALID_ARG;
+    }
+
+    const cJSON *op_j = cJSON_GetObjectItemCaseSensitive(root, "op");
+    if (!cJSON_IsString(op_j) || !op_j->valuestring || !op_j->valuestring[0]) {
+        ESP_LOGW(TAG, "CMD missing/invalid 'op' field");
+        cJSON_Delete(root);
+        return ESP_ERR_INVALID_ARG;
+    }
+
+    const char  *op = op_j->valuestring;
+    const cJSON *a  = cJSON_GetObjectItemCaseSensitive(root, "a");
+
+    esp_err_t err;
+    if (strcmp(op, "play") == 0) {
+        err = exec_play(a);
+    } else if (strcmp(op, "screen") == 0) {
+        err = exec_screen(a);
+    } else if (strcmp(op, "evt") == 0) {
+        err = exec_evt(a);
+    } else if (strcmp(op, "led") == 0) {
+        err = exec_led(a);
+    } else {
+        ESP_LOGW(TAG, "CMD op=\"%s\" unknown — ignored", op);
+        err = ESP_ERR_NOT_SUPPORTED;
+    }
+
+    cJSON_Delete(root);
+    return err;
+}

plip_voice/main/cmd_exec.h

@@ -0,0 +1,28 @@
+#pragma once
+/*
+ * cmd_exec.h — ESP-NOW CMD executor for the PLIP voice annex (D5 contract).
+ *
+ * Handles JSON CMD frames: { "op": "play"|"screen"|"evt"|"led", "a": {...} }
+ * Routing:
+ *   play   → audio_play_async(a.p)
+ *   screen → log only (no display on PLIP)
+ *   evt    → log only
+ *   led    → log only
+ */
+
+#include "esp_err.h"
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Parse and dispatch a CMD JSON payload received from the master via ESP-NOW.
+ * `data` is a NUL-terminated JSON string of `len` bytes.
+ * Returns ESP_OK on success, ESP_ERR_INVALID_ARG on parse error,
+ * ESP_ERR_NOT_SUPPORTED for unknown ops. */
+esp_err_t cmd_exec_handle(const char *data, size_t len);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/main/conversation.c

@@ -0,0 +1,588 @@
+/*
+ * conversation.c — PLIP telephone conversation state machine.
+ *
+ * States: IDLE → DIALTONE → DIALING → RINGBACK | BUSY
+ *         RINGBACK (after ~2 s) → GREET → CONNECTED
+ *
+ * Transitions:
+ *   IDLE        + off-hook              → DIALTONE   (tones_dialtone_start)
+ *   DIALTONE    + first digit           → DIALING    (tones_stop)
+ *   DIALING     + ms_since_last > 3000  → RINGBACK or BUSY (route decision)
+ *   RINGBACK    + 2 s elapsed           → GREET      (tones_stop, turn_client_greeting, play WAV)
+ *   GREET       + play enqueued         → CONNECTED  (Stage 3: listen loop)
+ *   any state   + on-hook              → IDLE        (tones_stop, audio_stop, PA off)
+ *
+ * Routing table (known numbers → ringback; unknown → busy after 3 s silence):
+ *   "12", "3615", "15", "17", "18", "0142738200"
+ */
+
+#include "conversation.h"
+#include "dialer.h"
+#include "tones.h"
+#include "audio.h"
+#include "phone.h"
+#include "slic.h"
+#include "turn_client.h"
+#include "plip_gamebook.h"
+#if CONFIG_PLIP_DIAL_DTMF
+#include "dtmf.h"
+#endif
+
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "esp_log.h"
+#include "esp_timer.h"
+#include "esp_heap_caps.h"
+#include "esp_random.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#if CONFIG_PLIP_VOICE_REPLY
+/* Capture buffer sizing.
+ * PSRAM target: 8 s of 16kHz mono S16 + 44-byte WAV header.
+ *   8 s × 16000 samples/s × 2 bytes = 256000 bytes + 44 = 256044 → round to 256 KB.
+ * Fallback (internal heap, 4 s max):
+ *   4 s × 16000 × 2 + 44 = 128044 → round to 128 KB.
+ */
+#define CAPTURE_MAX_PSRAM   (256 * 1024)
+#define CAPTURE_MAX_IRAM    (128 * 1024)
+#define CAPTURE_MAX_MS_PSRAM  8000
+#define CAPTURE_MAX_MS_IRAM   4000
+#define CAPTURE_SILENCE_MS     900   /* end-of-speech: needs to ride through brief
+                                      * mid-sentence pauses so a whole phrase is
+                                      * captured (quiet handset voice). 900 ms. */
+#define REPLY_POLL_MS          200   /* interval for checking hook during playback */
+#define REPLY_PLAYBACK_EXTRA_MS 500  /* safety margin added to computed WAV duration */
+#define BETWEEN_TURNS_MS       300   /* short pause between capture rounds */
+#endif /* CONFIG_PLIP_VOICE_REPLY */
+
+#define TAG "conversation"
+
+/* Duration of ringback before picking up and fetching the greeting */
+#define RINGBACK_GREET_MS  2000
+
+typedef enum {
+    STATE_IDLE,
+    STATE_DIALTONE,
+    STATE_DIALING,
+    STATE_RINGBACK,
+    STATE_BUSY,
+    STATE_GREET,       /* fetching + playing NPC greeting (Stage 2) */
+    STATE_CONNECTED,   /* in-call — Stage 3 will add listen loop     */
+    STATE_GAMEBOOK,    /* audio "livre dont vous êtes le héros"      */
+} conv_state_t;
+
+static volatile conv_state_t s_state    = STATE_IDLE;
+static volatile bool         s_offhook  = false;
+static volatile bool         s_hook_changed = false;
+
+/* Ringback start timestamp (µs, from esp_timer_get_time) */
+static int64_t s_ringback_start_us = 0;
+
+/* Session ID for the current call (generated at ringback → greet transition) */
+static char s_sid[32] = {0};
+/* Dialed number LOCKED at routing time. The dialer can keep accumulating
+ * spurious rotary pulses (marginal hook contact) during the call, so we must
+ * NOT re-read dialer_current() for the greeting/reply — that polluted number
+ * would 404 at the gateway. Capture the clean routed number here once. */
+static char s_number[16] = {0};
+/* Scene this call hints on (e.g. "SCENE_WARNING"), locked at pickup like the
+ * number. Empty for player-dialled calls. Drives the local SD hint clip. */
+static char s_scene[40] = {0};
+
+/* Incoming-call mode: an NPC "calls" the phone. conversation_arm_incoming()
+ * stores the caller's persona number (and optional scene) and rings; when the
+ * handset is then picked up from IDLE, we skip the outgoing dialtone/dial/
+ * ringback and go straight to GREET with this number (the NPC speaks first). */
+static char s_incoming_number[16] = {0};
+static char s_incoming_scene[40] = {0};
+static volatile bool s_incoming_armed = false;
+
+/* Auto story-ring: the phone rings on its own at a random interval; picking up
+ * launches a random audio story straight away. If nobody answers, the bell
+ * stops after STORY_RING_TIMEOUT_MS and the next ring is rescheduled. */
+#define STORY_RING_MIN_MS      (15 * 60 * 1000)   /* 15 min */
+#define STORY_RING_SPAN_MS     (15 * 60 * 1000)   /* + up to 15 min → 30 max */
+#define STORY_RING_TIMEOUT_MS  (60 * 1000)        /* give up after 1 min */
+static volatile bool s_story_ring_armed = false;
+static int64_t       s_story_ring_start_us = 0;
+
+/* Known numbers: ringback when dialed */
+static const char *KNOWN[] = {
+    "12", "3615", "15", "17", "18", "0142738200", NULL
+};
+
+static bool is_known(const char *num)
+{
+    for (int i = 0; KNOWN[i] != NULL; i++) {
+        if (strcmp(num, KNOWN[i]) == 0) return true;
+    }
+    return false;
+}
+
+/* "SCENE_WARNING" → "warning": strip the SCENE_ prefix and lowercase, matching
+ * the SD pack filenames from tools/tts/generate_plip_sd_pack.py. */
+static void scene_slug(const char *scene, char *out, size_t cap)
+{
+    const char *p = scene;
+    if (strncmp(p, "SCENE_", 6) == 0) p += 6;
+    size_t i = 0;
+    for (; p[i] && i + 1 < cap; i++) {
+        char c = p[i];
+        out[i] = (c >= 'A' && c <= 'Z') ? (char)(c - 'A' + 'a') : c;
+    }
+    out[i] = '\0';
+}
+
+static void go_idle(void)
+{
+    if (s_story_ring_armed) {   /* hung up mid auto-ring → silence the bell */
+        phone_ring_stop();
+        slic_ring_stop();
+        s_story_ring_armed = false;
+    }
+    tones_stop();
+    audio_stop();
+    audio_pa_set(false);
+    dialer_reset();
+    plip_gamebook_end();   /* no-op if the gamebook wasn't running */
+#if CONFIG_PLIP_DIAL_DTMF
+    dtmf_stop();
+#endif
+    s_scene[0] = '\0';   /* clear scripted-call scene so a later dial-out call
+                          * doesn't inherit it (CONNECTED would skip listening). */
+    s_state = STATE_IDLE;
+    ESP_LOGI(TAG, "-> IDLE");
+}
+
+/* Begin an answered INCOMING call: stop ringing, lock the persona number, and
+ * jump straight to GREET (the NPC speaks first). Drives the conversation's own
+ * s_offhook directly because phone.c's debounced GPIO detection is unreliable
+ * during ringing — we trust the SLIC's raw off-hook reading instead. */
+static void enter_incoming_greet(void)
+{
+    s_incoming_armed = false;
+    phone_ring_stop();
+    slic_ring_stop();   /* belt-and-suspenders: kill the physical bell directly in
+                         * case phone.c's s_ringing flag desynced from slic.c's
+                         * (otherwise phone_ring_stop early-returns and the bell
+                         * keeps ringing through the answered call). */
+    tones_stop();
+    dialer_reset();
+    s_offhook = true;
+    snprintf(s_number, sizeof(s_number), "%s", s_incoming_number);
+    snprintf(s_scene, sizeof(s_scene), "%s", s_incoming_scene);
+    snprintf(s_sid, sizeof(s_sid), "%lld", (long long)esp_timer_get_time());
+    audio_pa_set(true);
+    s_state = STATE_GREET;
+    ESP_LOGI(TAG, "INCOMING pickup -> GREET num=%s sid=%s", s_number, s_sid);
+}
+
+/* Answered an auto story-ring: stop the bell and launch a random story right
+ * into STATE_GAMEBOOK (no menu). Drives s_offhook directly because the
+ * debounced GPIO hook detection is unreliable while the bell rings. */
+static void enter_story_from_ring(void)
+{
+    s_story_ring_armed = false;
+    phone_ring_stop();
+    slic_ring_stop();
+    tones_stop();
+    dialer_reset();
+    s_offhook = true;
+#if CONFIG_PLIP_DIAL_DTMF
+    dtmf_start();
+#endif
+    audio_pa_set(true);
+    plip_gamebook_begin_random();
+    s_state = STATE_GAMEBOOK;
+    ESP_LOGI(TAG, "story-ring pickup -> GAMEBOOK (random story)");
+}
+
+#if CONFIG_PLIP_AUTO_RING
+/* Rings the phone at a random 15–30 min interval. Only rings when the line is
+ * truly idle (on-hook, no other call/ring armed) and a story pack is present.
+ * The 1-minute no-answer timeout is handled in conv_task. */
+static void story_ring_task(void *arg)
+{
+    (void)arg;
+    for (;;) {
+        uint32_t wait_ms = STORY_RING_MIN_MS + (esp_random() % STORY_RING_SPAN_MS);
+        ESP_LOGI(TAG, "auto story-ring: next attempt in %u min",
+                 (unsigned)(wait_ms / 60000));
+        vTaskDelay(pdMS_TO_TICKS(wait_ms));
+
+        if (s_state == STATE_IDLE && !s_offhook && !s_incoming_armed
+            && !s_story_ring_armed && plip_gamebook_available()) {
+            s_story_ring_armed = true;
+            s_story_ring_start_us = esp_timer_get_time();
+            phone_ring_start();
+            ESP_LOGI(TAG, "auto story-ring: ringing (up to %d s)",
+                     STORY_RING_TIMEOUT_MS / 1000);
+        } else {
+            ESP_LOGI(TAG, "auto story-ring: line busy/unavailable — skipped");
+        }
+    }
+}
+#endif /* CONFIG_PLIP_AUTO_RING */
+
+static void conv_task(void *arg)
+{
+    (void)arg;
+    ESP_LOGI(TAG, "conversation task ready");
+
+    for (;;) {
+        vTaskDelay(pdMS_TO_TICKS(50));
+
+        /* Handle hook change events */
+        if (s_hook_changed) {
+            s_hook_changed = false;
+            if (!s_offhook) {
+                /* On-hook: always go idle regardless of current state */
+                if (s_state != STATE_IDLE) {
+                    ESP_LOGI(TAG, "on-hook -> IDLE");
+                    go_idle();
+                }
+                continue;
+            } else {
+                /* Off-hook from idle */
+                if (s_state == STATE_IDLE) {
+                    if (s_story_ring_armed) {
+                        /* Answered an auto story-ring → straight into a story. */
+                        enter_story_from_ring();
+                    } else if (s_incoming_armed) {
+                        /* INCOMING call answered (phone.c detected the edge). */
+                        enter_incoming_greet();
+                    } else if (plip_gamebook_available()) {
+                        /* Gamebook phone: a story pack is on the SD → pick up
+                         * launches the audio "livre dont vous êtes le héros".
+                         * Digits dialed choose the story then the answers. */
+                        dialer_reset();
+#if CONFIG_PLIP_DIAL_DTMF
+                        dtmf_start();
+#endif
+                        plip_gamebook_begin();
+                        s_state = STATE_GAMEBOOK;
+                        ESP_LOGI(TAG, "off-hook -> GAMEBOOK");
+                    } else {
+                        /* Outgoing call: dial tone, wait for digits. */
+                        dialer_reset();
+                        tones_dialtone_start();
+#if CONFIG_PLIP_DIAL_DTMF
+                        dtmf_start();
+#endif
+                        s_state = STATE_DIALTONE;
+                        ESP_LOGI(TAG, "off-hook -> DIALTONE");
+                    }
+                }
+                continue;
+            }
+        }
+
+        /* State machine polling */
+        switch (s_state) {
+        case STATE_IDLE:
+            /* Incoming-call pickup: phone.c's debounced edge detection is
+             * unreliable while the bell rings, so poll the SLIC's raw off-hook
+             * line directly — it reads the pickup cleanly mid-ring. */
+            if (s_incoming_armed && slic_is_offhook()) {
+                enter_incoming_greet();
+            } else if (s_story_ring_armed) {
+                /* Same raw-SLIC poll for the auto story-ring, plus the
+                 * no-answer timeout: stop the bell after 1 min unanswered. */
+                if (slic_is_offhook()) {
+                    enter_story_from_ring();
+                } else if ((esp_timer_get_time() - s_story_ring_start_us) / 1000
+                           >= STORY_RING_TIMEOUT_MS) {
+                    phone_ring_stop();
+                    slic_ring_stop();
+                    s_story_ring_armed = false;
+                    ESP_LOGI(TAG, "auto story-ring: no answer after %d s — stop",
+                             STORY_RING_TIMEOUT_MS / 1000);
+                }
+            }
+            break;
+
+        case STATE_DIALTONE:
+            if (!s_offhook) {
+                go_idle();
+                break;
+            }
+            /* First digit received → stop dialtone, enter dialing */
+            if (!dialer_idle()) {
+                tones_stop();
+                s_state = STATE_DIALING;
+                ESP_LOGI(TAG, "first digit -> DIALING");
+            }
+            break;
+
+        case STATE_GAMEBOOK:
+            if (!s_offhook) {
+                go_idle();
+                break;
+            }
+            /* Each dialed digit drives one choice. Consume a single digit at a
+             * time (take the first, reset) so the player can dial again right
+             * away — even over the narration, which gets interrupted. */
+            if (!dialer_idle()) {
+                int d = dialer_current()[0] - '0';
+                dialer_reset();
+                plip_gamebook_feed_digit(d);
+            }
+            break;
+
+        case STATE_DIALING:
+            if (!s_offhook) {
+                go_idle();
+                break;
+            }
+            /* Wait for 3 s of silence after last digit, then route */
+            if (dialer_ms_since_last() > 3000) {
+                const char *num = dialer_current();
+                if (is_known(num)) {
+                    ESP_LOGI(TAG, "route %s -> known (ringback)", num);
+                    /* Lock the routed number now — the dialer may pick up
+                     * spurious pulses later and we must keep posting "17". */
+                    snprintf(s_number, sizeof(s_number), "%s", num);
+                    tones_ringback_start();
+                    s_ringback_start_us = esp_timer_get_time();
+                    s_state = STATE_RINGBACK;
+                } else {
+                    ESP_LOGI(TAG, "route %s -> unknown (busy)", num);
+                    tones_busy_start();
+                    s_state = STATE_BUSY;
+                }
+            }
+            break;
+
+        case STATE_RINGBACK:
+            if (!s_offhook) {
+                go_idle();
+                break;
+            }
+            {
+                int64_t elapsed_ms =
+                    (esp_timer_get_time() - s_ringback_start_us) / 1000;
+                if (elapsed_ms >= RINGBACK_GREET_MS) {
+                    /* Stop ringback tone synchronously before fetching */
+                    tones_stop();
+#if CONFIG_PLIP_DIAL_DTMF
+                    /* Disarm DTMF before entering voice-capture phase */
+                    dtmf_stop();
+#endif
+                    /* Generate a session ID from timer ticks */
+                    snprintf(s_sid, sizeof(s_sid), "%lld",
+                             (long long)esp_timer_get_time());
+                    ESP_LOGI(TAG, "ringback done -> GREET (sid=%s num=%s)",
+                             s_sid, dialer_current());
+                    s_state = STATE_GREET;
+                }
+            }
+            break;
+
+        case STATE_GREET:
+            if (!s_offhook) {
+                go_idle();
+                break;
+            }
+            /* Prefer the local SD voice pack (no gateway, no model in RAM):
+             * play /sdcard/voice/greet_<number>.wav when present. Fall back to
+             * the live gateway greeting only when the pack has no clip for this
+             * NPC number (see tools/tts/generate_plip_sd_pack.py). */
+            {
+                char sd_greet[64];
+                snprintf(sd_greet, sizeof(sd_greet),
+                         "/sdcard/voice/greet_%s.wav", s_number);
+                FILE *tf = NULL;
+                if (audio_ensure_sd() && (tf = fopen(sd_greet, "rb")) != NULL) {
+                    fclose(tf);
+                    audio_play_async(sd_greet);
+                    ESP_LOGI(TAG, "GREET: local SD pack %s", sd_greet);
+                } else if (turn_client_greeting(s_sid, s_number,
+                                                "/spiffs/turn.wav")) {
+                    audio_play_async("/spiffs/turn.wav");
+                } else {
+                    ESP_LOGW(TAG, "no SD clip + gateway greeting failed — silent");
+                }
+            }
+            /* Scripted offline call: chain this scene's hint from the SD pack
+             * right after the greeting (the audio queue plays them back-to-back).
+             * Fully local — no gateway, no model. */
+            if (s_scene[0] != '\0' && audio_ensure_sd()) {
+                char slug[40], hint[80];
+                scene_slug(s_scene, slug, sizeof(slug));
+                snprintf(hint, sizeof(hint),
+                         "/sdcard/voice/hint_%s_l1_0.wav", slug);
+                FILE *hf = fopen(hint, "rb");
+                if (hf != NULL) {
+                    fclose(hf);
+                    audio_play_async(hint);
+                    ESP_LOGI(TAG, "GREET: chained SD hint %s", hint);
+                } else {
+                    ESP_LOGI(TAG, "GREET: no SD hint for scene %s (%s)",
+                             s_scene, hint);
+                }
+            }
+            s_state = STATE_CONNECTED;
+            ESP_LOGI(TAG, "-> CONNECTED");
+            break;
+
+        case STATE_CONNECTED:
+            /* Scripted offline call (greeting + SD hint already queued): there's
+             * no live two-way conversation to run — just let the clips play out
+             * and wait for the player to hang up. Skips the gateway/model listen
+             * loop entirely. The global on-hook check returns us to IDLE. */
+            if (s_scene[0] != '\0') {
+                vTaskDelay(pdMS_TO_TICKS(REPLY_POLL_MS));
+                break;
+            }
+#if CONFIG_PLIP_VOICE_REPLY
+            /*
+             * Stage 3 — LISTEN loop.
+             *
+             * Allocate capture buffer once from PSRAM (preferred) or internal
+             * heap.  Then loop: capture → POST reply → play → wait → repeat.
+             * Exit on any on-hook event.  Buffer freed before leaving.
+             */
+            {
+                /* --- Allocate capture buffer -------------------------------- */
+                uint8_t *cap_buf   = NULL;
+                size_t   cap_max   = 0;
+                int      cap_ms    = 0;
+
+                cap_buf = heap_caps_malloc(CAPTURE_MAX_PSRAM, MALLOC_CAP_SPIRAM);
+                if (cap_buf) {
+                    cap_max = CAPTURE_MAX_PSRAM;
+                    cap_ms  = CAPTURE_MAX_MS_PSRAM;
+                    ESP_LOGI(TAG, "listen: cap_buf %zu B from PSRAM", cap_max);
+                } else {
+                    cap_buf = malloc(CAPTURE_MAX_IRAM);
+                    if (cap_buf) {
+                        cap_max = CAPTURE_MAX_IRAM;
+                        cap_ms  = CAPTURE_MAX_MS_IRAM;
+                        ESP_LOGW(TAG, "listen: PSRAM unavail, cap_buf %zu B from heap (max %d s)",
+                                 cap_max, cap_ms / 1000);
+                    } else {
+                        ESP_LOGE(TAG, "listen: cap_buf alloc failed — staying silent");
+                        /* Remain in CONNECTED without looping */
+                        if (!s_offhook) go_idle();
+                        break;
+                    }
+                }
+
+                /* --- LISTEN loop ------------------------------------------- */
+                ESP_LOGI(TAG, "listen: entering loop (max %d s / silence %d ms)",
+                         cap_ms, CAPTURE_SILENCE_MS);
+
+                while (s_offhook) {
+                    /* HALF-DUPLEX: a telephone handset couples the earpiece into
+                     * the mic. Never capture while anything is playing, or the
+                     * playback feeds back and the line saturates. Wait for the
+                     * greeting/filler/reply to finish, then let the line settle. */
+                    vTaskDelay(pdMS_TO_TICKS(120));   /* let a just-queued clip start (was 250) */
+                    while (s_offhook && audio_is_playing()) {
+                        vTaskDelay(pdMS_TO_TICKS(50));
+                    }
+                    if (!s_offhook) break;
+                    vTaskDelay(pdMS_TO_TICKS(200));   /* let the I2S DMA tail drain */
+
+                    /* Capture the caller utterance with the PA LEFT ON. The loop
+                     * already waits for any playback to finish above, so nothing
+                     * is driving the earpiece during capture — there is no echo to
+                     * mute. Cutting the PA here was found to collapse the mic level
+                     * (captures came back near-silent / DC only), so keep it on. */
+                    int n = audio_capture_wav(cap_buf, cap_max,
+                                              cap_ms, CAPTURE_SILENCE_MS);
+                    if (!s_offhook) break;   /* hung up during capture */
+
+                    if (n <= 44) {
+                        ESP_LOGD(TAG, "listen: no voice (n=%d)", n);
+                        continue;
+                    }
+
+                    ESP_LOGI(TAG, "listen: captured %d bytes, posting to gateway", n);
+
+                    /* Filler "un instant, je traite votre demande" plays while the
+                     * reply is synthesised (the POST blocks for several seconds). */
+                    audio_play_async("/spiffs/wait.wav");
+
+                    esp_err_t ret = turn_client_reply(s_sid, s_number,
+                                                      cap_buf, (size_t)n,
+                                                      "/spiffs/reply.wav");
+                    if (!s_offhook) break;   /* hung up during HTTP round-trip */
+
+                    if (ret != ESP_OK) {
+                        ESP_LOGW(TAG, "listen: turn_client_reply failed (%s) — skipping",
+                                 esp_err_to_name(ret));
+                        continue;
+                    }
+
+                    /* Let the filler finish before the reply (no overlap), then play
+                     * the reply. The loop top waits for it to end before re-capturing. */
+                    while (s_offhook && audio_is_playing()) {
+                        vTaskDelay(pdMS_TO_TICKS(50));
+                    }
+                    if (!s_offhook) break;
+                    audio_play_async("/spiffs/reply.wav");
+                }
+
+                /* --- Cleanup ----------------------------------------------- */
+                free(cap_buf);
+                ESP_LOGI(TAG, "listen: loop exited (offhook=%d)", (int)s_offhook);
+
+                if (!s_offhook) go_idle();
+            }
+#else
+            /* Stage 3 disabled — STATE_CONNECTED is terminal */
+            if (!s_offhook) {
+                go_idle();
+            }
+#endif /* CONFIG_PLIP_VOICE_REPLY */
+            break;
+
+        case STATE_BUSY:
+            if (!s_offhook) {
+                go_idle();
+            }
+            break;
+        }
+    }
+}
+
+void conversation_init(void)
+{
+    s_state        = STATE_IDLE;
+    s_offhook      = false;
+    s_hook_changed = false;
+    /* Stack: STATE_GREET needs 6144 (esp_http_client + file I/O).
+     * STATE_CONNECTED listen loop (Stage 3) adds turn_client_reply (~1100 B
+     * locals) + stat() call → bump to 8192 when Stage 3 is compiled in. */
+#if CONFIG_PLIP_VOICE_REPLY
+    xTaskCreate(conv_task, "conv", 8192, NULL, 4, NULL);
+#else
+    xTaskCreate(conv_task, "conv", 6144, NULL, 4, NULL);
+#endif
+#if CONFIG_PLIP_AUTO_RING
+    /* Auto story-ring scheduler: rings every 15–30 min when idle. */
+    xTaskCreate(story_ring_task, "storyring", 3072, NULL, 3, NULL);
+#endif
+    ESP_LOGI(TAG, "conversation init");
+}
+
+void conversation_on_hook_change(bool offhook)
+{
+    s_offhook      = offhook;
+    s_hook_changed = true;
+}
+
+void conversation_arm_incoming(const char *number, const char *scene)
+{
+    snprintf(s_incoming_number, sizeof(s_incoming_number), "%s",
+             (number && number[0]) ? number : "17");
+    snprintf(s_incoming_scene, sizeof(s_incoming_scene), "%s",
+             (scene && scene[0]) ? scene : "");
+    s_incoming_armed = true;
+    phone_ring_start();
+    ESP_LOGI(TAG, "incoming call armed (num=%s scene=%s) — ringing until pickup",
+             s_incoming_number, s_incoming_scene[0] ? s_incoming_scene : "-");
+}

