feat: gamebook mode (CYOA) on the master

Context:
We wanted a "livre dont vous etes le heros" playable on the Freenove
master with NO model in RAM: narration as say()-rendered WAV staged on
the SD card, and navigation on the physical 5-way pad. Nothing in the
firmware drove a branching, button-navigated story before this.

Approach:
A self-contained gamebook component reads a JSON book from the SD,
drives the display and audio, and owns the 5-way pad through a
registered display_ui key hook (so display_ui keeps no dependency on
gamebook — no cycle). A dedicated full-screen gamebook view renders
each page (title / wrapped body / choice list).

Changes:
- New gamebook component: loads /sdcard/gamebook/gamebook.json; per
  passage plays its WAV (media_manager, absolute /sdcard path) and
  renders the title, wrapped text, and a choice list with a ">" cursor.
- Navigation: D-pad up/down (and left/right) move the cursor, the
  center click confirms; endings (no choices) restart on click; a
  choice interrupts the current narration (media_manager_stop + play).
- display_ui: add display_ui_set_key_hook (consume keys when a takeover
  mode is active) and a dedicated gamebook view (title + height-bounded
  body band + reserved bottom menu band) with display_ui_gamebook_show
  / display_ui_gamebook_hide; the gamebook view wins the view selector
  while open.
- game_endpoint: POST /game/gamebook[?action=stop] to start/stop.
- ota_server: bump max_uri_handlers 20 -> 22 for the new endpoint.
- main: call gamebook_init() once the buttons are up.

Impact:
The Freenove plays a branching audio gamebook driven entirely by the
pad, fully offline (no gateway, no model). Hardware-validated end to
end: cursor navigation, click select, WAV narration from SD, endings
restart.

idf_zacus/components/display_ui/display_ui.cpp

@@ -248,6 +248,16 @@ 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 char s_gb_title_buf[64];
+static char s_gb_body_buf[512];
+static char s_gb_menu_buf[160];
 static volatile bool s_browser_reload = false;
 
 // ─── Intro — faithful port of the original cracktro ──────────────────────────
@@ -761,6 +771,44 @@ scroller:
     lv_obj_set_pos(s_intro_scroll, sx, sy);
 }
 
+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);
+
+    s_gb_title = lv_label_create(s_scr_gamebook);
+    lv_obj_set_style_text_color(s_gb_title, lv_color_hex(COL_VALUE), 0);
+    lv_obj_set_style_text_font(s_gb_title, &lv_font_montserrat_24, 0);
+    lv_label_set_long_mode(s_gb_title, LV_LABEL_LONG_DOT);
+    lv_obj_set_width(s_gb_title, DUI_HOR_RES - 16);
+    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, 6);
+
+    // Body band: top area, height-bounded so it can never run into the menu.
+    s_gb_body = lv_label_create(s_scr_gamebook);
+    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_size(s_gb_body, DUI_HOR_RES - 24, 196);
+    lv_label_set_text(s_gb_body, "");
+    lv_obj_align(s_gb_body, LV_ALIGN_TOP_LEFT, 12, 44);
+
+    // Menu band: reserved strip at the bottom, drawn last (foreground), with a
+    // faint background so the choices stand out from the body text.
+    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(0x101820), 0);
+    lv_obj_set_style_bg_opa(s_gb_menu, LV_OPA_COVER, 0);
+    lv_obj_set_style_pad_all(s_gb_menu, 6, 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_WRAP);
+    lv_obj_set_width(s_gb_menu, DUI_HOR_RES);
+    lv_label_set_text(s_gb_menu, "");
+    lv_obj_align(s_gb_menu, LV_ALIGN_BOTTOM_LEFT, 0, 0);
+    lv_obj_move_foreground(s_gb_menu);
+}
+
 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 +901,19 @@ 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.
+    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);
+    }
+
     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_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 +1021,7 @@ static void display_task(void *arg) {
     build_status_screen();
     build_scene_screen();
     build_shell_screen();
+    build_gamebook_screen();
     build_browser_screen();
     build_intro_screen();
 