plip_voice/main/conversation.h

@@ -0,0 +1,13 @@
+#pragma once
+#include <stdbool.h>
+
+void conversation_init(void);
+void conversation_on_hook_change(bool offhook);
+
+/* Arm an INCOMING call from NPC persona `number` and start ringing. When the
+ * handset is picked up from idle, the conversation jumps straight to the
+ * greeting (persona speaks first) instead of the outgoing dial-tone path.
+ * `scene` (optional, e.g. "SCENE_WARNING") makes the phone play that scene's
+ * hint from its local SD pack right after the greeting; pass NULL/"" for a
+ * greeting-only call. */
+void conversation_arm_incoming(const char *number, const char *scene);

plip_voice/main/dialer.c

@@ -0,0 +1,73 @@
+/*
+ * dialer.c — Digit accumulator for the PLIP telephone state machine.
+ *
+ * Digits are pushed from two sources:
+ *   1. HTTP /debug/dial?number=NNNN  (net.c handler)
+ *   2. Rotary pulse detection (phone.c, CONFIG_PLIP_DIAL_PULSE)
+ *
+ * Thread-safety: all state is protected by atomic operations on s_last_us
+ * and simple writes to s_len / s_num (called from a single phone task or
+ * HTTP task at a time — no concurrent push expected).
+ */
+
+#include "dialer.h"
+
+#include "esp_log.h"
+#include "esp_timer.h"
+
+#include <string.h>
+#include <stdint.h>
+
+#define TAG "dialer"
+
+#define MAX_DIGITS 12
+
+static char    s_num[MAX_DIGITS + 1];
+static int     s_len;
+static int64_t s_last_us;  /* esp_timer_get_time() at last push */
+
+void dialer_init(void)
+{
+    memset(s_num, 0, sizeof(s_num));
+    s_len     = 0;
+    s_last_us = 0;
+    ESP_LOGI(TAG, "dialer init");
+}
+
+void dialer_push_digit(int d)
+{
+    if (d < 0 || d > 9) return;
+    if (s_len >= MAX_DIGITS) return;  /* silently discard overflow */
+
+    s_num[s_len++] = (char)('0' + d);
+    s_num[s_len]   = '\0';
+    s_last_us      = esp_timer_get_time();
+
+    ESP_LOGI(TAG, "digit %d -> number so far: \"%s\"", d, s_num);
+}
+
+void dialer_reset(void)
+{
+    memset(s_num, 0, sizeof(s_num));
+    s_len     = 0;
+    s_last_us = 0;
+    ESP_LOGI(TAG, "dialer reset");
+}
+
+const char *dialer_current(void)
+{
+    return s_num;
+}
+
+bool dialer_idle(void)
+{
+    return s_len == 0;
+}
+
+int dialer_ms_since_last(void)
+{
+    if (s_last_us == 0) return 0;
+    int64_t elapsed_us = esp_timer_get_time() - s_last_us;
+    int ms = (int)(elapsed_us / 1000);
+    return (ms < 0) ? 0 : ms;
+}

plip_voice/main/dialer.h

@@ -0,0 +1,9 @@
+#pragma once
+#include <stdbool.h>
+
+void        dialer_init(void);
+void        dialer_push_digit(int d);
+void        dialer_reset(void);
+const char *dialer_current(void);
+bool        dialer_idle(void);
+int         dialer_ms_since_last(void);

plip_voice/main/dtmf.c

@@ -0,0 +1,323 @@
+/*
+ * dtmf.c — DTMF (touch-tone) detector using the Goertzel algorithm.
+ *
+ * Detection pipeline per 20 ms frame (320 samples @ 16 kHz):
+ *   1. Compute Goertzel power for 8 DTMF frequencies (4 low + 4 high groups).
+ *   2. Find the strongest frequency in each group (best_low, best_high).
+ *   3. Apply three guards:
+ *        a) Absolute energy threshold — both must exceed DTMF_ENERGY_THRESH.
+ *        b) Group dominance ratio    — winner must be > DTMF_DOMINANT_RATIO×
+ *                                      times the second-best in the same group.
+ *        c) Twist guard             — energy ratio (low/high) must be within
+ *                                      [1/DTMF_TWIST_MAX, DTMF_TWIST_MAX].
+ *   4. Debounce:
+ *        - Require DTMF_CONFIRM_FRAMES consecutive matching frames to emit.
+ *        - Require DTMF_RELEASE_FRAMES of silence/mismatch before re-arming.
+ *
+ * Threshold rationale:
+ *   DTMF_ENERGY_THRESH  = 4000000
+ *     Goertzel power is mean-squared × N².  A -30 dBFS sine at 16-bit PCM
+ *     (amplitude ≈ 1000 LSB) gives power ≈ (1000²/2) × 320² / 320 ≈ 1.6e8.
+ *     -50 dBFS (amplitude ≈ 100) gives ≈ 1.6e6.  We set the floor at 4e6
+ *     (≈ -47 dBFS) to reject noise while allowing quiet handset microphones.
+ *
+ *   DTMF_DOMINANT_RATIO = 4.0f  (≈ 6 dB separation within a group)
+ *     A real DTMF tone drives exactly one row and one column.  If two
+ *     frequencies in the same group are within 6 dB of each other, it is
+ *     more likely noise or voice than a keypad press.
+ *
+ *   DTMF_TWIST_MAX = 8.0f  (≈ 9 dB)
+ *     ITU-T Q.24 allows up to ±8 dB twist between low/high group.  We use
+ *     8× power ratio which corresponds to ≈ 9 dB — slightly relaxed to
+ *     accommodate the varied mic responses of vintage telephone handsets.
+ *
+ *   DTMF_CONFIRM_FRAMES = 2  (2 × 20 ms = 40 ms minimum tone duration)
+ *     ITU-T Q.24 specifies ≥ 40 ms tone duration for valid DTMF.
+ *
+ *   DTMF_RELEASE_FRAMES = 1  (≥ 20 ms inter-digit silence required)
+ *     Prevents a single sustained keypress from re-triggering.
+ */
+
+/* sdkconfig.h must be included before any CONFIG_* test so the preprocessor
+ * has the symbol defined when the #if guard below is evaluated. */
+#include "sdkconfig.h"
+
+#if CONFIG_PLIP_DIAL_DTMF
+
+#include "dtmf.h"
+#include "audio.h"
+#include "dialer.h"
+
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "esp_log.h"
+
+#include <math.h>
+#include <string.h>
+#include <stdbool.h>
+#include <stdint.h>
+
+#define TAG "dtmf"
+
+/* -------------------------------------------------------------------------
+ * Detection thresholds (see rationale in file header)
+ * ---------------------------------------------------------------------- */
+
+/* Minimum Goertzel power (mean-squared × N) for a frequency to be considered
+ * "present".  Both the row and column candidate must exceed this. */
+#define DTMF_ENERGY_THRESH    4000000LL
+
+/* Minimum ratio of best/second-best power within a group.  Below this the
+ * group is ambiguous (noise / voice) and the frame is rejected. */
+#define DTMF_DOMINANT_RATIO   4.0f
+
+/* Maximum ratio of low-group / high-group power (and its inverse).
+ * Exceeding this means one group is far stronger than expected for DTMF. */
+#define DTMF_TWIST_MAX        8.0f
+
+/* Consecutive frames required before a digit is reported. */
+#define DTMF_CONFIRM_FRAMES   2
+
+/* Frames of "no valid tone" required between two reports. */
+#define DTMF_RELEASE_FRAMES   1
+
+/* -------------------------------------------------------------------------
+ * DTMF frequency table
+ * ---------------------------------------------------------------------- */
+
+#define NUM_LOW   4
+#define NUM_HIGH  4
+
+static const float LOW_FREQS[NUM_LOW]  = { 697.0f, 770.0f, 852.0f, 941.0f };
+static const float HIGH_FREQS[NUM_HIGH] = { 1209.0f, 1336.0f, 1477.0f, 1633.0f };
+
+/* DTMF matrix: [low_idx][high_idx] → character.
+ * Column 3 (1633 Hz) maps to ABCD which are unused on standard phones → '\0'. */
+static const char DTMF_MATRIX[NUM_LOW][NUM_HIGH] = {
+    { '1', '2', '3', '\0' },  /* 697 Hz row */
+    { '4', '5', '6', '\0' },  /* 770 Hz row */
+    { '7', '8', '9', '\0' },  /* 852 Hz row */
+    { '*', '0', '#', '\0' },  /* 941 Hz row */
+};
+
+/* -------------------------------------------------------------------------
+ * Goertzel coefficient cache (precomputed at first call)
+ * ---------------------------------------------------------------------- */
+
+#define FS     16000  /* sample rate */
+#define N_SAMP 320    /* frame size  */
+
+static bool  s_coeff_ready = false;
+static float s_low_coeff[NUM_LOW];
+static float s_high_coeff[NUM_HIGH];
+
+static void precompute_coeffs(void)
+{
+    for (int i = 0; i < NUM_LOW; i++) {
+        float k = (float)N_SAMP * LOW_FREQS[i] / (float)FS;
+        s_low_coeff[i] = 2.0f * cosf(2.0f * (float)M_PI * k / (float)N_SAMP);
+    }
+    for (int i = 0; i < NUM_HIGH; i++) {
+        float k = (float)N_SAMP * HIGH_FREQS[i] / (float)FS;
+        s_high_coeff[i] = 2.0f * cosf(2.0f * (float)M_PI * k / (float)N_SAMP);
+    }
+    s_coeff_ready = true;
+}
+
+/* -------------------------------------------------------------------------
+ * Goertzel power for a single frequency
+ * Power = Q1² + Q2² − Q1·Q2·coeff  (unnormalised, proportional to amplitude²)
+ * ---------------------------------------------------------------------- */
+
+static float goertzel_power(const int16_t *samples, int n, float coeff)
+{
+    float q1 = 0.0f, q2 = 0.0f;
+    for (int i = 0; i < n; i++) {
+        float q0 = coeff * q1 - q2 + (float)samples[i];
+        q2 = q1;
+        q1 = q0;
+    }
+    return q1 * q1 + q2 * q2 - q1 * q2 * coeff;
+}
+
+/* -------------------------------------------------------------------------
+ * Public API: dtmf_detect_frame
+ * ---------------------------------------------------------------------- */
+
+char dtmf_detect_frame(const int16_t *mono, int n)
+{
+    if (n <= 0 || !mono) return '\0';
+
+    if (!s_coeff_ready) precompute_coeffs();
+
+    /* Compute Goertzel power for all 8 frequencies */
+    float low_pow[NUM_LOW], high_pow[NUM_HIGH];
+    for (int i = 0; i < NUM_LOW; i++)
+        low_pow[i]  = goertzel_power(mono, n, s_low_coeff[i]);
+    for (int i = 0; i < NUM_HIGH; i++)
+        high_pow[i] = goertzel_power(mono, n, s_high_coeff[i]);
+
+    /* Find strongest in each group */
+    int   best_low = 0, best_high = 0;
+    float max_low  = low_pow[0], max_high = high_pow[0];
+    for (int i = 1; i < NUM_LOW; i++) {
+        if (low_pow[i] > max_low)  { max_low  = low_pow[i];  best_low  = i; }
+    }
+    for (int i = 1; i < NUM_HIGH; i++) {
+        if (high_pow[i] > max_high) { max_high = high_pow[i]; best_high = i; }
+    }
+
+    /* Guard (a): absolute energy threshold */
+    if ((int64_t)max_low < DTMF_ENERGY_THRESH || (int64_t)max_high < DTMF_ENERGY_THRESH)
+        goto no_tone;
+
+    /* Guard (b): group dominance — find second-best in each group */
+    {
+        float second_low = 0.0f, second_high = 0.0f;
+        for (int i = 0; i < NUM_LOW; i++) {
+            if (i != best_low && low_pow[i] > second_low)
+                second_low = low_pow[i];
+        }
+        for (int i = 0; i < NUM_HIGH; i++) {
+            if (i != best_high && high_pow[i] > second_high)
+                second_high = high_pow[i];
+        }
+        /* If second-best is within DTMF_DOMINANT_RATIO of best, ambiguous */
+        if (second_low  > 0.0f && max_low  < DTMF_DOMINANT_RATIO * second_low)
+            goto no_tone;
+        if (second_high > 0.0f && max_high < DTMF_DOMINANT_RATIO * second_high)
+            goto no_tone;
+    }
+
+    /* Guard (c): twist — power ratio must be within [1/TWIST_MAX, TWIST_MAX] */
+    {
+        float ratio = max_low / max_high;
+        if (ratio > DTMF_TWIST_MAX || ratio < (1.0f / DTMF_TWIST_MAX))
+            goto no_tone;
+    }
+
+    /* --- Debounce state (static) --- */
+    {
+        static char  s_candidate      = '\0';
+        static int   s_confirm_count  = 0;
+        static int   s_release_count  = 0;
+        static bool  s_armed          = true;  /* true = ready to report */
+
+        char sym = DTMF_MATRIX[best_low][best_high];
+        if (sym == '\0') goto no_tone;  /* 1633 Hz column — ignored */
+
+        /* Reset release counter: we have a tone */
+        s_release_count = 0;
+
+        if (!s_armed) {
+            /* Waiting for silence/release before accepting next press */
+            return '\0';
+        }
+
+        if (sym == s_candidate) {
+            s_confirm_count++;
+        } else {
+            s_candidate     = sym;
+            s_confirm_count = 1;
+        }
+
+        if (s_confirm_count >= DTMF_CONFIRM_FRAMES) {
+            /* Confirmed — report and disarm until release */
+            s_confirm_count = 0;
+            s_candidate     = '\0';
+            s_armed         = false;
+            return sym;
+        }
+        return '\0';
+
+no_tone:
+        /* No valid tone detected: advance release counter */
+        ;  /* label must precede a statement */
+        s_confirm_count = 0;
+        s_candidate     = '\0';
+        if (!s_armed) {
+            s_release_count++;
+            if (s_release_count >= DTMF_RELEASE_FRAMES) {
+                s_armed         = true;
+                s_release_count = 0;
+            }
+        }
+        return '\0';
+    }
+}
+
+/* -------------------------------------------------------------------------
+ * Background capture task
+ * ---------------------------------------------------------------------- */
+
+#define DTMF_TASK_STACK  4096
+#define DTMF_TASK_PRIO   3
+#define DTMF_FRAME_SIZE  320
+
+static volatile bool s_armed_flag = false;  /* true = task should process frames */
+static TaskHandle_t  s_task_handle = NULL;
+
+static void dtmf_task(void *arg)
+{
+    (void)arg;
+    int16_t  frame[DTMF_FRAME_SIZE];
+    int64_t  rms_sq;
+
+    ESP_LOGI(TAG, "dtmf_task started");
+
+    /* Open the capture stream once and keep it open.
+     * Full-duplex: TX (speaker) stays active; RX is already enabled at boot.
+     * We pass generous max_ms / silence_ms since we never call capture_end
+     * while the call is in progress — dtmf_stop() just clears the armed flag. */
+    if (audio_capture_begin(3600000, 3600000) != 0) {
+        ESP_LOGE(TAG, "dtmf_task: capture_begin failed — task exits");
+        s_task_handle = NULL;
+        vTaskDelete(NULL);
+        return;
+    }
+
+    for (;;) {
+        if (!s_armed_flag) {
+            /* Disarmed: drain frames slowly so the RX FIFO doesn't overflow */
+            audio_capture_read_frame(frame, DTMF_FRAME_SIZE, &rms_sq);
+            vTaskDelay(pdMS_TO_TICKS(20));
+            continue;
+        }
+
+        int got = audio_capture_read_frame(frame, DTMF_FRAME_SIZE, &rms_sq);
+        if (got <= 0) continue;
+
+        char sym = dtmf_detect_frame(frame, got);
+        if (sym == '\0') continue;
+
+        ESP_LOGI(TAG, "DTMF detected: '%c'", sym);
+        if (sym >= '0' && sym <= '9') {
+            dialer_push_digit(sym - '0');
+        }
+        /* '*' and '#' are logged only — no dialer push for now */
+    }
+}
+
+void dtmf_start(void)
+{
+    if (!s_task_handle) {
+        /* Create the task once */
+        BaseType_t ok = xTaskCreatePinnedToCore(
+            dtmf_task, "dtmf", DTMF_TASK_STACK, NULL, DTMF_TASK_PRIO,
+            &s_task_handle, 1);
+        if (ok != pdPASS) {
+            ESP_LOGE(TAG, "dtmf_start: xTaskCreate failed");
+            return;
+        }
+    }
+    s_armed_flag = true;
+    ESP_LOGI(TAG, "dtmf_start: DTMF detection armed");
+}
+
+void dtmf_stop(void)
+{
+    s_armed_flag = false;
+    ESP_LOGI(TAG, "dtmf_stop: DTMF detection disarmed");
+}
+
+#endif /* CONFIG_PLIP_DIAL_DTMF */

plip_voice/main/dtmf.h

@@ -0,0 +1,57 @@
+#pragma once
+/*
+ * dtmf.h — DTMF (touch-tone) detector via Goertzel algorithm.
+ *
+ * API:
+ *   dtmf_detect_frame()  — pure signal processing, no I/O, testable standalone
+ *   dtmf_start() / dtmf_stop() — arm/disarm the background capture task
+ *
+ * The capture task reads 20 ms frames from audio_capture_read_frame() and calls
+ * dialer_push_digit() for confirmed digit presses (digits 0-9 only).
+ *
+ * Guard: all declarations and the task body are compiled only when
+ *        CONFIG_PLIP_DIAL_DTMF is set (see Kconfig.projbuild).
+ */
+
+#include "sdkconfig.h"
+
+#if CONFIG_PLIP_DIAL_DTMF
+
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * Analyse one 20 ms frame (320 samples at 16 kHz) for a DTMF tone.
+ *
+ * Returns the detected character ('0'-'9', '*', '#') once per confirmed press,
+ * or '\0' when nothing is detected or debounce is still pending.
+ *
+ * Debounce rules (internal static state):
+ *   - A symbol must appear in ≥ DTMF_CONFIRM_FRAMES consecutive frames to be
+ *     reported.
+ *   - After a report, at least DTMF_RELEASE_FRAMES of "no tone" must be seen
+ *     before the same (or another) symbol can be reported again.
+ */
+char dtmf_detect_frame(const int16_t *mono, int n);
+
+/*
+ * Start the DTMF background task (created once; idempotent re-arm).
+ * The task reads microphone frames and pushes confirmed 0-9 digits to the
+ * dialer.  Must be called after audio_init().
+ */
+void dtmf_start(void);
+
+/*
+ * Disarm the DTMF task.  The task suspends itself; the RX stream continues
+ * for the benefit of the voice capture path (full-duplex architecture).
+ */
+void dtmf_stop(void);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* CONFIG_PLIP_DIAL_DTMF */

plip_voice/main/es8388.c

@@ -0,0 +1,286 @@
+/*
+ * es8388.c — ES8388 codec driver (IDF 5.x, legacy I2C driver API).
+ *
+ * Register map reference: ES8388 datasheet rev 1.6 (Everest Semiconductor).
+ * Init sequence derived from:
+ *   - Espressif esp-adf es8388.c (Apache 2.0)
+ *   - schreibfaul1/ESP32-audioI2S ES8388 init (MIT)
+ *   - AI-Thinker SDK AudioKit driver
+ *
+ * The sequence configures the ES8388 for:
+ *   - Master clock: MCLK from ESP32 GPIO0 at 12.288 MHz (or 256*Fs at 16kHz)
+ *   - I2S format: I2S Philips, 16-bit, stereo
+ *   - DAC: LOUT1/ROUT1 → headphone amp (GPIO21 PA_ENABLE must be HIGH)
+ *   - ADC: LINPUT1/RINPUT1 mic (differential), gain 24 dB
+ *   - Sample rate: 16 kHz (MCLKDIV = 256Fs)
+ */
+
+#include "es8388.h"
+#include "board_config.h"
+
+#include <string.h>
+
+#include "driver/i2c.h"
+#include "driver/gpio.h"
+#include "esp_log.h"
+
+#define TAG "es8388"
+
+/* ── ES8388 register addresses ──────────────────────────────────────────── */
+#define ES8388_CHIP_CTL1        0x00  /* CHIP_CTL1 */
+#define ES8388_CHIP_CTL2        0x01  /* CHIP_CTL2 */
+#define ES8388_CHIP_POWER       0x02  /* CHIP_POWER (ADCPD, DACPD, etc.) */
+#define ES8388_ADC_POWER        0x03  /* ADCPOWER */
+#define ES8388_DAC_POWER        0x04  /* DACPOWER */
+#define ES8388_CHIP_LP          0x05  /* CHIP_LP */
+#define ES8388_CHIP_CTL3        0x06  /* CHIP_CTL3 */
+#define ES8388_ADC_CTL1         0x09  /* ADCCONTROL1 — PGA gain */
+#define ES8388_ADC_CTL2         0x0A  /* ADCCONTROL2 — input select (LINSEL/RINSEL) */
+#define ES8388_ADC_CTL3         0x0B  /* ADCCONTROL3 — DS/DS filter select */
+#define ES8388_ADC_CTL4         0x0C  /* ADCCONTROL4 — I2S format / word length */
+#define ES8388_ADC_CTL5         0x0D  /* ADCCONTROL5 — MCLK divider (RATIO=256) */
+#define ES8388_ADC_CTL7         0x0F  /* ADCCONTROL7 — HPF enable/config */
+#define ES8388_ADC_CTL8         0x10  /* ADCCONTROL8 — ADC L volume */
+#define ES8388_ADC_CTL9         0x11  /* ADCCONTROL9 — ADC R volume */
+#define ES8388_DAC_CTL1         0x17  /* DACCONTROL1 — I2S word len / format */
+#define ES8388_DAC_CTL2         0x18  /* DACCONTROL2 — MCLK divider */
+#define ES8388_DAC_CTL3         0x19  /* DACCONTROL3 — mute */
+#define ES8388_DAC_CTL4         0x1A  /* DACCONTROL4 — LDACVOL */
+#define ES8388_DAC_CTL5         0x1B  /* DACCONTROL5 — RDACVOL */
+#define ES8388_DAC_CTL16        0x26  /* DACCONTROL16 — L/R mixer config */
+#define ES8388_DAC_CTL17        0x27  /* DACCONTROL17 — L mixer gain */
+#define ES8388_DAC_CTL20        0x2A  /* DACCONTROL20 — R mixer gain */
+#define ES8388_DAC_CTL21        0x2B  /* DACCONTROL21 — ADC/DAC LRCK sync (bit7=1 for full-duplex) */
+#define ES8388_DAC_CTL22        0x2C  /* DACCONTROL22 — (unused in this config) */
+#define ES8388_DAC_CTL23        0x2D  /* DACCONTROL23 — (unused in this config) */
+#define ES8388_DAC_CTL24        0x2E  /* DACCONTROL24 — LOUT1VOL (OUT1 left volume) */
+#define ES8388_DAC_CTL25        0x2F  /* DACCONTROL25 — ROUT1VOL (OUT1 right volume) */
+#define ES8388_DAC_CTL26        0x30  /* DACCONTROL26 — LOUT2VOL (OUT2 left volume) */
+
+/* Volume register: 0x00=0dB (max), 0x21=-33dB step per 1.5dB, 0x24=mute. */
+#define ES8388_VOL_MAX  0x00
+#define ES8388_VOL_0DB  0x00
+#define ES8388_VOL_MUTE 0x24
+
+/* ── I2C helpers ─────────────────────────────────────────────────────────── */
+
+static esp_err_t i2c_write_reg(uint8_t reg, uint8_t value)
+{
+    uint8_t buf[2] = { reg, value };
+    esp_err_t ret = i2c_master_write_to_device(PLIP_I2C_PORT,
+                                                PLIP_ES8388_ADDR,
+                                                buf, sizeof(buf),
+                                                pdMS_TO_TICKS(100));
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "write reg 0x%02X=0x%02X failed: %s",
+                 reg, value, esp_err_to_name(ret));
+    }
+    return ret;
+}
+
+esp_err_t es8388_read_reg(uint8_t reg, uint8_t *value)
+{
+    return i2c_master_write_read_device(PLIP_I2C_PORT,
+                                        PLIP_ES8388_ADDR,
+                                        &reg, 1,
+                                        value, 1,
+                                        pdMS_TO_TICKS(100));
+}
+
+esp_err_t es8388_write_reg(uint8_t reg, uint8_t value)
+{
+    return i2c_write_reg(reg, value);
+}
+
+/* ── Public API ──────────────────────────────────────────────────────────── */
+
+esp_err_t es8388_init(void)
+{
+    /* Initialise I2C master on the A1S bus. */
+    i2c_config_t conf = {
+        .mode             = I2C_MODE_MASTER,
+        .sda_io_num       = PLIP_I2C_SDA,
+        .scl_io_num       = PLIP_I2C_SCL,
+        .sda_pullup_en    = GPIO_PULLUP_ENABLE,
+        .scl_pullup_en    = GPIO_PULLUP_ENABLE,
+        .master.clk_speed = PLIP_I2C_FREQ_HZ,
+    };
+    esp_err_t ret = i2c_param_config(PLIP_I2C_PORT, &conf);
+    if (ret != ESP_OK) return ret;
+    ret = i2c_driver_install(PLIP_I2C_PORT, conf.mode, 0, 0, 0);
+    if (ret != ESP_OK && ret != ESP_ERR_INVALID_STATE) {
+        /* ESP_ERR_INVALID_STATE means the driver is already installed — ok. */
+        ESP_LOGE(TAG, "i2c_driver_install: %s", esp_err_to_name(ret));
+        return ret;
+    }
+
+    /* Quick device presence check. */
+    uint8_t chip_id = 0;
+    ret = es8388_read_reg(ES8388_CHIP_CTL1, &chip_id);
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "ES8388 not found on I2C bus (addr=0x%02X): %s",
+                 PLIP_ES8388_ADDR, esp_err_to_name(ret));
+        return ret;
+    }
+    ESP_LOGI(TAG, "ES8388 detected: CHIP_CTL1=0x%02X", chip_id);
+
+    /* ── Full power-up + init sequence ─────────────────────────────────── */
+
+    /* 1. Reset. */
+    if (i2c_write_reg(ES8388_CHIP_CTL1, 0x80) != ESP_OK) return ESP_FAIL;
+    vTaskDelay(pdMS_TO_TICKS(20));
+    if (i2c_write_reg(ES8388_CHIP_CTL1, 0x00) != ESP_OK) return ESP_FAIL;
+
+    /* 2. Power down all blocks first (prevents pop on line-up). */
+    if (i2c_write_reg(ES8388_CHIP_POWER, 0xFF) != ESP_OK) return ESP_FAIL;
+
+    /* 3. CONTROL1: WORK_MODE = record+playback (0x10), VMIDSEL=10 (0x02).
+     * 0x12 = 0b00010010 — enables both ADC and DAC paths.
+     * Previous value 0x05 only enabled playback (WORK_MODE=00). */
+    if (i2c_write_reg(ES8388_CHIP_CTL1, 0x12) != ESP_OK) return ESP_FAIL;
+
+    /* 3b. CONTROL2 (0x01): VROI=0, LPVrefBuf=0, normal operation. */
+    if (i2c_write_reg(ES8388_CHIP_CTL2, 0x50) != ESP_OK) return ESP_FAIL;
+
+    /* 3c. Internal DLL stabilisation for LOW sample rates (16 kHz). UNDOCUMENTED
+     * registers 0x35/0x37/0x39 — the proven A252 driver (hardware/projects/
+     * slic-phone Es8388Driver.cpp) sets these to "disable internal DLL for low
+     * sample-rate stability". WITHOUT them the ADC clock domain is unstable at
+     * 16 kHz and the ADC outputs a frozen DC value (capture = flat, no AC) even
+     * though analog signal is present on the LIN pin — exactly our symptom. The
+     * DAC tolerates it, which is why playback worked but capture didn't. They
+     * must be set here in the boot sequence (a live poke can't re-lock the DLL). */
+    if (i2c_write_reg(0x35, 0xA0) != ESP_OK) return ESP_FAIL;
+    if (i2c_write_reg(0x37, 0xD0) != ESP_OK) return ESP_FAIL;
+    if (i2c_write_reg(0x39, 0xD0) != ESP_OK) return ESP_FAIL;
+
+    /* 4. Clock: MCLK divider = 256*Fs, DAC SRC = MCLK. */
+    if (i2c_write_reg(0x08, 0x00) != ESP_OK) return ESP_FAIL; /* MASTERMODE = slave */
+
+    /* 5. ADC power: power down to allow clean register writes. */
+    if (i2c_write_reg(ES8388_ADC_POWER, 0xFF) != ESP_OK) return ESP_FAIL;
+
+    /* 6. ADC config (follows Espressif esp-codec-dev reference sequence):
+     * - ADCCONTROL1 (0x09) = 0xBB: MIC PGA +24dB L and R
+     * - ADCCONTROL2 (0x0A) = 0x50: LINSEL=10 LIN2/RIN2 (telephone handset mic)
+     *   Value 0x00 = LIN1 differential, 0x50 = LIN2 single-ended (A1S combiné)
+     * - ADCCONTROL3 (0x0B) = 0x02: DS filter select (required by reference)
+     * - ADCCONTROL4 (0x0C) = 0x0C: I2S Philips 16-bit word length
+     * - ADCCONTROL5 (0x0D) = 0x02: ADCFsMode SINGLE SPEED RATIO=256 (16kHz@MCLK 4.096MHz)
+     * - ADCCONTROL8/9 (0x10/0x11) = 0x00: ADC digital volume 0dB */
+    if (i2c_write_reg(ES8388_ADC_CTL1, 0x88) != ESP_OK) return ESP_FAIL; /* MIC PGA +24dB L+R — the SLIC handset mic is QUIET (~2-5% FS at +12dB); +24dB lifts speech above the capture VAD onset. The old "+24dB too hot" note was an artifact of the broken ADC (pre-DLL-fix); measured clean at ~5% peak now. */
+    /* ADCCONTROL2 (0x0A): input select. The K50835F SLIC handset transmit audio is
+     * wired to LIN2/RIN2 on this bench — PROVEN: speech captured (ACrms 196, crest 10.3)
+     * on 0x50, vs DC-only floating offset on 0x00 (LIN1). */
+    if (i2c_write_reg(ES8388_ADC_CTL2, 0x50) != ESP_OK) return ESP_FAIL; /* LIN2/RIN2 — SLIC handset mic */
+    if (i2c_write_reg(ES8388_ADC_CTL3, 0x02) != ESP_OK) return ESP_FAIL; /* DS filter sel */
+    if (i2c_write_reg(ES8388_ADC_CTL4, 0x0C) != ESP_OK) return ESP_FAIL; /* I2S 16-bit */
+    if (i2c_write_reg(ES8388_ADC_CTL5, 0x02) != ESP_OK) return ESP_FAIL; /* RATIO=256 */
+    if (i2c_write_reg(0x0E, 0x00) != ESP_OK) return ESP_FAIL; /* ADCCONTROL6: clear ADCSMUTE bit5 (reset default 0x30 = ADC output muted) */
+    if (i2c_write_reg(ES8388_ADC_CTL8, 0x00) != ESP_OK) return ESP_FAIL; /* ADC vol L 0dB */
+    if (i2c_write_reg(ES8388_ADC_CTL9, 0x00) != ESP_OK) return ESP_FAIL; /* ADC vol R 0dB */
+
+    /* 8. DAC I2S: 16-bit I2S Philips, MCLK/256, no softmute. */
+    if (i2c_write_reg(ES8388_DAC_CTL1, 0x18) != ESP_OK) return ESP_FAIL;
+    if (i2c_write_reg(ES8388_DAC_CTL2, 0x02) != ESP_OK) return ESP_FAIL; /* DACLRCKDIV=256 */
+    if (i2c_write_reg(ES8388_DAC_CTL3, 0x00) != ESP_OK) return ESP_FAIL; /* mute off */
+
+    /* 9. DAC volume: 0 dB on both channels. */
+    if (i2c_write_reg(ES8388_DAC_CTL4, ES8388_VOL_0DB) != ESP_OK) return ESP_FAIL;
+    if (i2c_write_reg(ES8388_DAC_CTL5, ES8388_VOL_0DB) != ESP_OK) return ESP_FAIL;
+
+    /* 10. Mixer: L→LOUT, R→ROUT (straight through, no cross-mix). */
+    if (i2c_write_reg(ES8388_DAC_CTL16, 0x1B) != ESP_OK) return ESP_FAIL;
+    if (i2c_write_reg(ES8388_DAC_CTL17, 0x90) != ESP_OK) return ESP_FAIL;
+    if (i2c_write_reg(ES8388_DAC_CTL20, 0x90) != ESP_OK) return ESP_FAIL;
+
+    /* 11. DACCONTROL21 (0x2B): ADC+DAC LRCK sync.
+     * Espressif reference es8388_start():
+     *   0xC0 = LINE mode (analog bypass, not ADC digital path)
+     *   0x80 = DAC+ADC digital record+playback mode (bit7 only)
+     * We want digital ADC recording + DAC playback → use 0x80. */
+    if (i2c_write_reg(ES8388_DAC_CTL21, 0x80) != ESP_OK) return ESP_FAIL;
+
+    /* 11b. Restart internal state machine (required after DACCONTROL21 change).
+     * Without this pulse, the ADC clock domain may not synchronise properly.
+     * Espressif reference sequence: CHIPPOWER=0xF0 then 0x00 to restart FSM. */
+    if (i2c_write_reg(ES8388_CHIP_POWER, 0xF0) != ESP_OK) return ESP_FAIL;
+    vTaskDelay(pdMS_TO_TICKS(5));
+    if (i2c_write_reg(ES8388_CHIP_POWER, 0x00) != ESP_OK) return ESP_FAIL;
+    vTaskDelay(pdMS_TO_TICKS(10));
+
+    /* DACCONTROL23 (0x2D): VROI=0, normal output. */
+    if (i2c_write_reg(ES8388_DAC_CTL23, 0x00) != ESP_OK) return ESP_FAIL;
+
+    /* 12. Volume: OUT1L/R = 0x1E (0dB), OUT2L = 0 (speaker vol set by PA separately). */
+    if (i2c_write_reg(ES8388_DAC_CTL24, 0x1E) != ESP_OK) return ESP_FAIL;  /* LOUT1VOL */
+    if (i2c_write_reg(ES8388_DAC_CTL25, 0x1E) != ESP_OK) return ESP_FAIL;  /* ROUT1VOL */
+    if (i2c_write_reg(ES8388_DAC_CTL26, 0x00) != ESP_OK) return ESP_FAIL;  /* LOUT2VOL */
+
+    /* 13. DAC power: power up DAC L+R. */
+    if (i2c_write_reg(ES8388_DAC_POWER, 0x3C) != ESP_OK) return ESP_FAIL;
+
+    /* 14. ADC power: full power-up (all Pdn bits cleared = 0x00).
+     * 0x09 is the intermediate state (Espressif es8388_open end-state), but
+     * the full ADC + analog input power-up requires 0x00 (Espressif es8388_start(ADC)).
+     * Must come AFTER DACCONTROL21 state machine restart and ADCCONTROL6 unmute. */
+    if (i2c_write_reg(ES8388_ADC_POWER, 0x00) != ESP_OK) return ESP_FAIL;
+    vTaskDelay(pdMS_TO_TICKS(10));  /* let ADC analog settle after full power-up */
+
+    /* 15. Enable PA (power amplifier for speaker). */
+    gpio_set_direction(PLIP_PA_ENABLE, GPIO_MODE_OUTPUT);
+    gpio_set_level(PLIP_PA_ENABLE, 1);
+
+    /* Verify key register values to confirm ADC path is configured correctly. */
+    uint8_t ctl1=0, adcpwr=0, adcinsel=0, dacctl21=0, chippower=0, adcctl3=0;
+    es8388_read_reg(ES8388_CHIP_CTL1,    &ctl1);
+    es8388_read_reg(ES8388_ADC_POWER,    &adcpwr);
+    es8388_read_reg(ES8388_ADC_CTL2,     &adcinsel);  /* ADCCONTROL2 = input sel (0x50=LIN2) */
+    es8388_read_reg(ES8388_ADC_CTL3,     &adcctl3);   /* ADCCONTROL3 = DS filter (0x02) */
+    es8388_read_reg(ES8388_DAC_CTL21,    &dacctl21);  /* DACCONTROL21 = ADC+DAC LRCK sync (0xC0) */
+    es8388_read_reg(ES8388_CHIP_POWER,   &chippower);
+    ESP_LOGI(TAG, "ES8388 regs: CTL1=0x%02X ADCPWR=0x%02X ADCINSEL=0x%02X ADCCTL3=0x%02X DACCTL21=0x%02X CHIPPOWER=0x%02X",
+             ctl1, adcpwr, adcinsel, adcctl3, dacctl21, chippower);
+    ESP_LOGI(TAG, "ES8388 init OK — PA enabled, DAC @ 0dB, ADC PGA +12dB, input=LIN2/RIN2 (SLIC handset), DACCTL21=0x80");
+    return ESP_OK;
+}
+
+esp_err_t es8388_set_volume(uint8_t vol)
+{
+    /* ES8388 OUTx volume registers are GAIN, not attenuation: 0x00 = -45 dB
+     * (min) .. 0x21 = 0 dB (max); higher value = louder (>0x21 = mute/reserved).
+     * Map 0..100 → 0x00..0x21. (The previous code inverted this, so vol=100
+     * produced 0x00 = quietest — confirmed at the bench.)
+     * DACCONTROL24 (0x2E)=OUT1L, 25 (0x2F)=OUT1R, 26 (0x30)=OUT2L, 27 (0x31)=OUT2R.
+     * DACCONTROL21 (0x2B) is the ADC/DAC LRCK sync register — DO NOT touch here. */
+    if (vol > 100) vol = 100;
+    uint8_t reg_val = (uint8_t)((int)vol * 0x21 / 100);
+    if (reg_val > 0x21) reg_val = 0x21;
+    ESP_LOGI(TAG, "set_volume: %d%% -> reg=0x%02X (0x21=max,0dB)", vol, reg_val);
+    esp_err_t r = ESP_OK;
+    r |= i2c_write_reg(ES8388_DAC_CTL24, reg_val);  /* OUT1 L volume */
+    r |= i2c_write_reg(ES8388_DAC_CTL25, reg_val);  /* OUT1 R volume */
+    r |= i2c_write_reg(ES8388_DAC_CTL26, reg_val);  /* OUT2 L volume */
+    r |= i2c_write_reg(0x31,             reg_val);  /* OUT2 R volume (DACCONTROL27) */
+    return r;
+}
+
+esp_err_t es8388_set_dac_volume(uint8_t atten)
+{
+    /* DACCONTROL4 (0x04) = LDACVOL, DACCONTROL5 (0x05) = RDACVOL: DIGITAL DAC
+     * volume, applied BEFORE the analog output stages. 0x00 = 0 dB, each step
+     * = -0.5 dB, up to 0xC0 = -96 dB (mute). Lowering this gives analog
+     * headroom while keeping the output-stage volume (es8388_set_volume) high. */
+    if (atten > 0xC0) atten = 0xC0;
+    ESP_LOGI(TAG, "set_dac_volume: atten=0x%02X (-%.1f dB)", atten, atten * 0.5f);
+    esp_err_t r = i2c_write_reg(ES8388_DAC_CTL4, atten);
+    r |= i2c_write_reg(ES8388_DAC_CTL5, atten);
+    return r;
+}
+
+esp_err_t es8388_mute(bool mute)
+{
+    uint8_t val = mute ? 0x04 : 0x00;  /* bit2 = DACMUTE */
+    ESP_LOGI(TAG, "mute: %s", mute ? "on" : "off");
+    return i2c_write_reg(ES8388_DAC_CTL3, val);
+}

plip_voice/main/es8388.h

@@ -0,0 +1,49 @@
+#pragma once
+/*
+ * es8388.h — Minimal ES8388 codec driver for AI-Thinker ESP32-A1S (IDF 5.x).
+ *
+ * Controls the ES8388 via I2C (addr 0x10) for playback (DAC) and capture (ADC).
+ * I2S is configured separately via the standard IDF driver/i2s_std.h API.
+ *
+ * Thread safety: all public functions are non-reentrant; call from a single task.
+ */
+
+#include <stdint.h>
+#include <stdbool.h>
+#include "esp_err.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Initialise the I2C master bus and configure the ES8388 registers for
+ * stereo playback + mic capture at 16 kHz.  Must be called once before any
+ * other es8388_* function and after driver_i2c is available.
+ *
+ * Returns ESP_OK on success.  On I2C error the register that failed is logged
+ * and ESP_FAIL is returned — the codec is in an undefined state.
+ */
+esp_err_t es8388_init(void);
+
+/* Set output volume.  vol: 0 (mute) … 100 (full).  Mapped to ES8388
+ * OUT1 (headphone) + OUT2 (speaker) attenuation registers. */
+esp_err_t es8388_set_volume(uint8_t vol);
+
+/* Set the DIGITAL DAC volume (DACCONTROL4/5), applied before the analog output
+ * stages. atten: 0 = 0 dB (full), each step -0.5 dB, 0xC0 = mute. Lowering it
+ * gives analog headroom while keeping es8388_set_volume() high. */
+esp_err_t es8388_set_dac_volume(uint8_t atten);
+
+/* Mute / unmute DAC output. */
+esp_err_t es8388_mute(bool mute);
+
+/* Read a single ES8388 register for diagnostic purposes. */
+esp_err_t es8388_read_reg(uint8_t reg, uint8_t *value);
+
+/* Write a single ES8388 register for diagnostic purposes (bench input/gain
+ * sweeps via /debug/es8388?reg=&val= — find which input the SLIC feeds). */
+esp_err_t es8388_write_reg(uint8_t reg, uint8_t value);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/main/hook_client.c

@@ -0,0 +1,159 @@
+/*
+ * hook_client.c — Ported from PLIP_FIRMWARE/src/zacus_hook_client.cpp.
+ *
+ * Uses esp_http_client (IDF) instead of Arduino HTTPClient.
+ * Failure policy: 3 s timeout, one retry at +250 ms, then drop.
+ */
+
+#include "hook_client.h"
+
+#include <string.h>
+#include <stdio.h>
+
+#include "esp_http_client.h"
+#include "esp_log.h"
+#include "esp_wifi.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/queue.h"
+#include "freertos/task.h"
+
+#define TAG "hook_client"
+
+#define HOOK_PATH         "/voice/hook"
+#define TIMEOUT_MS        3000
+#define RETRY_DELAY_MS    250
+#define QUEUE_DEPTH       4
+#define WIFI_WAIT_MS      1500
+#define WIFI_WAIT_STEP_MS 50
+
+typedef struct {
+    char state[8];   /* "off" | "on" */
+    char reason[32]; /* free-form */
+} hook_event_t;
+
+static QueueHandle_t s_queue   = NULL;
+static char          s_url[192] = {0};  /* full URL: <base>/voice/hook */
+
+/* ── Helpers ─────────────────────────────────────────────────────────────── */
+
+static bool wifi_ready_within(uint32_t budget_ms)
+{
+    uint32_t waited = 0;
+    wifi_ap_record_t ap;
+    while (waited < budget_ms) {
+        if (esp_wifi_sta_get_ap_info(&ap) == ESP_OK) return true;
+        vTaskDelay(pdMS_TO_TICKS(WIFI_WAIT_STEP_MS));
+        waited += WIFI_WAIT_STEP_MS;
+    }
+    return false;
+}
+
+static bool post_hook(const hook_event_t *ev)
+{
+    char body[96];
+    snprintf(body, sizeof(body),
+             "{\"state\":\"%s\",\"reason\":\"%s\"}",
+             ev->state, ev->reason);
+
+    esp_http_client_config_t cfg = {
+        .url             = s_url,
+        .method          = HTTP_METHOD_POST,
+        .timeout_ms      = TIMEOUT_MS,
+        .buffer_size     = 512,
+    };
+    esp_http_client_handle_t client = esp_http_client_init(&cfg);
+    if (!client) {
+        ESP_LOGW(TAG, "http_client_init failed");
+        return false;
+    }
+
+    esp_http_client_set_header(client, "Content-Type", "application/json");
+    esp_http_client_set_post_field(client, body, (int)strlen(body));
+
+    esp_err_t ret = esp_http_client_perform(client);
+    int code = 0;
+    if (ret == ESP_OK) {
+        code = esp_http_client_get_status_code(client);
+    }
+    esp_http_client_cleanup(client);
+
+    if (ret == ESP_OK && code >= 200 && code < 300) {
+        ESP_LOGI(TAG, "POST %s -> %d (state=%s reason=%s)",
+                 s_url, code, ev->state, ev->reason);
+        return true;
+    }
+    ESP_LOGW(TAG, "POST %s failed: ret=%s code=%d (state=%s reason=%s)",
+             s_url, esp_err_to_name(ret), code, ev->state, ev->reason);
+    return false;
+}
+
+/* ── Worker task ─────────────────────────────────────────────────────────── */
+
+static void worker_task(void *arg)
+{
+    (void)arg;
+    ESP_LOGI(TAG, "worker ready, target=%s", s_url);
+    hook_event_t ev;
+    for (;;) {
+        if (xQueueReceive(s_queue, &ev, portMAX_DELAY) != pdTRUE) continue;
+
+        if (!wifi_ready_within(WIFI_WAIT_MS)) {
+            ESP_LOGW(TAG, "WiFi down, dropping hook event (state=%s)", ev.state);
+            continue;
+        }
+
+        if (post_hook(&ev)) continue;
+
+        vTaskDelay(pdMS_TO_TICKS(RETRY_DELAY_MS));
+        if (!post_hook(&ev)) {
+            ESP_LOGW(TAG, "giving up after retry (state=%s)", ev.state);
+        }
+    }
+}
+
+/* ── Public API ──────────────────────────────────────────────────────────── */
+
+esp_err_t hook_client_init(const char *master_url)
+{
+    if (s_queue) return ESP_OK;   /* idempotent */
+
+    const char *base = (master_url && *master_url)
+                           ? master_url : CONFIG_PLIP_MASTER_URL;
+
+    size_t blen = strnlen(base, sizeof(s_url) - sizeof(HOOK_PATH) - 1);
+    if (blen == 0 || blen >= sizeof(s_url) - sizeof(HOOK_PATH)) {
+        ESP_LOGE(TAG, "invalid master_url");
+        return ESP_ERR_INVALID_ARG;
+    }
+    memcpy(s_url, base, blen);
+    if (s_url[blen - 1] == '/') blen--;
+    s_url[blen] = '\0';
+    strncat(s_url, HOOK_PATH, sizeof(s_url) - strlen(s_url) - 1);
+
+    s_queue = xQueueCreate(QUEUE_DEPTH, sizeof(hook_event_t));
+    if (!s_queue) return ESP_ERR_NO_MEM;
+
+    if (xTaskCreate(worker_task, "hook_client", 4096, NULL, 5, NULL) != pdPASS) {
+        vQueueDelete(s_queue);
+        s_queue = NULL;
+        return ESP_ERR_NO_MEM;
+    }
+    return ESP_OK;
+}
+
+bool hook_client_report(const char *state, const char *reason)
+{
+    if (!s_queue) {
+        ESP_LOGW(TAG, "report() before init() — dropping");
+        return false;
+    }
+    hook_event_t ev = {0};
+    strncpy(ev.state,  state  ? state  : "", sizeof(ev.state)  - 1);
+    strncpy(ev.reason, reason ? reason : "", sizeof(ev.reason) - 1);
+
+    if (xQueueSend(s_queue, &ev, 0) != pdTRUE) {
+        ESP_LOGW(TAG, "queue full, dropping (state=%s)", ev.state);
+        return false;
+    }
+    return true;
+}

plip_voice/main/hook_client.h

@@ -0,0 +1,28 @@
+#pragma once
+/*
+ * hook_client.h — POST hook transitions to the Zacus master (Phase D).
+ *
+ * Ported from PLIP_FIRMWARE/src/zacus_hook_client.cpp to plain C / IDF.
+ * Endpoint: POST <master_url>/voice/hook  { "state": "off"|"on", "reason": "..." }
+ */
+
+#include "esp_err.h"
+#include <stdbool.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Initialise (spawns worker task).  master_url: base URL e.g.
+ * "http://192.168.0.188".  Pass NULL to use CONFIG_PLIP_MASTER_URL.
+ * Idempotent — safe to call once from app_main. */
+esp_err_t hook_client_init(const char *master_url);
+
+/* Enqueue a state report (non-blocking).  state = "off"|"on",
+ * reason = short free-form tag ("pickup", "hangup", "boot", ...).
+ * Returns true if enqueued, false if queue was full. */
+bool hook_client_report(const char *state, const char *reason);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/main/idf_component.yml