@@ -1145,9 +1203,36 @@ 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) {
+    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  : "");
+    s_gamebook_open = true;
+    s_dirty = true;
+    if (s_mutex) xSemaphoreGive(s_mutex);
+}
+
+extern "C" void display_ui_gamebook_hide(void) {
+    s_gamebook_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 +78,36 @@ 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);
+
+/** @brief Leave the gamebook view and return to the normal scene/status flow. */
+void display_ui_gamebook_hide(void);
+
 /**
  * @brief Pop the pending shell-app launch request, if any.
  *

idf_zacus/components/game_endpoint/CMakeLists.txt

@@ -17,6 +17,7 @@ idf_component_register(
         puzzle_state
         media_manager
         sd_storage
+        gamebook
     PRIV_REQUIRES
         local_puzzles
         p7_coffre

idf_zacus/components/game_endpoint/game_endpoint.c

@@ -27,6 +27,7 @@
 #include "nvs_flash.h"
 
 #include "hints_client.h"
+#include "gamebook.h"
 #include "npc_engine.h"
 #include "scenario_mesh.h"
 #include "puzzle_binding.h"
@@ -994,6 +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"}.
@@ -1232,6 +1256,12 @@ 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
@@ -1308,6 +1338,10 @@ 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/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 +1,188 @@
+// gamebook.c — see gamebook.h. Standalone "livre dont vous êtes le héros".
+#include "gamebook.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "cJSON.h"
+#include "esp_log.h"
+
+#include "display_ui.h"
+#include "media_manager.h"
+
+static const char *TAG = "gamebook";
+
+#define GAMEBOOK_DIR   "/sdcard/gamebook"
+#define GAMEBOOK_JSON  GAMEBOOK_DIR "/gamebook.json"
+#define GAMEBOOK_MAX   (32 * 1024)   /* JSON sanity cap */
+
+/* 5-way pad codes (mirror display_ui): 1=SELECT 2=DOWN 3=MENU 4=L/R 5=UP. */
+#define KEY_MENU 3
+
+static cJSON   *s_root      = NULL;   /* owns the parsed JSON while active */
+static cJSON   *s_passages  = NULL;   /* borrowed: s_root->"passages"      */
+static char     s_title[48] = {0};
+static char     s_current[48] = {0};
+static int      s_sel       = 0;      /* highlighted choice index           */
+static volatile bool s_active = false;
+
+bool gamebook_active(void) { return s_active; }
+
+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: title + wrapped text + choice list with a "> "
+ * cursor on s_sel. Does NOT touch audio — call on every cursor move. */
+static void render_page(void)
+{
+    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[256];
+    if (n == 0) {
+        snprintf(menu, sizeof(menu), "~ Fin ~   (clic = recommencer)");
+    } else {
+        size_t off = 0;
+        for (int i = 0; i < n && off < sizeof(menu); i++) {
+            const cJSON *lbl = cJSON_GetObjectItem(cJSON_GetArrayItem(choices, i),
+                                                   "label");
+            off += snprintf(menu + off, sizeof(menu) - off, "%s%s\n",
+                            i == s_sel ? "> " : "  ",
+                            cJSON_IsString(lbl) ? lbl->valuestring : "?");
+        }
+    }
+    display_ui_gamebook_show(
+        cJSON_IsString(screen) ? screen->valuestring : s_title,
+        cJSON_IsString(text)   ? text->valuestring   : "",
+        menu);
+}
+
+/* Enter a passage: reset the cursor, draw the page, play its WAV from the SD. */
+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();
+
+    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());
+}
+