@@ -0,0 +1,6 @@
+## IDF Component Manager manifest for plip_voice main component.
+## Managed components are fetched during `idf.py build` if not already present.
+dependencies:
+  ## cJSON for CMD/EVT JSON parsing
+  idf:
+    version: ">=5.1.0"

plip_voice/main/main.c

@@ -0,0 +1,162 @@
+/*
+ * main.c — PLIP voice annex application entry point.
+ *
+ * Board: AI-Thinker ESP32-A1S Audio Kit V2.2 (ESP32, ES8388, SD, mic, HP)
+ * Target: ESP-IDF 5.4, esp32 target (NOT esp32s3)
+ *
+ * Boot sequence:
+ *   1. NVS init
+ *   2. audio_init()  — ES8388 + I2S (Phase A)
+ *   3. audio_play_tone(440Hz, 1s)  — Phase A proof (sound before WiFi)
+ *   4. net_init()   — WiFi STA + HTTP server (Phase B)
+ *   5. scenario_mesh_init()  — ESP-NOW CMD receiver (Phase C)
+ *   6. hook_client_init()    — hook event forwarder (Phase D)
+ *   7. phone_init()          — off-hook GPIO monitor (Phase D)
+ */
+
+#include <stdio.h>
+#include <string.h>
+
+#include "esp_log.h"
+#include "esp_err.h"
+#include "esp_system.h"
+#include "nvs_flash.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+
+#include "board_config.h"
+#include "audio.h"
+#include "net.h"
+#include "cmd_exec.h"
+#include "hook_client.h"
+#include "phone.h"
+#include "slic.h"
+#include "scenario_mesh.h"
+#include "dialer.h"
+#include "conversation.h"
+
+static const char *TAG = "plip-main";
+
+/* ── ESP-NOW CMD/EVT text callback ─────────────────────────────────────────── */
+
+static void on_mesh_text(uint8_t kind, const uint8_t src_mac[6], const char *text)
+{
+    if (kind == SCENARIO_MESH_TEXT_CMD) {
+        ESP_LOGI(TAG, "ESP-NOW CMD from %02x:%02x:%02x:%02x:%02x:%02x: %.80s",
+                 src_mac[0], src_mac[1], src_mac[2],
+                 src_mac[3], src_mac[4], src_mac[5], text);
+        esp_err_t err = cmd_exec_handle(text, strlen(text));
+        if (err != ESP_OK && err != ESP_ERR_NOT_SUPPORTED) {
+            ESP_LOGW(TAG, "cmd_exec_handle: %s", esp_err_to_name(err));
+        }
+    } else {
+        ESP_LOGI(TAG, "ESP-NOW EVT from %02x:%02x:%02x:%02x:%02x:%02x: %.80s",
+                 src_mac[0], src_mac[1], src_mac[2],
+                 src_mac[3], src_mac[4], src_mac[5], text);
+    }
+}
+
+/* ── Boot task — runs all phases with sufficient stack ────────────────────── */
+
+static void boot_task(void *arg)
+{
+    (void)arg;
+
+    /* ── 1b. SLIC power-up (before audio — PD must be released early) ── */
+    esp_err_t ret = slic_init();
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "slic_init FAILED: %s — SLIC remains in power-down (no audio path, no hook sense)",
+                 esp_err_to_name(ret));
+    } else {
+        ESP_LOGI(TAG, "SLIC powered up: PD=GPIO%d OD-HIGH, SHK=GPIO%d active-HIGH, RM=GPIO%d, FR=GPIO%d",
+                 PLIP_SLIC_PD, PLIP_SLIC_SHK, PLIP_SLIC_RM, PLIP_SLIC_FR);
+    }
+
+    /* ── 2. Audio init (Phase A) ── */
+    ret = audio_init();
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "audio_init FAILED: %s", esp_err_to_name(ret));
+        ESP_LOGE(TAG, "PHASE A BLOCKED — check ES8388 I2C wiring (SDA=33, SCL=32)");
+        /* Continue anyway — WiFi + hook still work. */
+    } else {
+        /* ── 3. Phase A proof: 440 Hz test tone ── */
+        ESP_LOGI(TAG, "Phase A: playing 440 Hz test tone (1s) — LISTEN FOR SOUND");
+        audio_play_tone(440.0f, 1000);
+        vTaskDelay(pdMS_TO_TICKS(200));
+        ESP_LOGI(TAG, "Phase A: tone done. If you heard it, Phase A is VERIFIED.");
+    }
+
+    /* ── 4. Network + HTTP server (Phase B) ── */
+    ret = net_init();
+    if (ret == ESP_OK) {
+        ESP_LOGI(TAG, "Phase B: WiFi + httpd up. Verify: curl http://<PLIP_IP>/status");
+    } else if (ret == ESP_ERR_TIMEOUT) {
+        ESP_LOGW(TAG, "Phase B: WiFi timeout — running offline (ESP-NOW still active)");
+    } else {
+        ESP_LOGW(TAG, "Phase B: net_init error %s", esp_err_to_name(ret));
+    }
+
+    /* ── 5. ESP-NOW CMD receiver + scenario mesh (Phase C) ── */
+    /* scenario_mesh_init() requires WiFi to be started (done in net_init). */
+    ret = scenario_mesh_init(NULL);   /* receive CMD only — no scenario apply */
+    if (ret != ESP_OK) {
+        ESP_LOGW(TAG, "scenario_mesh_init: %s — ESP-NOW unavailable",
+                 esp_err_to_name(ret));
+    } else {
+        ret = scenario_mesh_set_text_cb(on_mesh_text);
+        if (ret == ESP_OK) {
+            ESP_LOGI(TAG, "Phase C: ESP-NOW CMD receiver up");
+        }
+    }
+
+    /* ── 6. Hook client (Phase D) ── */
+    ret = hook_client_init(NULL);  /* uses CONFIG_PLIP_MASTER_URL */
+    if (ret != ESP_OK) {
+        ESP_LOGW(TAG, "hook_client_init: %s", esp_err_to_name(ret));
+    }
+
+    /* ── Stage 1: dialer + conversation state machine ── */
+    dialer_init();
+    conversation_init();
+    ESP_LOGI(TAG, "Stage 1: dialer + conversation state machine up");
+
+    /* ── 7. Phone task (Phase D) ── */
+    ret = phone_init();
+    if (ret != ESP_OK) {
+        ESP_LOGW(TAG, "phone_init: %s", esp_err_to_name(ret));
+    } else {
+        ESP_LOGI(TAG, "Phase D: off-hook GPIO monitor up (GPIO=%d)",
+                 CONFIG_PLIP_HOOK_GPIO);
+    }
+
+    ESP_LOGI(TAG, "Boot complete. Free heap: %lu bytes", esp_get_free_heap_size());
+    ESP_LOGI(TAG, "Press BOOT button (GPIO%d) to simulate hook pickup.",
+             CONFIG_PLIP_HOOK_GPIO);
+    vTaskDelete(NULL);
+}
+
+/* ── Application entry point ─────────────────────────────────────────────── */
+
+void app_main(void)
+{
+    ESP_LOGI(TAG, "============================================");
+    ESP_LOGI(TAG, "  Zacus PLIP Voice Annex");
+    ESP_LOGI(TAG, "  Board: AI-Thinker ESP32-A1S (ES8388)");
+    ESP_LOGI(TAG, "  IDF:   %s", esp_get_idf_version());
+    ESP_LOGI(TAG, "  Free heap: %lu bytes", esp_get_free_heap_size());
+    ESP_LOGI(TAG, "============================================");
+
+    /* ── 1. NVS ── */
+    esp_err_t ret = nvs_flash_init();
+    if (ret == ESP_ERR_NVS_NO_FREE_PAGES ||
+        ret == ESP_ERR_NVS_NEW_VERSION_FOUND) {
+        ESP_ERROR_CHECK(nvs_flash_erase());
+        ret = nvs_flash_init();
+    }
+    ESP_ERROR_CHECK(ret);
+
+    /* Delegate all subsequent phases to a task with adequate stack.
+     * app_main's default stack (8192) is too tight for I2S + WiFi init
+     * when called from the main task (IDF internal stack usage adds up). */
+    xTaskCreatePinnedToCore(boot_task, "boot", 12288, NULL, 5, NULL, 0);
+}

plip_voice/main/net.c

@@ -0,0 +1,981 @@
+/*
+ * net.c — WiFi STA + minimal HTTP server for the PLIP voice annex.
+ *
+ * WiFi credentials: read from NVS namespace "wifi" (keys "ssid"/"pwd")
+ * first; fall back to Kconfig compile-time defaults if NVS is empty.
+ * This mirrors the pattern used in the idf_zacus master (main.c:load_wifi_creds).
+ *
+ * HTTP endpoints (Phase B):
+ *   GET  /status              → { "ip":"...", "playing":false, "off_hook":false }
+ *   POST /game/scenario       → store Runtime 3 IR to SPIFFS (same as box3_voice)
+ *   POST /game/file?path=...  → write an arbitrary file to SPIFFS (Phase B proof)
+ *
+ * The httpd handle is exposed via net_httpd_handle() so other modules
+ * (hook_client, cmd_exec) can register additional routes.
+ */
+
+#include "net.h"
+#include "dialer.h"
+
+#include <inttypes.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <errno.h>
+#include <sys/stat.h>
+
+#include "audio.h"
+#include "tones.h"
+#include "cmd_exec.h"
+#include "phone.h"
+#include "conversation.h"
+#include "es8388.h"
+#include "slic.h"
+#include "board_config.h"
+#include "driver/gpio.h"
+#include "esp_err.h"
+#include "esp_event.h"
+#include "esp_http_server.h"
+#include "esp_log.h"
+#include "esp_netif.h"
+#include "esp_spiffs.h"
+#include "esp_timer.h"
+#include "esp_wifi.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "nvs.h"
+#include "nvs_flash.h"
+#include "cJSON.h"
+
+#define TAG "net"
+
+#define WIFI_CONNECT_TIMEOUT_MS 30000
+#define WIFI_POLL_MS            250
+
+#define SPIFFS_LABEL  "storage"
+#define SPIFFS_BASE   "/spiffs"
+#define MAX_BODY      (3 * 1024 * 1024) /* streamed in 2 KB chunks → RAM-safe;
+                                          raised to fit full-length neural-TTS
+                                          (Kyutai) gamebook narration WAVs */
+
+static volatile bool    s_connected  = false;
+static httpd_handle_t   s_httpd      = NULL;
+static bool             s_spiffs_ok  = false;
+
+/* ── NVS WiFi cred loader (mirrors idf_zacus main.c) ─────────────────────── */
+
+static esp_err_t load_wifi_creds(char *ssid, size_t ssid_len,
+                                  char *pwd,  size_t pwd_len)
+{
+    nvs_handle_t h;
+    esp_err_t err = nvs_open("wifi", NVS_READONLY, &h);
+    if (err != ESP_OK) return err;
+
+    size_t sl = ssid_len, pl = pwd_len;
+    err  = nvs_get_str(h, "ssid", ssid, &sl);
+    err |= nvs_get_str(h, "pwd",  pwd,  &pl);
+    nvs_close(h);
+
+    if (err != ESP_OK || ssid[0] == '\0') return ESP_ERR_NOT_FOUND;
+    return ESP_OK;
+}
+
+/* ── WiFi event handler ───────────────────────────────────────────────────── */
+
+static void wifi_event_handler(void *arg, esp_event_base_t base,
+                               int32_t event_id, void *data)
+{
+    if (base == WIFI_EVENT && event_id == WIFI_EVENT_STA_START) {
+        esp_wifi_connect();
+    } else if (base == WIFI_EVENT && event_id == WIFI_EVENT_STA_DISCONNECTED) {
+        /* Retry on EVERY disconnect, including failures during the initial boot
+         * attempt: on a mesh the first association frequently fails (wrong node
+         * picked / channel roam) and must be retried, otherwise the boot connect
+         * just times out. Previously the initial loop gave up on first failure. */
+        s_connected = false;
+        ESP_LOGW(TAG, "WiFi disconnected/assoc failed, reconnecting...");
+        vTaskDelay(pdMS_TO_TICKS(1500));
+        esp_wifi_connect();
+    } else if (base == IP_EVENT && event_id == IP_EVENT_STA_GOT_IP) {
+        ip_event_got_ip_t *ev = (ip_event_got_ip_t *)data;
+        ESP_LOGI(TAG, "WiFi connected — IP: " IPSTR, IP2STR(&ev->ip_info.ip));
+        s_connected = true;
+    }
+}
+
+/* ── SPIFFS lazy mount ────────────────────────────────────────────────────── */
+
+static void ensure_spiffs(void)
+{
+    if (s_spiffs_ok) return;
+    esp_vfs_spiffs_conf_t conf = {
+        .base_path              = SPIFFS_BASE,
+        .partition_label        = SPIFFS_LABEL,
+        .max_files              = 8,
+        .format_if_mount_failed = true,
+    };
+    esp_err_t ret = esp_vfs_spiffs_register(&conf);
+    if (ret == ESP_OK || ret == ESP_ERR_INVALID_STATE) {
+        s_spiffs_ok = true;
+        ESP_LOGI(TAG, "SPIFFS mounted at %s", SPIFFS_BASE);
+    } else {
+        ESP_LOGE(TAG, "SPIFFS mount failed: %s", esp_err_to_name(ret));
+    }
+}
+
+/* ── HTTP helpers ─────────────────────────────────────────────────────────── */
+
+static esp_err_t send_json(httpd_req_t *req, const char *status,
+                            const char *body)
+{
+    httpd_resp_set_status(req, status);
+    httpd_resp_set_type(req, "application/json");
+    httpd_resp_set_hdr(req, "Access-Control-Allow-Origin", "*");
+    return httpd_resp_sendstr(req, body);
+}
+
+/* ── GET /status ──────────────────────────────────────────────────────────── */
+
+static esp_err_t handle_status(httpd_req_t *req)
+{
+    char buf[128];
+    esp_netif_ip_info_t ip_info = {0};
+    esp_netif_t *sta = esp_netif_get_handle_from_ifkey("WIFI_STA_DEF");
+    if (sta) esp_netif_get_ip_info(sta, &ip_info);
+
+    /* Report REAL state (this used to hardcode playing/off_hook to false, which
+     * was badly misleading during bring-up). off_hook = firmware debounced view
+     * (phone_is_offhook); shk = raw SLIC line, the single source of truth that
+     * gates audio playback — sound routes to the handset only when shk is true. */
+    snprintf(buf, sizeof(buf),
+             "{\"board\":\"plip\",\"ip\":\"" IPSTR "\",\"playing\":%s,"
+             "\"off_hook\":%s,\"shk\":%s}",
+             IP2STR(&ip_info.ip),
+             audio_is_playing() ? "true" : "false",
+             phone_is_offhook() ? "true" : "false",
+             slic_is_offhook() ? "true" : "false");
+    return send_json(req, "200 OK", buf);
+}
+
+/* ── POST /game/scenario ──────────────────────────────────────────────────── */
+
+static esp_err_t handle_scenario_post(httpd_req_t *req)
+{
+    if (req->content_len <= 0 || req->content_len > MAX_BODY)
+        return send_json(req, "413 Payload Too Large", "{\"error\":\"too large\"}");
+
+    ensure_spiffs();
+
+    char *body = malloc((size_t)req->content_len + 1);
+    if (!body) return send_json(req, "500 Internal Server Error", "{\"error\":\"oom\"}");
+
+    int total = 0;
+    while (total < (int)req->content_len) {
+        int got = httpd_req_recv(req, body + total, req->content_len - total);
+        if (got <= 0) { free(body); return ESP_FAIL; }
+        total += got;
+    }
+    body[total] = '\0';
+
+    const char *path = SPIFFS_BASE "/scenario.json";
+    FILE *f = fopen(path, "wb");
+    if (!f) {
+        free(body);
+        return send_json(req, "500 Internal Server Error", "{\"error\":\"write failed\"}");
+    }
+    fwrite(body, 1, (size_t)total, f);
+    fclose(f);
+    free(body);
+
+    char resp[128];
+    snprintf(resp, sizeof(resp),
+             "{\"status\":\"ok\",\"board\":\"plip\",\"bytes\":%d}", total);
+    ESP_LOGI(TAG, "scenario stored: %d bytes", total);
+    return send_json(req, "200 OK", resp);
+}
+
+/* ── POST /game/file?path=...  (Phase B proof: write file to SPIFFS) ─────── */
+
+/* Streaming chunk size for /game/file — keeps heap usage under 4 KB. */
+#define FILE_RECV_CHUNK 2048
+
+static esp_err_t handle_file_post(httpd_req_t *req)
+{
+    /* Extract ?path= query parameter */
+    char query[256] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    char fpath[128] = {0};
+    if (httpd_query_key_value(query, "path", fpath, sizeof(fpath)) != ESP_OK) {
+        return send_json(req, "400 Bad Request", "{\"error\":\"missing path param\"}");
+    }
+
+    if (req->content_len <= 0 || req->content_len > MAX_BODY)
+        return send_json(req, "413 Payload Too Large", "{\"error\":\"too large\"}");
+
+    /* Route to the microSD when the path is sd-prefixed ("sd/…", "sdcard/…"
+     * or "/sdcard/…"); otherwise stage in SPIFFS (default). The SD is mounted
+     * on demand here so a voice pack can be staged without an off-hook play. */
+    const char *base;
+    const char *rel = fpath;
+    bool to_sd = false;
+    if      (strncmp(fpath, "sd/", 3) == 0)       { rel = fpath + 3; to_sd = true; }
+    else if (strncmp(fpath, "sdcard/", 7) == 0)   { rel = fpath + 7; to_sd = true; }
+    else if (strncmp(fpath, "/sdcard/", 8) == 0)  { rel = fpath + 8; to_sd = true; }
+
+    if (to_sd) {
+        if (!audio_ensure_sd())
+            return send_json(req, "503 Service Unavailable", "{\"error\":\"no sd card\"}");
+        base = PLIP_SD_MOUNT;   /* "/sdcard" */
+    } else {
+        ensure_spiffs();
+        base = SPIFFS_BASE;
+    }
+    while (*rel == '/') rel++;   /* avoid a double slash when joining */
+
+    char full[192];
+    snprintf(full, sizeof(full), "%s/%s", base, rel);
+
+    /* Create the immediate parent dir (one level) so a pack can live in a
+     * subfolder like /sdcard/voice/<file>.wav — fopen won't create it. FAT
+     * supports mkdir; SPIFFS is flat so we only do this for the SD path. */
+    if (to_sd) {
+        char *slash = strrchr(full, '/');
+        if (slash && slash != full) {
+            *slash = '\0';
+            mkdir(full, 0775);   /* best-effort; ignore EEXIST */
+            *slash = '/';
+        }
+    }
+
+    FILE *f = fopen(full, "wb");
+    if (!f) {
+        ESP_LOGW(TAG, "fopen %s failed: errno=%d", full, errno);
+        return send_json(req, "500 Internal Server Error", "{\"error\":\"write failed\"}");
+    }
+
+    /* Stream body to file using a small stack buffer — no large malloc needed. */
+    char chunk[FILE_RECV_CHUNK];
+    int remaining = (int)req->content_len;
+    size_t written = 0;
+    while (remaining > 0) {
+        int to_recv = (remaining < FILE_RECV_CHUNK) ? remaining : FILE_RECV_CHUNK;
+        int got = httpd_req_recv(req, chunk, to_recv);
+        if (got <= 0) {
+            fclose(f);
+            return ESP_FAIL;
+        }
+        size_t fw = fwrite(chunk, 1, (size_t)got, f);
+        if ((int)fw != got) {
+            ESP_LOGW(TAG, "fwrite short: %zu/%d (disk full?)", fw, got);
+            fclose(f);
+            return send_json(req, "500 Internal Server Error", "{\"error\":\"write failed\"}");
+        }
+        written += fw;
+        remaining -= got;
+    }
+    fclose(f);
+
+    char resp[256];
+    snprintf(resp, sizeof(resp),
+             "{\"status\":\"ok\",\"path\":\"%s\",\"bytes\":%zu}", full, written);
+    ESP_LOGI(TAG, "file stored: %s (%zu bytes)", full, written);
+    return send_json(req, "200 OK", resp);
+}
+
+/* ── DELETE /game/file?path=...  (remove a staged file; SD or SPIFFS) ─────── */
+/* Same path routing as the POST handler (sd/ prefix → /sdcard, else /spiffs).
+ * Lets the host clean stale clips when regenerating the SD voice pack. */
+static esp_err_t handle_file_delete(httpd_req_t *req)
+{
+    char query[128] = {0}, fpath[128] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    if (httpd_query_key_value(query, "path", fpath, sizeof(fpath)) != ESP_OK)
+        return send_json(req, "400 Bad Request", "{\"error\":\"missing path param\"}");
+
+    const char *base, *rel = fpath;
+    bool to_sd = false;
+    if      (strncmp(fpath, "sd/", 3) == 0)       { rel = fpath + 3; to_sd = true; }
+    else if (strncmp(fpath, "sdcard/", 7) == 0)   { rel = fpath + 7; to_sd = true; }
+    else if (strncmp(fpath, "/sdcard/", 8) == 0)  { rel = fpath + 8; to_sd = true; }
+
+    if (to_sd) {
+        if (!audio_ensure_sd())
+            return send_json(req, "503 Service Unavailable", "{\"error\":\"no sd card\"}");
+        base = PLIP_SD_MOUNT;
+    } else {
+        ensure_spiffs();
+        base = SPIFFS_BASE;
+    }
+    while (*rel == '/') rel++;
+
+    char full[192];
+    snprintf(full, sizeof(full), "%s/%s", base, rel);
+    if (remove(full) != 0) {
+        ESP_LOGW(TAG, "remove %s failed: errno=%d", full, errno);
+        return send_json(req, "404 Not Found", "{\"error\":\"remove failed\"}");
+    }
+    ESP_LOGI(TAG, "file removed: %s", full);
+    char resp[256];
+    snprintf(resp, sizeof(resp), "{\"status\":\"ok\",\"removed\":\"%s\"}", full);
+    return send_json(req, "200 OK", resp);
+}
+
+/* ── POST /voice/capture ──────────────────────────────────────────────────── */
+
+#define CAP_MAX_MS_DEFAULT      5000
+#define CAP_SILENCE_MS_DEFAULT   800
+/* Max capture duration: 10 s (no large buffer needed — streaming approach). */
+#define CAP_MAX_ALLOWED_MS  10000
+
+/*
+ * Streaming WAV capture handler.
+ *
+ * Memory budget: only two small buffers on stack:
+ *   - WAV header : 44 bytes
+ *   - mono frame : 320 × 2 = 640 bytes
+ * The audio driver uses an additional 1280-byte scratch buf for the stereo I2S read.
+ * Total heap usage during capture: ~2 KB (vs 128+ KB for buffer-at-once approach).
+ *
+ * The response uses chunked transfer encoding via httpd_resp_send_chunk().
+ * We send the WAV header first (with data_size = 0xFFFFFFFF as placeholder since
+ * length is unknown), then PCM frames, then finalize with httpd_resp_send_chunk(NULL,0).
+ * Most WAV players / ffmpeg handle the placeholder length gracefully.
+ */
+static esp_err_t handle_voice_capture(httpd_req_t *req)
+{
+    /* Parse optional query params: max_ms, silence_ms */
+    char query[128] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+
+    char val[16] = {0};
+    int max_ms     = CAP_MAX_MS_DEFAULT;
+    int silence_ms = CAP_SILENCE_MS_DEFAULT;
+    if (httpd_query_key_value(query, "max_ms", val, sizeof(val)) == ESP_OK) {
+        int v = atoi(val);
+        if (v > 0 && v <= CAP_MAX_ALLOWED_MS) max_ms = v;
+        else if (v > CAP_MAX_ALLOWED_MS) max_ms = CAP_MAX_ALLOWED_MS;
+    }
+    memset(val, 0, sizeof(val));
+    if (httpd_query_key_value(query, "silence_ms", val, sizeof(val)) == ESP_OK) {
+        int v = atoi(val);
+        if (v > 0 && v <= 5000) silence_ms = v;
+    }
+    ESP_LOGI(TAG, "POST /voice/capture max_ms=%d silence_ms=%d (streaming)", max_ms, silence_ms);
+
+    /* Start capture (enables I2S RX, disables TX). */
+    if (audio_capture_begin(max_ms, silence_ms) != 0) {
+        httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "capture init failed");
+        return ESP_FAIL;
+    }
+
+    /* Send HTTP headers. */
+    httpd_resp_set_type(req, "audio/wav");
+    httpd_resp_set_hdr(req, "Access-Control-Allow-Origin", "*");
+
+    /* WAV header with placeholder data_size (0xFFFFFFFF = unknown length). */
+    const uint32_t cap_rate    = 16000;
+    const uint16_t cap_ch      = 1;
+    const uint16_t cap_bits    = 16;
+    const uint32_t byte_rate   = cap_rate * cap_ch * (cap_bits / 8);
+    const uint16_t block_align = (uint16_t)(cap_ch * (cap_bits / 8));
+    const uint32_t data_size   = 0xFFFFFFFFu;  /* placeholder, updated below if possible */
+    const uint32_t riff_size   = 0xFFFFFFFFu;
+    uint16_t fmt_pcm  = 1;
+    uint32_t fmt_size = 16;
+
+    uint8_t hdr[44];
+    memcpy(hdr +  0, "RIFF", 4);  memcpy(hdr +  4, &riff_size,   4);
+    memcpy(hdr +  8, "WAVE", 4);  memcpy(hdr + 12, "fmt ", 4);
+    memcpy(hdr + 16, &fmt_size,   4);  memcpy(hdr + 20, &fmt_pcm,     2);
+    memcpy(hdr + 22, &cap_ch,     2);  memcpy(hdr + 24, &cap_rate,    4);
+    memcpy(hdr + 28, &byte_rate,  4);  memcpy(hdr + 32, &block_align, 2);
+    memcpy(hdr + 34, &cap_bits,   2);
+    memcpy(hdr + 36, "data", 4);  memcpy(hdr + 40, &data_size,   4);
+
+    esp_err_t ret = httpd_resp_send_chunk(req, (const char *)hdr, sizeof(hdr));
+    if (ret != ESP_OK) { audio_capture_end(); return ESP_FAIL; }
+
+    /* VAD state (mirrors audio_capture_wav logic). */
+    const int64_t vad_onset_sq   = (int64_t)328 * 328;
+    const int64_t vad_silence_sq  = (int64_t)100 * 100;
+    const int     max_frames      = max_ms / 20;
+    const int     silence_frames  = silence_ms / 20;
+    bool          voice_started   = false;
+    int           silent_count    = 0;
+    int           total_frames    = 0;
+    size_t        total_pcm_bytes = 0;
+
+    /* Per-frame mono buffer: 320 × int16_t = 640 bytes on stack. */
+    int16_t mono_frame[320];
+
+    for (int f = 0; f < max_frames; f++) {
+        int64_t rms_sq = 0;
+        int n = audio_capture_read_frame(mono_frame, 320, &rms_sq);
+        if (n < 0) { ESP_LOGW(TAG, "capture: read error at frame %d", f); break; }
+        if (n == 0) continue;  /* timeout, retry */
+
+        /* Send PCM chunk. */
+        size_t frame_bytes = (size_t)n * sizeof(int16_t);
+        ret = httpd_resp_send_chunk(req, (const char *)mono_frame, (ssize_t)frame_bytes);
+        if (ret != ESP_OK) { ESP_LOGW(TAG, "capture: send_chunk failed"); break; }
+        total_pcm_bytes += frame_bytes;
+        total_frames = f + 1;
+
+        /* VAD. */
+        if (!voice_started) {
+            if (rms_sq >= vad_onset_sq) {
+                voice_started = true;
+                silent_count  = 0;
+                ESP_LOGI(TAG, "capture: voice onset frame %d rms²=%"PRId64, f, rms_sq);
+            }
+        } else {
+            if (rms_sq < vad_silence_sq) {
+                if (++silent_count >= silence_frames) {
+                    ESP_LOGI(TAG, "capture: VAD end frame %d", f);
+                    break;
+                }
+            } else {
+                silent_count = 0;
+            }
+        }
+    }
+
+    audio_capture_end();
+
+    /* Finalize chunked transfer. */
+    httpd_resp_send_chunk(req, NULL, 0);
+
+    float dur = (float)total_pcm_bytes / (float)(cap_rate * (cap_bits / 8) * cap_ch);
+    ESP_LOGI(TAG, "capture done: %d frames %.2fs %zu PCM bytes sent", total_frames, dur, total_pcm_bytes);
+    return ESP_OK;
+}
+
+/* ── GET /debug/regs  (ES8388 register dump for ADC diagnosis) ─────────────── */
+
+static esp_err_t handle_debug_regs(httpd_req_t *req)
+{
+    /* Read key ES8388 registers to diagnose ADC path. */
+    uint8_t regs[0x40] = {0};
+    for (int i = 0; i < 0x40; i++) {
+        es8388_read_reg((uint8_t)i, &regs[i]);
+    }
+    char buf[1024];
+    /* All register names use Espressif official numbering (ADCCONTROL1=reg0x09, etc.) */
+    int n = snprintf(buf, sizeof(buf),
+        "{\"CTL1\":\"0x%02X\",\"CTL2\":\"0x%02X\",\"CHIPPOWER\":\"0x%02X\","
+        "\"ADCPOWER\":\"0x%02X\",\"DACPOWER\":\"0x%02X\","
+        "\"ADCCONTROL1\":\"0x%02X\",\"ADCCONTROL2\":\"0x%02X\",\"ADCCONTROL3\":\"0x%02X\","
+        "\"ADCCONTROL4\":\"0x%02X\",\"ADCCONTROL5\":\"0x%02X\",\"ADCCONTROL6\":\"0x%02X\","
+        "\"ADCCONTROL7\":\"0x%02X\",\"ADCCONTROL8\":\"0x%02X\",\"ADCCONTROL9\":\"0x%02X\","
+        "\"DACCTL1\":\"0x%02X\",\"DACCTL2\":\"0x%02X\",\"DACCTL3\":\"0x%02X\","
+        "\"DACCTL21\":\"0x%02X\",\"DACCTL24\":\"0x%02X\",\"DACCTL25\":\"0x%02X\","
+        "\"MASTERMODE\":\"0x%02X\",\"CHIPCTL3\":\"0x%02X\",\"CHIPLP\":\"0x%02X\"}",
+        regs[0x00], regs[0x01], regs[0x02],
+        regs[0x03], regs[0x04],
+        regs[0x09], regs[0x0A], regs[0x0B],
+        regs[0x0C], regs[0x0D], regs[0x0E],
+        regs[0x0F], regs[0x10], regs[0x11],
+        regs[0x17], regs[0x18], regs[0x19],
+        regs[0x2B], regs[0x2E], regs[0x2F],
+        regs[0x08], regs[0x06], regs[0x05]);
+    httpd_resp_set_type(req, "application/json");
+    httpd_resp_set_hdr(req, "Access-Control-Allow-Origin", "*");
+    return httpd_resp_send(req, buf, n);
+}
+
+/* ── GET /debug/es8388?reg=0x0A&val=0x00  (poke a codec register) ──────────────
+ * Bench tool to find which ADC input the SLIC handset mic feeds: sweep the
+ * input-select (ADCCONTROL2 = reg 0x0A) and PGA gain (ADCCONTROL1 = reg 0x09)
+ * live, then re-capture, without reflashing per experiment. reg/val accept
+ * decimal or 0x-hex. Returns the written value read back. */
+static esp_err_t handle_debug_es8388(httpd_req_t *req)
+{
+    char query[64] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    char rs[12] = {0}, vs[12] = {0};
+    if (httpd_query_key_value(query, "reg", rs, sizeof(rs)) != ESP_OK)
+        return send_json(req, "400 Bad Request", "{\"error\":\"missing reg param\"}");
+    uint8_t reg = (uint8_t)strtol(rs, NULL, 0);
+    char resp[96];
+    if (httpd_query_key_value(query, "val", vs, sizeof(vs)) == ESP_OK) {
+        uint8_t val = (uint8_t)strtol(vs, NULL, 0);
+        esp_err_t err = es8388_write_reg(reg, val);
+        uint8_t rb = 0;
+        es8388_read_reg(reg, &rb);
+        ESP_LOGI(TAG, "debug/es8388: reg 0x%02X <- 0x%02X (readback 0x%02X, %s)",
+                 reg, val, rb, esp_err_to_name(err));
+        snprintf(resp, sizeof(resp),
+                 "{\"reg\":\"0x%02X\",\"wrote\":\"0x%02X\",\"readback\":\"0x%02X\",\"ok\":%s}",
+                 reg, val, rb, err == ESP_OK ? "true" : "false");
+    } else {
+        uint8_t rb = 0;
+        es8388_read_reg(reg, &rb);
+        snprintf(resp, sizeof(resp), "{\"reg\":\"0x%02X\",\"value\":\"0x%02X\"}", reg, rb);
+    }
+    return send_json(req, "200 OK", resp);
+}
+
+/* ── POST /game/cmd  (WiFi-direct CMD endpoint) ───────────────────────────── */
+
+#define MAX_CMD_BYTES 512
+
+static esp_err_t handle_cmd_post(httpd_req_t *req)
+{
+    if (req->content_len <= 0 || req->content_len > MAX_CMD_BYTES)
+        return send_json(req, "400 Bad Request", "{\"error\":\"body must be 1..512 bytes\"}");
+
+    char body[MAX_CMD_BYTES + 1];
+    int total = 0;
+    while (total < (int)req->content_len) {
+        int got = httpd_req_recv(req, body + total, req->content_len - total);
+        if (got <= 0) {
+            if (got == HTTPD_SOCK_ERR_TIMEOUT) continue;
+            return send_json(req, "400 Bad Request", "{\"error\":\"recv failed\"}");
+        }
+        total += got;
+    }
+    body[total] = '\0';
+
+    ESP_LOGI(TAG, "POST /game/cmd (%d B): %.*s", total, total < 80 ? total : 80, body);
+    esp_err_t err = cmd_exec_handle(body, (size_t)total);
+    if (err == ESP_ERR_INVALID_ARG)
+        return send_json(req, "400 Bad Request", "{\"error\":\"malformed cmd json or missing op\"}");
+    return send_json(req, "200 OK", "{\"ok\":true}");
+}
+
+/* ── GET /debug/ring  (trigger ring tone over HTTP) ──────────────────────── */
+
+static esp_err_t handle_debug_ring(httpd_req_t *req)
+{
+    /* /debug/ring[?number=NN] — arm an INCOMING call from persona NN (default
+     * 17) and ring. On pickup the conversation jumps straight to the greeting. */
+    char query[96] = {0}, number[16] = {0}, scene[40] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    httpd_query_key_value(query, "number", number, sizeof(number));
+    httpd_query_key_value(query, "scene", scene, sizeof(scene));
+    /* scene (optional) lets the phone play that scene's hint from its SD pack
+     * after the greeting — fully offline, no model. */
+    conversation_arm_incoming(number, scene[0] ? scene : NULL);
+    char resp[64];
+    snprintf(resp, sizeof(resp), "{\"ok\":true,\"ringing\":true,\"incoming\":\"%s\"}",
+             number[0] ? number : "17");
+    ESP_LOGI(TAG, "debug/ring: incoming armed (num=%s), ringing", number[0] ? number : "17");
+    return send_json(req, "200 OK", resp);
+}
+
+/* ── GET /debug/ringstop  (stop ring tone over HTTP) ─────────────────────── */
+
+static esp_err_t handle_debug_ringstop(httpd_req_t *req)
+{
+    phone_ring_stop();
+    ESP_LOGI(TAG, "debug/ringstop: ring stopped");
+    return send_json(req, "200 OK", "{\"ok\":true,\"ringing\":false}");
+}
+
+/* ── GET /debug/slic  (read GPIO levels of all 4 SLIC pins) ──────────────── */
+
+static esp_err_t handle_debug_slic(httpd_req_t *req)
+{
+    char buf[192];
+    /* gpio_get_level reads the actual pad level even for output pins. */
+    snprintf(buf, sizeof(buf),
+             "{\"rm\":%d,\"fr\":%d,\"shk\":%d,\"pd\":%d,\"offhook\":%s,\"ringing\":%s}",
+             gpio_get_level(PLIP_SLIC_RM), gpio_get_level(PLIP_SLIC_FR),
+             gpio_get_level(PLIP_SLIC_SHK), gpio_get_level(PLIP_SLIC_PD),
+             slic_is_offhook() ? "true" : "false",
+             slic_is_ringing() ? "true" : "false");
+    ESP_LOGI(TAG, "debug/slic: %s", buf);
+    return send_json(req, "200 OK", buf);
+}
+
+/* ── GET /debug/hookmon  (fast-sample the hook GPIO, log transitions) ─────────
+ * Samples CONFIG_PLIP_HOOK_GPIO every 2 ms for ~6 s and records every level
+ * transition with a timestamp.  Catches both slow hook toggles AND fast rotary
+ * pulses (~60-100 ms) that the 0.7 s HTTP poll of /debug/slic cannot see.
+ * The user performs the physical action (lift/hang/dial) during the window.    */
+
+#define HOOKMON_SAMPLE_MS   2
+#define HOOKMON_WINDOW_MS   6000
+#define HOOKMON_MAX_EDGES   64
+
+static esp_err_t handle_debug_hookmon(httpd_req_t *req)
+{
+    const int gpio = CONFIG_PLIP_HOOK_GPIO;
+    int       last = gpio_get_level(gpio);
+    const int initial = last;
+
+    int64_t  t0 = esp_timer_get_time();
+    int      edges_t[HOOKMON_MAX_EDGES];
+    int      edges_l[HOOKMON_MAX_EDGES];
+    int      n_edges = 0;
+    int      lo = last, hi = last;   /* track min/max seen */
+
+    ESP_LOGI(TAG, "hookmon: start, gpio=%d initial=%d (sample %dms / window %dms)",
+             gpio, initial, HOOKMON_SAMPLE_MS, HOOKMON_WINDOW_MS);
+
+    for (;;) {
+        int64_t now = esp_timer_get_time();
+        int     dt  = (int)((now - t0) / 1000);
+        if (dt >= HOOKMON_WINDOW_MS) break;
+
+        int lvl = gpio_get_level(gpio);
+        if (lvl < lo) lo = lvl;
+        if (lvl > hi) hi = lvl;
+        if (lvl != last) {
+            if (n_edges < HOOKMON_MAX_EDGES) {
+                edges_t[n_edges] = dt;
+                edges_l[n_edges] = lvl;
+                n_edges++;
+            }
+            last = lvl;
+        }
+        vTaskDelay(pdMS_TO_TICKS(HOOKMON_SAMPLE_MS));
+    }
+
+    /* Build JSON: { gpio, initial, final, lo, hi, edges:[{t,l},...], count } */
+    char buf[1024];
+    int  off = 0;
+    off += snprintf(buf + off, sizeof(buf) - off,
+                    "{\"gpio\":%d,\"initial\":%d,\"final\":%d,\"lo\":%d,\"hi\":%d,"
+                    "\"count\":%d,\"edges\":[",
+                    gpio, initial, last, lo, hi, n_edges);
+    for (int i = 0; i < n_edges && off < (int)sizeof(buf) - 32; i++) {
+        off += snprintf(buf + off, sizeof(buf) - off, "%s{\"t\":%d,\"l\":%d}",
+                        i ? "," : "", edges_t[i], edges_l[i]);
+    }
+    off += snprintf(buf + off, sizeof(buf) - off, "]}");
+
+    ESP_LOGI(TAG, "hookmon: done, %d edges, lo=%d hi=%d", n_edges, lo, hi);
+    return send_json(req, "200 OK", buf);
+}
+
+/* ── GET /debug/dacvol?a=N  (set ES8388 DIGITAL DAC volume, live tuning) ─────── */
+
+static esp_err_t handle_debug_dacvol(httpd_req_t *req)
+{
+    char query[32] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    char astr[8] = {0};
+    if (httpd_query_key_value(query, "a", astr, sizeof(astr)) != ESP_OK) {
+        return send_json(req, "400 Bad Request", "{\"error\":\"missing a param (atten steps, 0=0dB)\"}");
+    }
+    int a = atoi(astr);
+    if (a < 0) a = 0;
+    if (a > 192) a = 192;
+    esp_err_t r = es8388_set_dac_volume((uint8_t)a);
+    char resp[80];
+    snprintf(resp, sizeof(resp), "{\"ok\":%s,\"atten_steps\":%d,\"db\":-%.1f}",
+             (r == ESP_OK) ? "true" : "false", a, a * 0.5);
+    ESP_LOGI(TAG, "debug/dacvol: atten=%d (-%.1f dB)", a, a * 0.5);
+    return send_json(req, "200 OK", resp);
+}
+
+/* ── GET /debug/miccap?ms=4000  (TEMP DIAG: raw fixed-duration mic capture) ───
+ * No VAD, DC-blocked, PA on, tones stopped. Lets us measure the true handset
+ * mic level while the caller speaks continuously, independent of VAD timing. */
+static esp_err_t handle_debug_miccap(httpd_req_t *req)
+{
+    char query[64] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    char val[16] = {0};
+    int ms = 4000;
+    if (httpd_query_key_value(query, "ms", val, sizeof(val)) == ESP_OK) {
+        int v = atoi(val);
+        if (v > 0 && v <= 8000) ms = v;
+    }
+    tones_stop();
+    audio_pa_set(true);
+    if (audio_capture_begin(ms, ms) != 0) {
+        httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "capture init failed");
+        return ESP_FAIL;
+    }
+    httpd_resp_set_type(req, "audio/wav");
+    const uint32_t cap_rate = 16000; const uint16_t cap_ch = 1, cap_bits = 16;
+    const uint32_t byte_rate = cap_rate * cap_ch * (cap_bits / 8);
+    const uint16_t block_align = (uint16_t)(cap_ch * (cap_bits / 8));
+    const uint32_t big = 0xFFFFFFFFu; uint16_t fmt_pcm = 1; uint32_t fmt_size = 16;
+    uint8_t hdr[44];
+    memcpy(hdr, "RIFF", 4);  memcpy(hdr + 4, &big, 4);
+    memcpy(hdr + 8, "WAVE", 4);  memcpy(hdr + 12, "fmt ", 4);
+    memcpy(hdr + 16, &fmt_size, 4);  memcpy(hdr + 20, &fmt_pcm, 2);
+    memcpy(hdr + 22, &cap_ch, 2);  memcpy(hdr + 24, &cap_rate, 4);
+    memcpy(hdr + 28, &byte_rate, 4);  memcpy(hdr + 32, &block_align, 2);
+    memcpy(hdr + 34, &cap_bits, 2);  memcpy(hdr + 36, "data", 4);  memcpy(hdr + 40, &big, 4);
+    httpd_resp_send_chunk(req, (const char *)hdr, sizeof(hdr));
+
+    const int max_frames = ms / 20;
+    int16_t mono[320];
+    float dc_x1 = 0.0f, dc_y1 = 0.0f; const float dc_R = 0.97f;
+    for (int f = 0; f < max_frames; f++) {
+        int64_t rms_sq = 0;
+        int n = audio_capture_read_frame(mono, 320, &rms_sq);
+        if (n < 0) break;
+        if (n == 0) continue;
+        for (int i = 0; i < n; i++) {
+            float xf = (float)mono[i];
+            float yf = xf - dc_x1 + dc_R * dc_y1;
+            dc_x1 = xf; dc_y1 = yf;
+            if (yf > 32767.0f) yf = 32767.0f; else if (yf < -32768.0f) yf = -32768.0f;
+            mono[i] = (int16_t)yf;
+        }
+        if (httpd_resp_send_chunk(req, (const char *)mono, (ssize_t)(n * sizeof(int16_t))) != ESP_OK) break;
+    }
+    audio_capture_end();
+    httpd_resp_send_chunk(req, NULL, 0);
+    return ESP_OK;
+}
+
+/* ── GET /debug/getfile?path=/x.wav  (read a SPIFFS file back for diagnosis) ── */
+
+static esp_err_t handle_debug_getfile(httpd_req_t *req)
+{
+    char query[160] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    char fpath[96] = {0};
+    if (httpd_query_key_value(query, "path", fpath, sizeof(fpath)) != ESP_OK) {
+        return send_json(req, "400 Bad Request", "{\"error\":\"missing path param\"}");
+    }
+    ensure_spiffs();
+    char full[160];
+    if (fpath[0] == '/') snprintf(full, sizeof(full), "%s%s", SPIFFS_BASE, fpath);
+    else                 snprintf(full, sizeof(full), "%s/%s", SPIFFS_BASE, fpath);
+
+    FILE *f = fopen(full, "rb");
+    if (!f) return send_json(req, "404 Not Found", "{\"error\":\"not found\"}");
+
+    httpd_resp_set_type(req, "application/octet-stream");
+    char chunk[1024];
+    size_t r;
+    while ((r = fread(chunk, 1, sizeof(chunk), f)) > 0) {
+        if (httpd_resp_send_chunk(req, chunk, r) != ESP_OK) { fclose(f); return ESP_FAIL; }
+    }
+    fclose(f);
+    httpd_resp_send_chunk(req, NULL, 0);  /* terminate */
+    return ESP_OK;
+}
+
+/* ── GET /debug/vol?v=N  (set ES8388 output volume 0..100, live tuning) ─────── */
+
+static esp_err_t handle_debug_vol(httpd_req_t *req)
+{
+    char query[32] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    char vstr[8] = {0};
+    if (httpd_query_key_value(query, "v", vstr, sizeof(vstr)) != ESP_OK) {
+        return send_json(req, "400 Bad Request", "{\"error\":\"missing v param (0..100)\"}");
+    }
+    int v = atoi(vstr);
+    if (v < 0) v = 0;
+    if (v > 100) v = 100;
+    esp_err_t r = es8388_set_volume((uint8_t)v);
+    char resp[64];
+    snprintf(resp, sizeof(resp), "{\"ok\":%s,\"volume\":%d}", (r == ESP_OK) ? "true" : "false", v);
+    ESP_LOGI(TAG, "debug/vol: set volume=%d (%s)", v, (r == ESP_OK) ? "ok" : "err");
+    return send_json(req, "200 OK", resp);
+}
+
+/* ── GET /debug/dial?number=NNNN  (push digits into the dialer) ───────────── */
+
+static esp_err_t handle_debug_dial(httpd_req_t *req)
+{
+    char query[64] = {0};
+    httpd_req_get_url_query_str(req, query, sizeof(query));
+    char number[16] = {0};
+    if (httpd_query_key_value(query, "number", number, sizeof(number)) != ESP_OK) {
+        return send_json(req, "400 Bad Request", "{\"error\":\"missing number param\"}");
+    }
+
+    int pushed = 0;
+    for (int i = 0; number[i] != '\0' && i < 12; i++) {
+        if (number[i] >= '0' && number[i] <= '9') {
+            dialer_push_digit(number[i] - '0');
+            pushed++;
+        }
+    }
+
+    char resp[64];
+    snprintf(resp, sizeof(resp), "{\"ok\":true,\"pushed\":%d,\"number\":\"%s\"}",
+             pushed, dialer_current());
+    ESP_LOGI(TAG, "debug/dial: pushed %d digits -> \"%s\"", pushed, dialer_current());
+    return send_json(req, "200 OK", resp);
+}
+
+/* ── Public API ───────────────────────────────────────────────────────────── */
+
+bool net_is_connected(void)
+{
+    return s_connected;
+}
+
+httpd_handle_t net_httpd_handle(void)
+{
+    return s_httpd;
+}
+
+esp_err_t net_init(void)
+{
+    /* WiFi credentials: NVS first, then Kconfig compile-time defaults. */
+    char ssid[64] = CONFIG_PLIP_WIFI_SSID;
+    char pwd[64]  = CONFIG_PLIP_WIFI_PASSWORD;
+
+    char nvs_ssid[64] = {0}, nvs_pwd[64] = {0};
+    if (load_wifi_creds(nvs_ssid, sizeof(nvs_ssid),
+                        nvs_pwd,  sizeof(nvs_pwd)) == ESP_OK) {
+        strncpy(ssid, nvs_ssid, sizeof(ssid) - 1);
+        strncpy(pwd,  nvs_pwd,  sizeof(pwd)  - 1);
+        ESP_LOGI(TAG, "WiFi creds from NVS: SSID=\"%s\"", ssid);
+    } else {
+        ESP_LOGI(TAG, "WiFi creds from Kconfig: SSID=\"%s\"", ssid);
+    }
+
+    /* Network stack init. */
+    ESP_ERROR_CHECK(esp_netif_init());
+    ESP_ERROR_CHECK(esp_event_loop_create_default());
+    esp_netif_create_default_wifi_sta();
+
+    wifi_init_config_t cfg = WIFI_INIT_CONFIG_DEFAULT();
+    ESP_ERROR_CHECK(esp_wifi_init(&cfg));
+
+    ESP_ERROR_CHECK(esp_event_handler_instance_register(
+        WIFI_EVENT, ESP_EVENT_ANY_ID, wifi_event_handler, NULL, NULL));
+    ESP_ERROR_CHECK(esp_event_handler_instance_register(
+        IP_EVENT, IP_EVENT_STA_GOT_IP, wifi_event_handler, NULL, NULL));
+
+    wifi_config_t wcfg = {
+        .sta = {
+            .threshold.authmode = (strlen(pwd) == 0)
+                                      ? WIFI_AUTH_OPEN : WIFI_AUTH_WPA2_PSK,
+            /* Mesh-friendly association. A fixed .channel breaks on mesh APs
+             * (Freebox/Orange "Les cils") that roam channels via band-steering
+             * and DFS: if the node isn't on that channel at boot, association
+             * silently fails. So scan ALL channels (channel=0) and connect to
+             * the STRONGEST node broadcasting our SSID. The PLIP talks HTTP to
+             * the gateway, not ESP-NOW, so there is no co-channel constraint. */
+            .channel = 0,
+            .scan_method = WIFI_ALL_CHANNEL_SCAN,
+            .sort_method = WIFI_CONNECT_AP_BY_SIGNAL,
+            .failure_retry_cnt = 5,
+        },
+    };
+    strncpy((char *)wcfg.sta.ssid,     ssid, sizeof(wcfg.sta.ssid) - 1);
+    strncpy((char *)wcfg.sta.password,  pwd,  sizeof(wcfg.sta.password) - 1);
+
+    ESP_ERROR_CHECK(esp_wifi_set_mode(WIFI_MODE_STA));
+    ESP_ERROR_CHECK(esp_wifi_set_config(WIFI_IF_STA, &wcfg));
+    ESP_ERROR_CHECK(esp_wifi_start());
+
+    ESP_LOGI(TAG, "WiFi STA started, connecting to '%s' (ch=%d)...",
+             ssid, CONFIG_PLIP_WIFI_CHANNEL);
+
+    /* Wait for IP with timeout. */
+    const uint32_t deadline = xTaskGetTickCount() +
+                              pdMS_TO_TICKS(WIFI_CONNECT_TIMEOUT_MS);
+    while (!s_connected && xTaskGetTickCount() < deadline) {
+        vTaskDelay(pdMS_TO_TICKS(WIFI_POLL_MS));
+    }
+    if (!s_connected) {
+        ESP_LOGW(TAG, "WiFi connect timeout — continuing offline");
+        return ESP_ERR_TIMEOUT;
+    }
+
+    /* Start HTTP server. */
+    httpd_config_t hcfg = HTTPD_DEFAULT_CONFIG();
+    hcfg.server_port      = 80;
+    hcfg.max_uri_handlers = 18;
+    hcfg.stack_size       = 8192;
+
+    esp_err_t ret = httpd_start(&s_httpd, &hcfg);
+    if (ret != ESP_OK) {
+        ESP_LOGE(TAG, "httpd_start: %s", esp_err_to_name(ret));
+        return ret;
+    }
+
+    static const httpd_uri_t uri_status = {
+        .uri = "/status", .method = HTTP_GET,
+        .handler = handle_status, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_scenario = {
+        .uri = "/game/scenario", .method = HTTP_POST,
+        .handler = handle_scenario_post, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_file = {
+        .uri = "/game/file", .method = HTTP_POST,
+        .handler = handle_file_post, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_file_delete = {
+        .uri = "/game/file", .method = HTTP_DELETE,
+        .handler = handle_file_delete, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_cmd = {
+        .uri = "/game/cmd", .method = HTTP_POST,
+        .handler = handle_cmd_post, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_capture = {
+        .uri = "/voice/capture", .method = HTTP_POST,
+        .handler = handle_voice_capture, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_regs = {
+        .uri = "/debug/regs", .method = HTTP_GET,
+        .handler = handle_debug_regs, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_dial = {
+        .uri = "/debug/dial", .method = HTTP_GET,
+        .handler = handle_debug_dial, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_vol = {
+        .uri = "/debug/vol", .method = HTTP_GET,
+        .handler = handle_debug_vol, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_getfile = {
+        .uri = "/debug/getfile", .method = HTTP_GET,
+        .handler = handle_debug_getfile, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_miccap = {
+        .uri = "/debug/miccap", .method = HTTP_GET,
+        .handler = handle_debug_miccap, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_dacvol = {
+        .uri = "/debug/dacvol", .method = HTTP_GET,
+        .handler = handle_debug_dacvol, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_ring = {
+        .uri = "/debug/ring", .method = HTTP_GET,
+        .handler = handle_debug_ring, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_ringstop = {
+        .uri = "/debug/ringstop", .method = HTTP_GET,
+        .handler = handle_debug_ringstop, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_slic = {
+        .uri = "/debug/slic", .method = HTTP_GET,
+        .handler = handle_debug_slic, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_hookmon = {
+        .uri = "/debug/hookmon", .method = HTTP_GET,
+        .handler = handle_debug_hookmon, .user_ctx = NULL,
+    };
+    static const httpd_uri_t uri_debug_es8388 = {
+        .uri = "/debug/es8388", .method = HTTP_GET,
+        .handler = handle_debug_es8388, .user_ctx = NULL,
+    };
+    httpd_register_uri_handler(s_httpd, &uri_status);
+    httpd_register_uri_handler(s_httpd, &uri_scenario);
+    httpd_register_uri_handler(s_httpd, &uri_file);
+    httpd_register_uri_handler(s_httpd, &uri_file_delete);
+    httpd_register_uri_handler(s_httpd, &uri_cmd);
+    httpd_register_uri_handler(s_httpd, &uri_capture);
+    httpd_register_uri_handler(s_httpd, &uri_debug_regs);
+    httpd_register_uri_handler(s_httpd, &uri_debug_dial);
+    httpd_register_uri_handler(s_httpd, &uri_debug_vol);
+    httpd_register_uri_handler(s_httpd, &uri_debug_getfile);
+    httpd_register_uri_handler(s_httpd, &uri_debug_miccap);
+    httpd_register_uri_handler(s_httpd, &uri_debug_dacvol);
+    httpd_register_uri_handler(s_httpd, &uri_debug_ring);
+    httpd_register_uri_handler(s_httpd, &uri_debug_ringstop);
+    httpd_register_uri_handler(s_httpd, &uri_debug_slic);
+    httpd_register_uri_handler(s_httpd, &uri_debug_hookmon);
+    httpd_register_uri_handler(s_httpd, &uri_debug_es8388);
+
+    ESP_LOGI(TAG, "httpd up on :80 (GET /status, GET /debug/regs, GET /debug/dial, GET /debug/ring, GET /debug/ringstop, GET /debug/slic, POST /game/scenario, POST /game/file, POST /game/cmd, POST /voice/capture)");
+    return ESP_OK;
+}

plip_voice/main/net.h

@@ -0,0 +1,26 @@
+#pragma once
+/* net.h — WiFi STA + HTTP server for PLIP voice annex. */
+#include "esp_err.h"
+#include "esp_http_server.h"
+#include <stdbool.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Start WiFi STA (creds from NVS or Kconfig fallback) + httpd on port 80.
+ * Blocks until connected or 30 s elapses.
+ * Returns ESP_OK on success; ESP_ERR_TIMEOUT if WiFi failed (board continues
+ * offline — httpd is NOT started in that case). */
+esp_err_t net_init(void);
+
+/* Returns true once the WiFi STA has obtained an IP address. */
+bool net_is_connected(void);
+
+/* Expose the httpd handle so other modules can register additional URI
+ * handlers after net_init() succeeds. Returns NULL if not yet started. */
+httpd_handle_t net_httpd_handle(void);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/main/phone.c

@@ -0,0 +1,355 @@
+/*
+ * phone.c — Off-hook GPIO debounce + rotary pulse decoding + ring control.
+ *
+ * ISR sets a flag; the task reads the GPIO level after a short debounce
+ * and reports the transition via hook_client_report() and
+ * conversation_on_hook_change().
+ *
+ * Rotary pulse decoding (CONFIG_PLIP_DIAL_PULSE):
+ *   While off-hook, a rotary dial produces brief open/close pulses on the
+ *   SHK line (~60-100 ms open per pulse).  A real hangup holds the line
+ *   open for >500 ms.  The task polls the GPIO at 5 ms intervals to catch
+ *   pulses that would be missed by the 30 ms debounce.
+ *
+ *   Pulse train end: if the line stays closed for > PLIP_DIAL_PULSE_MAX_GAP_MS
+ *   after at least one pulse, the count is emitted as a digit (10 pulses = 0).
+ */
+
+#include "phone.h"
+#include "audio.h"
+#include "hook_client.h"
+#include "conversation.h"
+#include "slic.h"
+
+#if CONFIG_PLIP_DIAL_PULSE
+#include "dialer.h"
+#endif
+
+#include "driver/gpio.h"
+#include "esp_log.h"
+#include "esp_timer.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+
+#define TAG "phone"
+
+#define DEBOUNCE_MS          30
+#define HANGUP_THRESHOLD_MS  2500  /* open > this → real hangup, not a pulse/flicker.
+                                    * HOOK DEBOUNCE: the A1S cradle contact is
+                                    * marginal and flickers open for up to ~2 s
+                                    * mid-call; treating those as a hangup dropped
+                                    * the live conversation. Requiring the line to
+                                    * stay open ≥ 2.5 s before hanging up rides
+                                    * through the flickers. A real hangup holds the
+                                    * line open indefinitely so it still fires
+                                    * (~2.5 s later). Rotary pulses (~60 ms) and a
+                                    * closed line are unaffected. */
+#define TASK_STACK           4096
+#define TASK_PRIO            5
+#define RESYNC_PICKUP_MS     600   /* off-hook (pickup) self-correct: fast, so an
+                                    * incoming call answers promptly. */
+#define RESYNC_HANGUP_MS     HANGUP_THRESHOLD_MS  /* on-hook (hangup) self-correct:
+                                    * same long debounce as the prolonged-open path
+                                    * so a flickering contact never drops the call. */
+
+/* Rotary pulse hardening (CONFIG_PLIP_DIAL_PULSE) */
+#define PULSE_MIN_WIDTH_MS   20    /* ignore opens shorter than this (glitch filter) */
+                                   /* real rotary pulses are ≥ 40 ms open            */
+#define HANGUP_VERIFY_COUNT  5     /* re-read GPIO this many times before declaring  */
+                                   /* hangup (guards against marginal cradle contact) */
+#define HANGUP_VERIFY_STEP_MS 20   /* delay between verification reads (ms);         */
+                                   /* total span = COUNT × STEP = 100 ms             */
+
+static volatile bool s_edge_pending = false;
+static volatile bool s_ringing      = false;
+static volatile bool s_offhook      = false;  /* true = handset picked up */
+
+/* IRAM_ATTR: ISR must live in IRAM on original ESP32. */
+static void IRAM_ATTR on_hook_isr(void *arg)
+{
+    (void)arg;
+    s_edge_pending = true;
+}
+
+void phone_ring_start(void)
+{
+    if (s_ringing) return;
+    s_ringing = true;
+    ESP_LOGI(TAG, "ring start: SLIC RM/FR + audio tone");
+    /* Physical bell via SLIC RM/FR (main ringer) */
+    slic_ring_start();
+    /* Optional in-earpiece audio tone (audible feedback when handset is up) */
+    audio_ring_start();
+}
+
+void phone_ring_stop(void)
+{
+    if (!s_ringing) return;
+    s_ringing = false;
+    ESP_LOGI(TAG, "ring stop: SLIC RM/FR off + audio stop");
+    slic_ring_stop();
+    audio_stop();
+}
+
+static void report_offhook(bool offhook)
+{
+    conversation_on_hook_change(offhook);
+    hook_client_report(offhook ? "off" : "on", offhook ? "pickup" : "hangup");
+}
+
+/*
+ * Hook polarity:
+ *   CONFIG_PLIP_HOOK_ACTIVE_HIGH=y  (default, SLIC SHK GPIO23): HIGH = off-hook
+ *   CONFIG_PLIP_HOOK_ACTIVE_HIGH=n  (legacy BOOT button GPIO4): LOW  = off-hook
+ *
+ * Rotary pulse notes (SLIC SHK, active-HIGH):
+ *   Off-hook base: SHK HIGH.  A rotary pulse briefly opens the loop → SHK drops LOW
+ *   for ~60-100 ms, then returns HIGH.  Hangup: SHK stays LOW for >500 ms.
+ *   So "open" (pulse event) = level drops LOW when active-HIGH polarity.
+ */
+#if CONFIG_PLIP_HOOK_ACTIVE_HIGH
+#define HOOK_OFFHOOK_LEVEL  1   /* HIGH = off-hook */
+#define HOOK_PULSE_OPEN     0   /* LOW  = rotary pulse open (loop broken) */
+#define HOOK_PULSE_CLOSED   1   /* HIGH = rotary pulse closed (loop restored) */
+#else
+#define HOOK_OFFHOOK_LEVEL  0   /* LOW  = off-hook (legacy pull-up + BOOT button) */
+#define HOOK_PULSE_OPEN     1   /* HIGH = rotary pulse open */
+#define HOOK_PULSE_CLOSED   0   /* LOW  = rotary pulse closed */
+#endif
+
+static void phone_task(void *arg)
+{
+    (void)arg;
+    const int hook_gpio = CONFIG_PLIP_HOOK_GPIO;
+
+    gpio_config_t io_conf = {
+        .pin_bit_mask = (1ULL << hook_gpio),
+        .mode         = GPIO_MODE_INPUT,
+        .pull_up_en   = GPIO_PULLUP_ENABLE,
+        .pull_down_en = GPIO_PULLDOWN_DISABLE,
+        .intr_type    = GPIO_INTR_ANYEDGE,
+    };
+    ESP_ERROR_CHECK(gpio_config(&io_conf));
+    ESP_ERROR_CHECK(gpio_install_isr_service(0));
+    ESP_ERROR_CHECK(gpio_isr_handler_add(hook_gpio, on_hook_isr, NULL));
+
+    /* Read and report initial level so master state machine is in sync. */
+    int last_level = gpio_get_level(hook_gpio);
+    s_offhook = (last_level == HOOK_OFFHOOK_LEVEL);
+    int resync_ms = 0;   /* accumulates while raw level stably disagrees with s_offhook */
+    ESP_LOGI(TAG, "phone task ready, hook GPIO=%d level=%d active_%s (%s)",
+             hook_gpio, last_level,
+             (HOOK_OFFHOOK_LEVEL == 1) ? "high" : "low",
+             s_offhook ? "off-hook" : "on-hook");
+
+    if (!s_offhook) {
+        ESP_LOGI(TAG, "boot: on-hook -> PA off (mute)");
+        audio_pa_set(false);
+    } else {
+        ESP_LOGI(TAG, "boot: off-hook -> PA on (unmute)");
+        audio_pa_set(true);
+    }
+
+    hook_client_report(s_offhook ? "off" : "on", "boot");
+    conversation_on_hook_change(s_offhook);
+
+#if CONFIG_PLIP_DIAL_PULSE
+    /* Rotary pulse state.
+     * With SLIC SHK (active-HIGH): off-hook = HIGH, pulse = brief LOW dip. */
+    int     pulse_count   = 0;
+    bool    in_pulse      = false;   /* true while SHK is in pulse-open state */
+    int64_t pulse_open_us = 0;       /* timestamp when pulse open started */
+    int64_t last_close_us = 0;       /* timestamp when pulse closed (SHK back to OFFHOOK) */
+#endif
+
+    for (;;) {
+        if (s_edge_pending) {
+            s_edge_pending = false;
+
+#if CONFIG_PLIP_DIAL_PULSE
+            /* While off-hook, use fast polling instead of 30ms debounce
+             * so rotary pulses (~60-100ms) are not missed. */
+            if (s_offhook) {
+                /* The ISR fired — re-read immediately to catch the edge. */
+                vTaskDelay(pdMS_TO_TICKS(5));  /* minimal settle */
+                int level = gpio_get_level(hook_gpio);
+
+                if (level == HOOK_PULSE_OPEN && last_level == HOOK_PULSE_CLOSED) {
+                    /* Pulse open: loop broken (SHK dropped from off-hook level) */
+                    last_level    = HOOK_PULSE_OPEN;
+                    in_pulse      = true;
+                    pulse_open_us = esp_timer_get_time();
+                    ESP_LOGD(TAG, "pulse: open");
+                } else if (level == HOOK_PULSE_CLOSED && last_level == HOOK_PULSE_OPEN) {
+                    /* Pulse closed: loop restored */
+                    last_level    = HOOK_PULSE_CLOSED;
+                    last_close_us = esp_timer_get_time();
+                    int64_t open_dur_ms = (last_close_us - pulse_open_us) / 1000;
+
+                    if (in_pulse && open_dur_ms < PULSE_MIN_WIDTH_MS) {
+                        /* Glitch filter: open was too brief to be a real pulse.
+                         * Ignore — do not count as pulse and do not declare hangup. */
+                        in_pulse = false;
+                        ESP_LOGD(TAG, "glitch ignored (%"PRId64"ms < %dms)",
+                                 open_dur_ms, PULSE_MIN_WIDTH_MS);
+                    } else if (in_pulse && open_dur_ms < HANGUP_THRESHOLD_MS) {
+                        /* Valid rotary pulse duration: count it */
+                        pulse_count++;
+                        in_pulse = false;
+                        ESP_LOGD(TAG, "pulse %d (open %"PRId64"ms)", pulse_count, open_dur_ms);
+                    } else if (open_dur_ms >= HANGUP_THRESHOLD_MS) {
+                        /* Was open too long: treat as hangup */
+                        ESP_LOGI(TAG, "on-hook (hangup) after %"PRId64"ms open", open_dur_ms);
+                        pulse_count   = 0;
+                        in_pulse      = false;
+                        s_offhook     = false;
+                        audio_stop();
+                        audio_pa_set(false);
+                        report_offhook(false);
+                    }
+                }
+                /* Don't fall through to the standard debounce below */
+                goto poll_sleep;
+            }
+#endif
+            /* Standard 30 ms debounce for on-hook/off-hook transitions */
+            vTaskDelay(pdMS_TO_TICKS(DEBOUNCE_MS));
+            int level = gpio_get_level(hook_gpio);
+            if (level != last_level) {
+                last_level = level;
+                if (level == HOOK_OFFHOOK_LEVEL) {
+                    /* Off-hook: handset picked up. */
+                    s_offhook = true;
+#if CONFIG_PLIP_DIAL_PULSE
+                    pulse_count   = 0;
+                    in_pulse      = false;
+                    last_close_us = esp_timer_get_time();
+#endif
+                    ESP_LOGI(TAG, "off-hook (pickup) detected — PA on, SLIC offhook=%d",
+                             slic_is_offhook());
+                    audio_pa_set(true);
+                    phone_ring_stop();
+                    report_offhook(true);
+                } else {
+                    /* On-hook: handset hung up. */
+                    s_offhook = false;
+                    ESP_LOGI(TAG, "on-hook (hangup) detected — PA off, stopping audio");
+                    audio_stop();
+                    audio_pa_set(false);
+                    report_offhook(false);
+                }
+            }
+        }
+
+#if CONFIG_PLIP_DIAL_PULSE
+        /* Check for inter-digit gap: if we accumulated pulses and the line
+         * has been at closed (off-hook) state for > MAX_GAP_MS, emit the digit. */
+        if (s_offhook && pulse_count > 0 && !in_pulse && last_close_us > 0) {
+            int64_t gap_ms = (esp_timer_get_time() - last_close_us) / 1000;
+            if (gap_ms > CONFIG_PLIP_DIAL_PULSE_MAX_GAP_MS) {
+                if (pulse_count > 10) {
+                    ESP_LOGW(TAG, "bad pulse count %d, ignored", pulse_count);
+                    pulse_count = 0;
+                } else {
+                    int digit = (pulse_count == 10) ? 0 : pulse_count;
+                    ESP_LOGI(TAG, "rotary digit: %d pulses -> %d", pulse_count, digit);
+                    dialer_push_digit(digit);
+                    pulse_count = 0;
+                }
+            }
+        }
+
+        /* Also detect prolonged open (hangup) even if no more edges arrive.
+         * Active-HIGH: hangup = level stays at HOOK_PULSE_OPEN (LOW) > 500 ms.
+         *
+         * Hardening: re-sample the GPIO HANGUP_VERIFY_COUNT times with
+         * HANGUP_VERIFY_STEP_MS gaps.  Only declare hangup if every sample
+         * confirms the open state — guards against a marginal cradle contact
+         * that briefly dips below HANGUP_THRESHOLD_MS and then bounces back. */
+        if (s_offhook && in_pulse && pulse_open_us > 0) {
+            int64_t open_ms = (esp_timer_get_time() - pulse_open_us) / 1000;
+            if (open_ms >= HANGUP_THRESHOLD_MS) {
+                /* Multi-sample verification */
+                bool confirmed = true;
+                for (int v = 0; v < HANGUP_VERIFY_COUNT; v++) {
+                    vTaskDelay(pdMS_TO_TICKS(HANGUP_VERIFY_STEP_MS));
+                    if (gpio_get_level(hook_gpio) != HOOK_PULSE_OPEN) {
+                        /* Line recovered during verification — not a real hangup */
+                        confirmed = false;
+                        ESP_LOGD(TAG, "hangup verify failed at sample %d — bounce", v);
+                        /* Treat the level as if the line just closed */
+                        last_level    = HOOK_PULSE_CLOSED;
+                        last_close_us = esp_timer_get_time();
+                        /* The open was shorter than HANGUP_THRESHOLD_MS in effect,
+                         * but we already passed the threshold — do not count as
+                         * a pulse (it's the same ambiguous situation we reject
+                         * elsewhere).  Simply clear in_pulse. */
+                        in_pulse = false;
+                        break;
+                    }
+                }
+                if (confirmed) {
+                    ESP_LOGI(TAG, "on-hook (prolonged open %"PRId64"ms, verified ×%d) detected",
+                             open_ms, HANGUP_VERIFY_COUNT);
+                    last_level  = HOOK_PULSE_OPEN;
+                    pulse_count = 0;
+                    in_pulse    = false;
+                    s_offhook   = false;
+                    audio_stop();
+                    audio_pa_set(false);
+                    report_offhook(false);
+                }
+            }
+        }
+
+poll_sleep:
+#endif
+        /* ── Self-healing resync ──────────────────────────────────────────────
+         * The marginal A1S cradle contact occasionally makes the edge detection
+         * miss/flap a transition, leaving s_offhook stuck (e.g. firmware thinks
+         * on-hook while the handset is physically off-hook → the call never
+         * starts). If the raw level stably disagrees with s_offhook for
+         * RESYNC_STABLE_MS — and we are NOT mid rotary-pulse — trust the
+         * physical level and correct s_offhook, firing the transition. */
+        {
+            bool pulse_active = false;
+#if CONFIG_PLIP_DIAL_PULSE
+            pulse_active = in_pulse;
+#endif
+            bool phys_off = (gpio_get_level(hook_gpio) == HOOK_OFFHOOK_LEVEL);
+            if (!pulse_active && phys_off != s_offhook) {
+                resync_ms += 10;
+                /* Hook debounce: a PICKUP (→off-hook) is confirmed fast so calls
+                 * answer promptly; a HANGUP (→on-hook) needs the long debounce so
+                 * a flickering cradle contact can't drop a live call. */
+                int need = phys_off ? RESYNC_PICKUP_MS : RESYNC_HANGUP_MS;
+                if (resync_ms >= need) {
+                    ESP_LOGW(TAG, "hook resync: physical=%s but s_offhook=%d — correcting",
+                             phys_off ? "off-hook" : "on-hook", (int)s_offhook);
+                    s_offhook  = phys_off;
+                    last_level = phys_off ? HOOK_OFFHOOK_LEVEL : !HOOK_OFFHOOK_LEVEL;
+                    audio_pa_set(phys_off);
+                    if (!phys_off) audio_stop();
+                    report_offhook(phys_off);
+                    resync_ms = 0;
+                }
+            } else {
+                resync_ms = 0;
+            }
+        }
+        vTaskDelay(pdMS_TO_TICKS(10));
+    }
+}
+
+esp_err_t phone_init(void)
+{
+    BaseType_t ok = xTaskCreatePinnedToCore(phone_task, "phone",
+                                             TASK_STACK, NULL,
+                                             TASK_PRIO, NULL, 1);
+    return (ok == pdPASS) ? ESP_OK : ESP_ERR_NO_MEM;
+}
+
+bool phone_is_offhook(void)
+{
+    return s_offhook;
+}

plip_voice/main/phone.h

@@ -0,0 +1,36 @@
+#pragma once
+/*
+ * phone.h — Off-hook GPIO monitor + ring control for PLIP (Phase D).
+ *
+ * On the ESP32-A1S dev kit: BOOT button (GPIO4 = KEY1) acts as the
+ * off-hook stand-in (active LOW via INPUT_PULLUP, CONFIG_PLIP_HOOK_GPIO).
+ * On the Si3210 PCB target: replace with the SLIC interrupt GPIO.
+ *
+ * Ring: drives audio_ring_start() / audio_stop().
+ * Hook events: forwarded to hook_client_report().
+ */
+
+#include <stdbool.h>
+#include "esp_err.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Initialise the off-hook GPIO and start the monitoring task.
+ * hook_client_init() and audio_init() must have been called first. */
+esp_err_t phone_init(void);
+
+/* Signal the phone to start ringing (called externally if needed). */
+void phone_ring_start(void);
+
+/* Stop ringing. */
+void phone_ring_stop(void);
+
+/* Returns true if the handset is currently off-hook (picked up).
+ * Safe to call from any task; backed by a volatile flag updated in phone_task. */
+bool phone_is_offhook(void);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/main/plip_gamebook.c

@@ -0,0 +1,190 @@
+// plip_gamebook.c — see plip_gamebook.h. Audio CYOA for the PLIP phone.
+#include "plip_gamebook.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "cJSON.h"
+#include "esp_log.h"
+#include "esp_random.h"
+
+#include "audio.h"
+
+static const char *TAG = "plip_gb";
+
+#define GB_DIR      "/sdcard/gamebook"
+#define GB_LIBRARY  GB_DIR "/library.json"
+#define GB_MENU_WAV GB_DIR "/menu.wav"
+#define GB_JSON_MAX (64 * 1024)
+
+typedef enum { GB_OFF, GB_MENU, GB_STORY } gb_mode_t;
+
+static cJSON   *s_lib_root  = NULL;   // owns library.json
+static cJSON   *s_lib       = NULL;   // borrowed: root->"library"
+static int      s_lib_n     = 0;
+
+static cJSON   *s_book      = NULL;   // owns the current <id>.json
+static cJSON   *s_passages  = NULL;   // borrowed: book->"passages"
+static char     s_current[48] = {0};
+
+static gb_mode_t s_mode = GB_OFF;
+
+bool plip_gamebook_active(void) { return s_mode != GB_OFF; }
+
+static cJSON *load_json(const char *path)
+{
+    FILE *f = fopen(path, "rb");
+    if (!f) { ESP_LOGW(TAG, "open %s failed", path); return NULL; }
+    fseek(f, 0, SEEK_END);
+    long sz = ftell(f);
+    rewind(f);
+    if (sz <= 0 || sz > GB_JSON_MAX) { fclose(f); return NULL; }
+    char *buf = malloc((size_t)sz + 1);
+    if (!buf) { fclose(f); return NULL; }
+    size_t rd = fread(buf, 1, (size_t)sz, f);
+    fclose(f);
+    buf[rd] = '\0';
+    cJSON *root = cJSON_Parse(buf);
+    free(buf);
+    if (!root) ESP_LOGW(TAG, "malformed JSON: %s", path);
+    return root;
+}
+
+static void play_wav(const char *file)   // file = bare name under GB_DIR
+{
+    char path[96];
+    snprintf(path, sizeof(path), "%s/%s", GB_DIR, file);
+    audio_stop();                 // interrupt any current narration
+    audio_play_async(path);
+}
+
+bool plip_gamebook_available(void)
+{
+    if (!audio_ensure_sd()) return false;
+    FILE *f = fopen(GB_LIBRARY, "rb");
+    if (!f) return false;
+    fclose(f);
+    return true;
+}
+
+static void free_story(void)
+{
+    if (s_book) { cJSON_Delete(s_book); s_book = NULL; }
+    s_passages = NULL;
+    s_current[0] = '\0';
+}
+
+static void enter_menu(void)
+{
+    free_story();
+    s_mode = GB_MENU;
+    audio_stop();
+    audio_play_async(GB_MENU_WAV);   // absolute path
+    ESP_LOGI(TAG, "menu (%d stories)", s_lib_n);
+}
+
+// Play a passage by id (its WAV narrates the text + numbered choices).
+static void enter_passage(const char *pid)
+{
+    const cJSON *p = cJSON_GetObjectItem(s_passages, pid);
+    if (!cJSON_IsObject(p)) { ESP_LOGW(TAG, "passage '%s' missing", pid); return; }
+    snprintf(s_current, sizeof(s_current), "%s", pid);
+    const cJSON *wav = cJSON_GetObjectItem(p, "wav");
+    if (cJSON_IsString(wav) && wav->valuestring[0]) play_wav(wav->valuestring);
+    const cJSON *ch = cJSON_GetObjectItem(p, "choices");
+    ESP_LOGI(TAG, "passage '%s' (%d choices)", pid,
+             cJSON_IsArray(ch) ? cJSON_GetArraySize(ch) : 0);
+}
+
+static void load_book(int idx)
+{
+    const cJSON *e = cJSON_GetArrayItem(s_lib, idx);
+    const cJSON *file = cJSON_GetObjectItem(e, "book");
+    if (!cJSON_IsString(file)) return;
+    char path[128];
+    snprintf(path, sizeof(path), "%s/%s", GB_DIR, file->valuestring);
+    cJSON *book = load_json(path);
+    if (!book) return;
+    const cJSON *passages = cJSON_GetObjectItem(book, "passages");
+    const cJSON *start = cJSON_GetObjectItem(book, "start");
+    if (!cJSON_IsObject(passages) || !cJSON_IsString(start)) {
+        cJSON_Delete(book); return;
+    }
+    free_story();
+    s_book = book;
+    s_passages = (cJSON *)passages;
+    s_mode = GB_STORY;
+    ESP_LOGI(TAG, "load book #%d @ '%s'", idx, start->valuestring);
+    enter_passage(start->valuestring);
+}
+
+void plip_gamebook_begin(void)
+{
+    plip_gamebook_end();   // clean any previous run
+
+    s_lib_root = load_json(GB_LIBRARY);
+    if (!s_lib_root) return;
+    s_lib = cJSON_GetObjectItem(s_lib_root, "library");
+    if (!cJSON_IsArray(s_lib) || cJSON_GetArraySize(s_lib) == 0) {
+        cJSON_Delete(s_lib_root); s_lib_root = NULL; s_lib = NULL;
+        return;
+    }
+    s_lib_n = cJSON_GetArraySize(s_lib);
+    audio_pa_set(true);    // earpiece amplifier on
+    enter_menu();
+}
+
+void plip_gamebook_begin_random(void)
+{
+    plip_gamebook_end();   // clean any previous run
+
+    s_lib_root = load_json(GB_LIBRARY);
+    if (!s_lib_root) return;
+    s_lib = cJSON_GetObjectItem(s_lib_root, "library");
+    if (!cJSON_IsArray(s_lib) || cJSON_GetArraySize(s_lib) == 0) {
+        cJSON_Delete(s_lib_root); s_lib_root = NULL; s_lib = NULL;
+        return;
+    }
+    s_lib_n = cJSON_GetArraySize(s_lib);
+    audio_pa_set(true);    // earpiece amplifier on
+    int idx = (int)(esp_random() % (uint32_t)s_lib_n);
+    ESP_LOGI(TAG, "auto-start random story #%d/%d", idx, s_lib_n);
+    load_book(idx);        // straight into the story, no menu
+}
+
+void plip_gamebook_feed_digit(int d)
+{
+    if (s_mode == GB_MENU) {
+        if (d >= 1 && d <= s_lib_n) load_book(d - 1);
+        return;
+    }
+    if (s_mode == GB_STORY) {
+        const cJSON *p = cJSON_GetObjectItem(s_passages, s_current);
+        const cJSON *ch = cJSON_GetObjectItem(p, "choices");
+        int n = cJSON_IsArray(ch) ? cJSON_GetArraySize(ch) : 0;
+        if (n == 0) {                 // ending: 0 (or any) → back to menu
+            enter_menu();
+            return;
+        }
+        if (d >= 1 && d <= n) {
+            const cJSON *g = cJSON_GetObjectItem(cJSON_GetArrayItem(ch, d - 1),
+                                                 "goto");
+            if (cJSON_IsString(g)) {
+                ESP_LOGI(TAG, "choice %d -> '%s'", d, g->valuestring);
+                enter_passage(g->valuestring);
+            }
+        }
+        // out-of-range digit: ignore (player can dial again)
+    }
+}
+
+void plip_gamebook_end(void)
+{
+    if (s_mode == GB_OFF && !s_lib_root && !s_book) return;
+    s_mode = GB_OFF;
+    audio_stop();
+    free_story();
+    if (s_lib_root) { cJSON_Delete(s_lib_root); s_lib_root = NULL; s_lib = NULL; }
+    ESP_LOGI(TAG, "ended");
+}

plip_voice/main/plip_gamebook.h

@@ -0,0 +1,48 @@
+#pragma once
+// plip_gamebook — "livre dont vous êtes le héros" AUDIO pour le téléphone PLIP.
+//
+// Pas d'écran : tout passe par l'écouteur. On décroche, le narrateur lit le
+// menu des histoires ("pour le dragon, faites le 1..."), on compose un chiffre
+// au cadran pour choisir une histoire, puis à chaque passage le narrateur lit
+// le texte ET les choix numérotés ("pour examiner la machine, faites le 1..."),
+// et on compose le chiffre du choix. Raccrocher arrête tout.
+//
+// Lit /sdcard/gamebook/{library.json, menu.wav, <id>.json, <id>_<passage>.wav}
+// (pack audio "téléphone" produit par tools/gamebook/build_phone_gamebook.py,
+// où la narration des WAV inclut déjà les invites "faites le N"). 100% local,
+// aucun modèle, aucun réseau.
+//
+// Intégration : conversation.c lance plip_gamebook_begin() au décroché (si un
+// pack est présent sur la SD), pousse chaque chiffre composé via
+// plip_gamebook_feed_digit(), et appelle plip_gamebook_end() au raccroché.
+
+#include <stdbool.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// La SD est-elle montée et un pack gamebook présent ? (à tester au décroché)
+bool plip_gamebook_available(void);
+
+// Démarre le mode livre-jeu : charge la bibliothèque et joue le menu audio.
+void plip_gamebook_begin(void);
+
+// Démarre directement UNE histoire tirée au hasard (sans passer par le menu) —
+// utilisé quand le téléphone sonne tout seul et qu'on décroche.
+void plip_gamebook_begin_random(void);
+
+// Traite un chiffre composé (0..9). Menu → choisit l'histoire ; en histoire →
+// choisit la réponse ; sur une fin → 0 revient au menu. Interrompt la
+// narration en cours pour enchaîner.
+void plip_gamebook_feed_digit(int digit);
+
+// Quitte le mode livre-jeu (raccroché) : stoppe l'audio, libère la mémoire.
+void plip_gamebook_end(void);
+
+// Vrai tant qu'une partie est en cours.
+bool plip_gamebook_active(void);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/main/slic.c

@@ -0,0 +1,184 @@
+/*
+ * slic.c — K50835F / AG1171-class SLIC control (ESP-IDF port).
+ *
+ * Ported from hardware/projects/slic-phone/src/slic/Ks0835SlicController.cpp
+ * (Arduino) to bare ESP-IDF 5.x GPIO driver.
+ *
+ * Key differences from Arduino original:
+ *  - Open-drain on PD achieved via GPIO_MODE_OUTPUT_OD + gpio_set_level(PD, 1)
+ *  - Ring FR toggle runs as a FreeRTOS task instead of a cooperative tick()
+ *  - hook_active_high is hard-coded TRUE (matches A252ConfigStore default)
+ */
+
+#include "slic.h"
+#include "board_config.h"
+
+#include "driver/gpio.h"
+#include "esp_log.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+
+#define TAG "slic"
+
+/* Empirically on THIS A1S+SLIC unit, SHK is ACTIVE-HIGH: off-hook (loop closed)
+ * drives SHK HIGH, on-hook drives it LOW.  Confirmed at the bench via
+ * /debug/hookmon: on-hook→0, pickup-during-ring→0→1 transition, off-hook→1.
+ * (The A252 reference was active-low; the polarity is inverted here, so the
+ * raw straight-wired GPIO reads off-hook = HIGH.) */
+#define SLIC_SHK_OFFHOOK_LEVEL  1
+
+static volatile bool s_ringing = false;
+static TaskHandle_t  s_ring_task = NULL;
+
+/* France Télécom ring cadence: 1.5 s burst ON, 3.5 s silent pause, repeating.
+ * (Single-ring cadence — distinct from the UK double-ring or US 2 s/4 s.)
+ * The bell is driven during the burst by toggling FR at ~25 Hz with RM HIGH;
+ * during the pause RM and FR are held low so the bell is silent. */
+#define RING_FR_TOGGLE_MS  20     /* 20 ms half-period → ~25 Hz bell drive */
+#define RING_BURST_MS      1500   /* FT: 1.5 s of ringing */
+#define RING_PAUSE_MS      3500   /* FT: 3.5 s of silence */
+
+/* ── FR toggle + cadence task ─────────────────────────────────────────────── */
+
+static void slic_ring_task(void *arg)
+{
+    (void)arg;
+    bool fr_state = false;
+    bool in_burst = true;     /* start each ring with a burst */
+    int  phase_ms = 0;        /* elapsed ms in the current burst/pause phase */
+
+    for (;;) {
+        if (s_ringing) {
+            if (in_burst) {
+                /* Ring burst: RM HIGH, FR toggling at ~25 Hz to swing the bell */
+                gpio_set_level(PLIP_SLIC_RM, 1);
+                fr_state = !fr_state;
+                gpio_set_level(PLIP_SLIC_FR, fr_state ? 1 : 0);
+                phase_ms += RING_FR_TOGGLE_MS;
+                if (phase_ms >= RING_BURST_MS) {
+                    /* End of burst → enter silent pause */
+                    in_burst = false;
+                    phase_ms = 0;
+                    fr_state = false;
+                    gpio_set_level(PLIP_SLIC_RM, 0);
+                    gpio_set_level(PLIP_SLIC_FR, 0);
+                    ESP_LOGV(TAG, "ring: burst end -> pause");
+                }
+            } else {
+                /* Silent pause: RM/FR stay low */
+                phase_ms += RING_FR_TOGGLE_MS;
+                if (phase_ms >= RING_PAUSE_MS) {
+                    in_burst = true;
+                    phase_ms = 0;
+                    ESP_LOGV(TAG, "ring: pause end -> burst");
+                }
+            }
+        } else {
+            /* Idle — keep RM/FR low and reset the cadence for the next ring */
+            gpio_set_level(PLIP_SLIC_RM, 0);
+            gpio_set_level(PLIP_SLIC_FR, 0);
+            fr_state = false;
+            in_burst = true;
+            phase_ms = 0;
+        }
+        vTaskDelay(pdMS_TO_TICKS(RING_FR_TOGGLE_MS));
+    }
+}
+
+/* ── Public API ──────────────────────────────────────────────────────────── */
+
+esp_err_t slic_init(void)
+{
+    /* RM: Ring Mode output, init LOW.
+     * INPUT_OUTPUT (not plain OUTPUT) so gpio_get_level() reads back the real
+     * pad level — plain OUTPUT disables the input buffer and always reads 0. */
+    gpio_config_t rm_cfg = {
+        .pin_bit_mask = (1ULL << PLIP_SLIC_RM),
+        .mode         = GPIO_MODE_INPUT_OUTPUT,
+        .pull_up_en   = GPIO_PULLUP_DISABLE,
+        .pull_down_en = GPIO_PULLDOWN_DISABLE,
+        .intr_type    = GPIO_INTR_DISABLE,
+    };
+    esp_err_t ret = gpio_config(&rm_cfg);
+    if (ret != ESP_OK) return ret;
+    gpio_set_level(PLIP_SLIC_RM, 0);
+
+    /* FR: Forward/Reverse output, init LOW (INPUT_OUTPUT for readback, see RM). */
+    gpio_config_t fr_cfg = {
+        .pin_bit_mask = (1ULL << PLIP_SLIC_FR),
+        .mode         = GPIO_MODE_INPUT_OUTPUT,
+        .pull_up_en   = GPIO_PULLUP_DISABLE,
+        .pull_down_en = GPIO_PULLDOWN_DISABLE,
+        .intr_type    = GPIO_INTR_DISABLE,
+    };
+    ret = gpio_config(&fr_cfg);
+    if (ret != ESP_OK) return ret;
+    gpio_set_level(PLIP_SLIC_FR, 0);
+
+    /* SHK: Switch Hook input with pull-up (physical line has no external pull) */
+    gpio_config_t shk_cfg = {
+        .pin_bit_mask = (1ULL << PLIP_SLIC_SHK),
+        .mode         = GPIO_MODE_INPUT,
+        .pull_up_en   = GPIO_PULLUP_ENABLE,
+        .pull_down_en = GPIO_PULLDOWN_DISABLE,
+        .intr_type    = GPIO_INTR_DISABLE,
+    };
+    ret = gpio_config(&shk_cfg);
+    if (ret != ESP_OK) return ret;
+
+    /* PD: Power Down — EXACT A252-proven sequence: open-drain output, HIGH = released
+     * = SLIC active (setPowerDown(false) in Ks0835SlicController). This is the config
+     * the working Arduino slic-phone project uses on the same chip. */
+    /* INPUT_OUTPUT_OD (not plain OUTPUT_OD) so gpio_get_level() reads the real
+     * pad level — OUTPUT_OD also disables the input buffer and would read 0. */
+    ret = gpio_set_direction(PLIP_SLIC_PD, GPIO_MODE_INPUT_OUTPUT_OD);
+    if (ret != ESP_OK) return ret;
+    ret = gpio_set_level(PLIP_SLIC_PD, 1);   /* open-drain released HIGH = active (A252-proven) */
+    if (ret != ESP_OK) return ret;
+
+    ESP_LOGI(TAG, "slic: PD GPIO%d open-drain HIGH (A252-proven active) — testing with fixed mic", PLIP_SLIC_PD);
+    ESP_LOGI(TAG, "slic: pins RM=%d FR=%d SHK=%d PD=%d hook_active_high=1",
+             PLIP_SLIC_RM, PLIP_SLIC_FR, PLIP_SLIC_SHK, PLIP_SLIC_PD);
+
+    /* Spawn the FR-toggle task (runs indefinitely, toggles only when s_ringing=true) */
+    BaseType_t ok = xTaskCreatePinnedToCore(slic_ring_task, "slic_ring",
+                                             2048, NULL, 4, &s_ring_task, 1);
+    if (ok != pdPASS) {
+        ESP_LOGE(TAG, "Failed to create slic_ring task");
+        return ESP_ERR_NO_MEM;
+    }
+
+    int shk = gpio_get_level(PLIP_SLIC_SHK);
+    ESP_LOGI(TAG, "SLIC init OK — SHK level=%d (%s)",
+             shk, (shk == SLIC_SHK_OFFHOOK_LEVEL) ? "off-hook" : "on-hook");
+    return ESP_OK;
+}
+
+bool slic_is_offhook(void)
+{
+    return gpio_get_level(PLIP_SLIC_SHK) == SLIC_SHK_OFFHOOK_LEVEL;
+}
+
+void slic_ring_start(void)
+{
+    if (s_ringing) return;
+    ESP_LOGI(TAG, "ring start: France Télécom cadence %d ms ON / %d ms OFF",
+             RING_BURST_MS, RING_PAUSE_MS);
+    /* The cadence task drives RM/FR; it starts on a burst (in_burst reset in
+     * the idle branch). Just arm it here. */
+    s_ringing = true;
+}
+
+void slic_ring_stop(void)
+{
+    if (!s_ringing) return;
+    ESP_LOGI(TAG, "ring stop: RM=LOW, FR=LOW");
+    s_ringing = false;
+    gpio_set_level(PLIP_SLIC_RM, 0);
+    gpio_set_level(PLIP_SLIC_FR, 0);
+}
+
+bool slic_is_ringing(void)
+{
+    return s_ringing;
+}

plip_voice/main/slic.h

@@ -0,0 +1,59 @@
+/*
+ * slic.h — K50835F / AG1171-class SLIC control (ESP-IDF port of Ks0835SlicController).
+ *
+ * Pins (A1S board, from board_config.h):
+ *   RM  GPIO18  Ring Mode output     — HIGH = ring burst active
+ *   FR  GPIO5   Forward/Reverse out  — toggled at ~25 Hz during ring to drive bell
+ *   SHK GPIO23  Switch Hook input    — HIGH = off-hook (active-high)
+ *   PD  GPIO19  Power Down (OD)      — HIGH = SLIC powered (released open-drain)
+ *
+ * Power-up: PD is open-drain, written HIGH → high-impedance → SLIC active.
+ * Power-down: PD driven LOW → SLIC off (no audio, no hook sense).
+ *
+ * Ring: slic_ring_start() sets RM=1 and spawns/enables a 20 ms FR-toggle task.
+ *       slic_ring_stop()  clears RM=0, FR=0, pauses the toggle task.
+ *
+ * Hook: slic_is_offhook() reads SHK; HIGH = off-hook (matches A252ConfigStore default).
+ */
+
+#pragma once
+
+#include <stdbool.h>
+#include "esp_err.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * Initialise SLIC GPIO and power up the SLIC.
+ * Must be called early in boot, before audio init.
+ * Returns ESP_OK on success.
+ */
+esp_err_t slic_init(void);
+
+/**
+ * Returns true when the handset is off-hook (SHK GPIO23 HIGH).
+ */
+bool slic_is_offhook(void);
+
+/**
+ * Activate ringing: RM=HIGH + FR toggles at ~25 Hz (20 ms period).
+ * No-op if already ringing.
+ */
+void slic_ring_start(void);
+
+/**
+ * Stop ringing: RM=LOW, FR=LOW, FR toggle suspended.
+ * No-op if not ringing.
+ */
+void slic_ring_stop(void);
+
+/**
+ * Returns true if slic_ring_start() was called and not yet stopped.
+ */
+bool slic_is_ringing(void);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/main/tones.c

@@ -0,0 +1,139 @@
+/*
+ * tones.c — France Télécom telephone tones (dial / busy / ringback).
+ *
+ * Generates 440 Hz sinusoid with on/off cadence written directly to the
+ * I2S speaker handle (stereo — ES8388 requires L+R pairs).
+ *
+ * Cadences (all at 440 Hz):
+ *   DIAL     : continuous
+ *   BUSY     : 0.5 s on / 0.5 s off (1 s cycle)
+ *   RINGBACK : 1.5 s on / 3.5 s off (5 s cycle)
+ */
+
+#include "tones.h"
+#include "audio.h"
+
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "driver/i2s_std.h"
+#include "esp_log.h"
+
+#include <math.h>
+#include <string.h>
+#include <stdbool.h>
+
+#define TAG "tones"
+
+typedef enum { TONE_NONE, TONE_DIAL, TONE_BUSY, TONE_RINGBACK } tone_mode_t;
+
+static volatile tone_mode_t s_mode      = TONE_NONE;
+static volatile bool        s_tone_idle = true;  /* true when tone_task is NOT in i2s_channel_write */
+static TaskHandle_t         s_task      = NULL;
+
+#define SR    16000   /* sample rate (must match audio_init) */
+#define FRAME 320     /* 20 ms @ 16 kHz */
+
+/*
+ * Fill `n` stereo frames (L=R = 440 Hz sine) into buf.
+ * buf must hold n * 2 int16_t (stereo interleaved).
+ * ph: running phase accumulator (mono sample index).
+ */
+static void fill_440_stereo(int16_t *buf, int n, int *ph)
+{
+    for (int i = 0; i < n; i++) {
+        int16_t v = (int16_t)(8000.0f * sinf(2.0f * (float)M_PI * 440.0f * (*ph) / SR));
+        buf[i * 2]     = v;  /* L */
+        buf[i * 2 + 1] = v;  /* R */
+        (*ph)++;
+    }
+}
+
+static void tone_task(void *a)
+{
+    (void)a;
+    /* Stereo buffer: FRAME mono samples × 2 channels × 2 bytes = FRAME*4 bytes */
+    int16_t buf[FRAME * 2];
+    int ph = 0;
+    int64_t t = 0;  /* ms elapsed in current tone state */
+
+    for (;;) {
+        tone_mode_t m = s_mode;
+        if (m == TONE_NONE) {
+            s_tone_idle = true;
+            vTaskDelay(pdMS_TO_TICKS(20));
+            t = 0;
+            ph = 0;
+            continue;
+        }
+
+        bool on = true;
+        if (m == TONE_BUSY) {
+            on = (t % 1000) < 500;          /* 500ms on / 500ms off */
+        } else if (m == TONE_RINGBACK) {
+            on = (t % 5000) < 1500;         /* 1.5s on / 3.5s off */
+        }
+        /* TONE_DIAL: always on */
+
+        if (on) {
+            fill_440_stereo(buf, FRAME, &ph);
+        } else {
+            memset(buf, 0, sizeof(buf));
+            /* Keep phase accumulator advancing so pitch is smooth on resumption */
+            ph += FRAME;
+        }
+
+        size_t written = 0;
+        s_tone_idle = false;
+        i2s_channel_write(audio_spk_handle(), buf, sizeof(buf), &written, pdMS_TO_TICKS(50));
+        t += 20;  /* each frame = 20 ms */
+    }
+}
+
+static void ensure_task(void)
+{
+    if (!s_task) {
+        xTaskCreate(tone_task, "tones", 3072, NULL, 5, &s_task);
+        ESP_LOGI(TAG, "tone_task created");
+    }
+}
+
+void tones_dialtone_start(void)
+{
+    ESP_LOGI(TAG, "dial tone start");
+    audio_pa_set(true);
+    ensure_task();
+    s_mode = TONE_DIAL;
+}
+
+void tones_busy_start(void)
+{
+    ESP_LOGI(TAG, "busy tone start");
+    audio_pa_set(true);
+    ensure_task();
+    s_mode = TONE_BUSY;
+}
+
+void tones_ringback_start(void)
+{
+    ESP_LOGI(TAG, "ringback tone start");
+    audio_pa_set(true);
+    ensure_task();
+    s_mode = TONE_RINGBACK;
+}
+
+void tones_stop(void)
+{
+    ESP_LOGI(TAG, "tones stop");
+    s_mode = TONE_NONE;
+    /* Wait until tone_task is no longer inside i2s_channel_write (I2S arbitration).
+     * Guard: max 200 ms so we never block indefinitely. */
+    const int max_wait_ms = 200;
+    int waited_ms = 0;
+    while (!s_tone_idle && waited_ms < max_wait_ms) {
+        vTaskDelay(pdMS_TO_TICKS(5));
+        waited_ms += 5;
+    }
+    if (waited_ms >= max_wait_ms) {
+        ESP_LOGW(TAG, "tones_stop: timeout waiting for tone_task idle");
+    }
+}

plip_voice/main/tones.h

@@ -0,0 +1,7 @@
+#pragma once
+#include "esp_err.h"
+
+void tones_dialtone_start(void);   /* 440 Hz continuous */
+void tones_busy_start(void);       /* 440 Hz 0.5s on/off */
+void tones_ringback_start(void);   /* 440 Hz 1.5s/3.5s */
+void tones_stop(void);

plip_voice/main/turn_client.c

@@ -0,0 +1,340 @@
+/*
+ * turn_client.c — NPC gateway client for /v1/voice/turn and /v1/voice/reply.
+ *
+ * Uses esp_http_client open/write/fetch/read streaming so the binary WAV
+ * body can be written directly to SPIFFS without a large heap buffer.
+ *
+ * Pattern mirrors hook_client.c but synchronous (called from conv_task).
+ * Timeout is generous (30 s) because TTS synthesis adds latency.
+ *
+ * Stage 3 adds turn_client_reply() which POSTs captured mic audio as
+ * multipart/form-data to /v1/voice/reply and streams the NPC response WAV.
+ */
+
+#include "turn_client.h"
+
+#include "sdkconfig.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#include "esp_http_client.h"
+#include "esp_log.h"
+
+#define TAG "turn_client"
+
+/* WAV header is 44 bytes; anything shorter is certainly not a valid response. */
+#define MIN_WAV_BYTES 44
+
+/* Read chunks when streaming the response body. */
+#define CHUNK_SIZE    1024
+
+/* Write chunks when uploading the captured WAV body (4 KB). */
+#define UPLOAD_CHUNK  4096
+
+/* Timeout for the full TTS round-trip (connect + synthesis + transfer). */
+#define TIMEOUT_MS    90000  /* reply TTS (Kyutai MLX ~0.3x realtime) can take
+                              * tens of seconds; was 30s → reply POST timed out
+                              * (HTTP -1). 90s covers worst-case generation. */
+
+/* Multipart boundary (must not appear in the WAV payload — 16kHz PCM is binary
+ * so any fixed ASCII boundary is safe). */
+#define BOUNDARY "----ZacusPlipBoundary7MA4YWxkTrZu0gW"
+
+bool turn_client_greeting(const char *session_id,
+                           const char *number,
+                           const char *out_path)
+{
+    /* --- Build URL -------------------------------------------------------- */
+    char url[256];
+    snprintf(url, sizeof(url), "%s/v1/voice/turn",
+             CONFIG_PLIP_GATEWAY_URL);
+
+    /* --- Build JSON body -------------------------------------------------- */
+    char body[256];
+    int body_len = snprintf(body, sizeof(body),
+        "{\"session_id\":\"%s\",\"number\":\"%s\",\"kind\":\"greeting\"}",
+        session_id ? session_id : "",
+        number     ? number     : "");
+
+    /* --- Configure client ------------------------------------------------- */
+    esp_http_client_config_t cfg = {
+        .url        = url,
+        .method     = HTTP_METHOD_POST,
+        .timeout_ms = TIMEOUT_MS,
+    };
+    esp_http_client_handle_t client = esp_http_client_init(&cfg);
+    if (!client) {
+        ESP_LOGE(TAG, "esp_http_client_init failed");
+        return false;
+    }
+
+    esp_http_client_set_header(client, "Content-Type", "application/json");
+
+    /* Authorization header — skip if token is empty */
+    const char *token = CONFIG_PLIP_GATEWAY_TOKEN;
+    if (token && token[0] != '\0') {
+        char auth[128];
+        snprintf(auth, sizeof(auth), "Bearer %s", token);
+        esp_http_client_set_header(client, "Authorization", auth);
+    }
+
+    /* --- Open connection and write body (streaming POST) ------------------ */
+    esp_err_t err = esp_http_client_open(client, body_len);
+    if (err != ESP_OK) {
+        ESP_LOGW(TAG, "open %s failed: %s", url, esp_err_to_name(err));
+        esp_http_client_cleanup(client);
+        return false;
+    }
+
+    int written = esp_http_client_write(client, body, body_len);
+    if (written < 0) {
+        ESP_LOGW(TAG, "write body failed (written=%d)", written);
+        esp_http_client_close(client);
+        esp_http_client_cleanup(client);
+        return false;
+    }
+
+    /* --- Fetch response headers ------------------------------------------ */
+    int content_len = esp_http_client_fetch_headers(client);
+    int status_code = esp_http_client_get_status_code(client);
+
+    if (status_code != 200) {
+        ESP_LOGW(TAG, "gateway returned HTTP %d (url=%s)", status_code, url);
+        esp_http_client_close(client);
+        esp_http_client_cleanup(client);
+        return false;
+    }
+
+    ESP_LOGI(TAG, "HTTP 200 from %s (content_length=%d)", url, content_len);
+
+    /* --- Stream binary body into SPIFFS file ------------------------------ */
+    FILE *fp = fopen(out_path, "wb");
+    if (!fp) {
+        ESP_LOGE(TAG, "fopen(%s, wb) failed", out_path);
+        esp_http_client_close(client);
+        esp_http_client_cleanup(client);
+        return false;
+    }
+
+    static uint8_t s_chunk[CHUNK_SIZE];   /* static: avoids stack pressure */
+    int total = 0;
+    int rd;
+    while ((rd = esp_http_client_read(client, (char *)s_chunk, sizeof(s_chunk))) > 0) {
+        size_t fw = fwrite(s_chunk, 1, (size_t)rd, fp);
+        if ((int)fw != rd) {
+            ESP_LOGW(TAG, "fwrite short: wrote %d of %d bytes", (int)fw, rd);
+            break;
+        }
+        total += rd;
+    }
+
+    fclose(fp);
+    esp_http_client_close(client);
+    esp_http_client_cleanup(client);
+
+    if (total <= MIN_WAV_BYTES) {
+        ESP_LOGW(TAG, "greeting WAV too short (%d bytes) — ignoring", total);
+        return false;
+    }
+
+    ESP_LOGI(TAG, "greeting WAV %d bytes written to %s", total, out_path);
+    return true;
+}
+
+/* ── turn_client_reply ─────────────────────────────────────────────────────
+ *
+ * POST multipart/form-data to /v1/voice/reply with the captured mic WAV.
+ *
+ * Multipart layout (each part is CRLF-delimited per RFC 2046):
+ *
+ *   --<boundary>\r\n
+ *   Content-Disposition: form-data; name="session_id"\r\n\r\n
+ *   <session_id>\r\n
+ *   --<boundary>\r\n
+ *   Content-Disposition: form-data; name="number"\r\n\r\n
+ *   <number>\r\n
+ *   --<boundary>\r\n
+ *   Content-Disposition: form-data; name="audio"; filename="rec.wav"\r\n
+ *   Content-Type: audio/wav\r\n\r\n
+ *   <wav bytes>
+ *   \r\n--<boundary>--\r\n
+ *
+ * Content-Length = len(preamble) + wav_len + len(epilogue)
+ * (known exactly before writing — allows non-chunked POST).
+ */
+esp_err_t turn_client_reply(const char *session_id,
+                             const char *number,
+                             const uint8_t *wav,
+                             size_t wav_len,
+                             const char *out_path)
+{
+    if (!session_id || !number || !wav || wav_len == 0 || !out_path) {
+        return ESP_ERR_INVALID_ARG;
+    }
+
+    /* --- Build URL -------------------------------------------------------- */
+    char url[256];
+    snprintf(url, sizeof(url), "%s/v1/voice/reply",
+             CONFIG_PLIP_GATEWAY_URL);
+
+    /* --- Build multipart preamble ---------------------------------------- */
+    /* preamble = 3 parts before the raw wav bytes:
+     *   part 1 — session_id (text field)
+     *   part 2 — number (text field)
+     *   part 3 — audio file header (up to but NOT including the file body)
+     */
+    char preamble[512];
+    int preamble_len = snprintf(preamble, sizeof(preamble),
+        "--%s\r\n"
+        "Content-Disposition: form-data; name=\"session_id\"\r\n\r\n"
+        "%s\r\n"
+        "--%s\r\n"
+        "Content-Disposition: form-data; name=\"number\"\r\n\r\n"
+        "%s\r\n"
+        "--%s\r\n"
+        "Content-Disposition: form-data; name=\"audio\"; filename=\"rec.wav\"\r\n"
+        "Content-Type: audio/wav\r\n\r\n",
+        BOUNDARY,
+        session_id,
+        BOUNDARY,
+        number,
+        BOUNDARY);
+
+    if (preamble_len <= 0 || preamble_len >= (int)sizeof(preamble)) {
+        ESP_LOGE(TAG, "preamble buffer overflow (len=%d)", preamble_len);
+        return ESP_ERR_INVALID_SIZE;
+    }
+
+    /* --- Build multipart epilogue ---------------------------------------- */
+    /* epilogue = CRLF after the wav data + closing boundary */
+    char epilogue[64];
+    int epilogue_len = snprintf(epilogue, sizeof(epilogue),
+        "\r\n--%s--\r\n",
+        BOUNDARY);
+
+    if (epilogue_len <= 0 || epilogue_len >= (int)sizeof(epilogue)) {
+        ESP_LOGE(TAG, "epilogue buffer overflow");
+        return ESP_ERR_INVALID_SIZE;
+    }
+
+    int content_length = preamble_len + (int)wav_len + epilogue_len;
+
+    ESP_LOGI(TAG, "reply POST %s preamble=%d wav=%zu epilogue=%d total=%d",
+             url, preamble_len, wav_len, epilogue_len, content_length);
+
+    /* --- Configure client ------------------------------------------------- */
+    esp_http_client_config_t cfg = {
+        .url        = url,
+        .method     = HTTP_METHOD_POST,
+        .timeout_ms = TIMEOUT_MS,
+    };
+    esp_http_client_handle_t client = esp_http_client_init(&cfg);
+    if (!client) {
+        ESP_LOGE(TAG, "esp_http_client_init failed");
+        return ESP_ERR_NO_MEM;
+    }
+
+    /* Content-Type with boundary */
+    char ct_header[128];
+    snprintf(ct_header, sizeof(ct_header),
+             "multipart/form-data; boundary=%s", BOUNDARY);
+    esp_http_client_set_header(client, "Content-Type", ct_header);
+
+    /* Authorization header — skip if token is empty */
+    const char *token = CONFIG_PLIP_GATEWAY_TOKEN;
+    if (token && token[0] != '\0') {
+        char auth[128];
+        snprintf(auth, sizeof(auth), "Bearer %s", token);
+        esp_http_client_set_header(client, "Authorization", auth);
+    }
+
+    /* --- Open connection and stream body ---------------------------------- */
+    esp_err_t err = esp_http_client_open(client, content_length);
+    if (err != ESP_OK) {
+        ESP_LOGW(TAG, "open %s failed: %s", url, esp_err_to_name(err));
+        esp_http_client_cleanup(client);
+        return err;
+    }
+
+    /* Write preamble */
+    int written = esp_http_client_write(client, preamble, preamble_len);
+    if (written < 0) {
+        ESP_LOGW(TAG, "write preamble failed (ret=%d)", written);
+        esp_http_client_close(client);
+        esp_http_client_cleanup(client);
+        return ESP_FAIL;
+    }
+
+    /* Write WAV data in 4 KB chunks */
+    size_t remaining = wav_len;
+    const uint8_t *ptr = wav;
+    while (remaining > 0) {
+        size_t to_send = (remaining < UPLOAD_CHUNK) ? remaining : UPLOAD_CHUNK;
+        int wr = esp_http_client_write(client, (const char *)ptr, (int)to_send);
+        if (wr < 0) {
+            ESP_LOGW(TAG, "write wav chunk failed (ret=%d)", wr);
+            esp_http_client_close(client);
+            esp_http_client_cleanup(client);
+            return ESP_FAIL;
+        }
+        ptr       += (size_t)wr;
+        remaining -= (size_t)wr;
+    }
+
+    /* Write epilogue */
+    written = esp_http_client_write(client, epilogue, epilogue_len);
+    if (written < 0) {
+        ESP_LOGW(TAG, "write epilogue failed (ret=%d)", written);
+        esp_http_client_close(client);
+        esp_http_client_cleanup(client);
+        return ESP_FAIL;
+    }
+
+    /* --- Fetch response headers ------------------------------------------ */
+    int content_len = esp_http_client_fetch_headers(client);
+    int status_code = esp_http_client_get_status_code(client);
+
+    ESP_LOGI(TAG, "reply HTTP %d content_length=%d", status_code, content_len);
+
+    if (status_code != 200) {
+        ESP_LOGW(TAG, "gateway returned HTTP %d for reply", status_code);
+        esp_http_client_close(client);
+        esp_http_client_cleanup(client);
+        return ESP_ERR_INVALID_RESPONSE;
+    }
+
+    /* --- Stream binary WAV response into SPIFFS file --------------------- */
+    FILE *fp = fopen(out_path, "wb");
+    if (!fp) {
+        ESP_LOGE(TAG, "fopen(%s, wb) failed", out_path);
+        esp_http_client_close(client);
+        esp_http_client_cleanup(client);
+        return ESP_ERR_NOT_FOUND;
+    }
+
+    static uint8_t s_reply_chunk[CHUNK_SIZE];  /* static: avoids stack pressure */
+    int total = 0;
+    int rd;
+    while ((rd = esp_http_client_read(client, (char *)s_reply_chunk,
+                                      sizeof(s_reply_chunk))) > 0) {
+        size_t fw = fwrite(s_reply_chunk, 1, (size_t)rd, fp);
+        if ((int)fw != rd) {
+            ESP_LOGW(TAG, "fwrite short: wrote %d of %d bytes", (int)fw, rd);
+            break;
+        }
+        total += rd;
+    }
+
+    fclose(fp);
+    esp_http_client_close(client);
+    esp_http_client_cleanup(client);
+
+    if (total <= MIN_WAV_BYTES) {
+        ESP_LOGW(TAG, "reply WAV too short (%d bytes) — ignoring", total);
+        return ESP_ERR_INVALID_SIZE;
+    }
+
+    ESP_LOGI(TAG, "reply WAV %d bytes written to %s", total, out_path);
+    return ESP_OK;
+}

plip_voice/main/turn_client.h

@@ -0,0 +1,66 @@
+#pragma once
+/*
+ * turn_client.h — POST /v1/voice/turn to the NPC gateway and retrieve a WAV response.
+ *
+ * Stage 2: greeting fetch (kind="greeting").
+ * Stage 3: reply fetch via /v1/voice/reply (multipart/form-data with captured WAV).
+ */
+#include <stdbool.h>
+#include <stdint.h>
+#include <stddef.h>
+#include "esp_err.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * turn_client_greeting — POST /v1/voice/turn with kind="greeting".
+ *
+ * Sends:  { "session_id": session_id, "number": number, "kind": "greeting" }
+ * Bearer: CONFIG_PLIP_GATEWAY_TOKEN (skipped if empty)
+ * URL:    CONFIG_PLIP_GATEWAY_URL/v1/voice/turn
+ *
+ * On success (HTTP 200, body > 44 bytes written to out_path):
+ *   - Streams the binary WAV response into fopen(out_path, "wb").
+ *   - Returns true.
+ *
+ * On failure (network error, non-200, short body):
+ *   - Logs a warning and returns false.
+ *   - Does not crash; caller may proceed silently.
+ */
+bool turn_client_greeting(const char *session_id,
+                           const char *number,
+                           const char *out_path);
+
+/*
+ * turn_client_reply — POST /v1/voice/reply as multipart/form-data.
+ *
+ * Endpoint: CONFIG_PLIP_GATEWAY_URL/v1/voice/reply
+ * Method:   POST multipart/form-data
+ * Fields:   session_id (text), number (text), audio (file, "rec.wav", audio/wav)
+ * Bearer:   CONFIG_PLIP_GATEWAY_TOKEN (skipped if empty)
+ *
+ * The function builds the multipart body in three segments:
+ *   preamble  = boundary + session_id part + boundary + number part +
+ *               boundary + audio file header
+ *   wav data  = wav bytes (wav_len bytes, sent in 4 KB chunks)
+ *   epilogue  = CRLF + closing boundary
+ *
+ * Content-Length = len(preamble) + wav_len + len(epilogue)
+ * The response body (WAV 16 kHz) is streamed into out_path.
+ *
+ * Response headers X-Zacus-Heard and X-Zacus-Said are logged at INFO level.
+ *
+ * Returns ESP_OK if HTTP 200 and a valid WAV (> 44 bytes) was written.
+ * On any failure: logs a warning and returns an ESP error code.
+ */
+esp_err_t turn_client_reply(const char *session_id,
+                             const char *number,
+                             const uint8_t *wav,
+                             size_t wav_len,
+                             const char *out_path);
+
+#ifdef __cplusplus
+}
+#endif

plip_voice/partitions.csv

@@ -0,0 +1,8 @@
+# Partitions for PLIP voice firmware — AI-Thinker ESP32-A1S (4 MB flash)
+# Matches PLIP_FIRMWARE/partitions/plip_4mb.csv (OTA dual-bank + SPIFFS data).
+# Name,    Type, SubType,  Offset,   Size,    Flags
+nvs,       data, nvs,      0x9000,   0x5000,
+otadata,   data, ota,      0xe000,   0x2000,
+app0,      app,  ota_0,    0x10000,  0x180000,
+app1,      app,  ota_1,    0x190000, 0x180000,
+storage,   data, spiffs,   0x310000, 0xF0000,

plip_voice/sdkconfig.defaults

@@ -0,0 +1,61 @@
+# Target: ESP32 (AI-Thinker ESP32-A1S, 4MB flash, 8MB PSRAM N4R8)
+CONFIG_IDF_TARGET="esp32"
+
+# Flash: 4MB DIO (A1S default)
+CONFIG_ESPTOOLPY_FLASHSIZE_4MB=y
+CONFIG_ESPTOOLPY_FLASHMODE_DIO=y
+CONFIG_ESPTOOLPY_FLASHFREQ_40M=y
+
+# Partition table
+CONFIG_PARTITION_TABLE_CUSTOM=y
+CONFIG_PARTITION_TABLE_CUSTOM_FILENAME="partitions.csv"
+
+# WiFi
+CONFIG_ESP_WIFI_STATIC_RX_BUFFER_NUM=10
+CONFIG_ESP_WIFI_DYNAMIC_RX_BUFFER_NUM=32
+CONFIG_ESP_WIFI_DYNAMIC_TX_BUFFER_NUM=32
+CONFIG_ESP_WIFI_NVS_ENABLED=y
+CONFIG_LWIP_LOCAL_HOSTNAME="plip"
+
+# FreeRTOS
+CONFIG_FREERTOS_HZ=1000
+
+# Logging
+CONFIG_LOG_DEFAULT_LEVEL_INFO=y
+CONFIG_LOG_COLORS=y
+
+# HTTPD
+CONFIG_HTTPD_MAX_REQ_HDR_LEN=1024
+CONFIG_HTTPD_MAX_URI_LEN=512
+
+# Watchdog — allow for codec init time
+CONFIG_ESP_TASK_WDT_TIMEOUT_S=15
+
+# FreeRTOS IDLE task stack — increase from default 1536 to avoid overflow
+# when I2S/WiFi interrupts spill into idle stack on original ESP32.
+CONFIG_FREERTOS_IDLE_TASK_STACKSIZE=2048
+
+# SLIC hook GPIO — SHK on GPIO23 (A1S KEY4, active-HIGH: HIGH = off-hook)
+CONFIG_PLIP_HOOK_GPIO=23
+CONFIG_PLIP_HOOK_ACTIVE_HIGH=y
+
+# WiFi credentials — set via `idf.py menuconfig` (PLIP Voice Configuration)
+# or override locally in sdkconfig (NOT committed).
+# CONFIG_PLIP_WIFI_SSID="..."
+# CONFIG_PLIP_WIFI_PASSWORD="..."
+# CONFIG_PLIP_WIFI_CHANNEL=11
+
+# Zacus master URL — default matches the lab IP
+# CONFIG_PLIP_MASTER_URL="http://192.168.0.188"
+
+# NPC gateway (Stage 2) — override in sdkconfig (NOT committed) for local testing.
+# Example: CONFIG_PLIP_GATEWAY_URL="http://192.168.0.175:8401"
+#          CONFIG_PLIP_GATEWAY_TOKEN="testtoken"
+# Default (Kconfig): http://192.168.0.50:8401 with empty token.
+
+# Long file names on the SD (voice sample pack uses >8.3 names)
+CONFIG_FATFS_LFN_HEAP=y
+CONFIG_FATFS_MAX_LFN=255
+
+# Speaker at max (earpiece)
+CONFIG_PLIP_SPEAKER_VOLUME=100