+/* Key hook (we own the 5-way pad while active). D-pad navigates the choice
+ * list, click (SELECT) confirms. Ladder codes: 1=click 2=down 3/4=left/right
+ * 5=up. up/left → previous, down/right → next. On an ending (no choices) the
+ * click restarts the book. Returns true to consume the key. */
+static bool gamebook_key_hook(uint8_t key)
+{
+    if (!s_active) return false;
+
+    int n = cur_choice_count();
+    if (n == 0) {                                  /* ending page */
+        if (key == 1) {                            /* click → restart */
+            const cJSON *start = cJSON_GetObjectItem(s_root, "start");
+            if (cJSON_IsString(start)) enter_passage(start->valuestring);
+        }
+        return true;
+    }
+
+    switch (key) {
+    case 5: case 3:                                /* up / left → previous */
+        s_sel = (s_sel - 1 + n) % n; render_page(); break;
+    case 2: case 4:                                /* down / right → next  */
+        s_sel = (s_sel + 1) % n; render_page(); break;
+    case 1: {                                      /* click → confirm */
+        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, "select #%d -> '%s'", s_sel, g->valuestring);
+            enter_passage(g->valuestring);
+        }
+        break;
+    }
+    default: break;
+    }
+    return true;
+}
+
+esp_err_t gamebook_start(void)
+{
+    gamebook_stop();   /* clean any previous run */
+
+    FILE *f = fopen(GAMEBOOK_JSON, "rb");
+    if (!f) {
+        ESP_LOGW(TAG, "open %s failed — SD card / pack missing?", GAMEBOOK_JSON);
+        return ESP_ERR_NOT_FOUND;
+    }
+    fseek(f, 0, SEEK_END);
+    long sz = ftell(f);
+    rewind(f);
+    if (sz <= 0 || sz > GAMEBOOK_MAX) { fclose(f); return ESP_ERR_INVALID_SIZE; }
+    char *buf = malloc((size_t)sz + 1);
+    if (!buf) { fclose(f); return ESP_ERR_NO_MEM; }
+    size_t rd = fread(buf, 1, (size_t)sz, f);
+    fclose(f);
+    buf[rd] = '\0';
+
+    s_root = cJSON_Parse(buf);
+    free(buf);
+    if (!s_root) { ESP_LOGW(TAG, "malformed gamebook.json"); return ESP_ERR_INVALID_ARG; }
+
+    s_passages = cJSON_GetObjectItem(s_root, "passages");
+    const cJSON *start = cJSON_GetObjectItem(s_root, "start");
+    const cJSON *title = cJSON_GetObjectItem(s_root, "title");
+    if (!cJSON_IsObject(s_passages) || !cJSON_IsString(start)) {
+        ESP_LOGW(TAG, "gamebook.json missing passages/start");
+        cJSON_Delete(s_root); s_root = NULL; s_passages = NULL;
+        return ESP_ERR_INVALID_ARG;
+    }
+    snprintf(s_title, sizeof(s_title), "%s",
+             cJSON_IsString(title) ? title->valuestring : "");
+
+    s_active = true;
+    ESP_LOGI(TAG, "start \"%s\" @ '%s'", s_title, start->valuestring);
+    enter_passage(start->valuestring);
+    return ESP_OK;
+}
+
+void gamebook_stop(void)
+{
+    if (!s_active && !s_root) return;
+    s_active = false;
+    media_manager_stop();
+    display_ui_gamebook_hide();
+    if (s_root) { cJSON_Delete(s_root); s_root = NULL; }
+    s_passages = NULL;
+    s_current[0] = '\0';
+    ESP_LOGI(TAG, "stopped");
+}
+
+void gamebook_init(void)
+{
+    display_ui_set_key_hook(gamebook_key_hook);
+    ESP_LOGI(TAG, "ready (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/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 = 22;   // ota + voice_hook + game (incl. /game/step,
+                                    // /game/puzzle_state, /game/file, relay,
+                                    // /game/gamebook)
                                     // + headroom
     config.uri_match_fn    = httpd_uri_match_wildcard;
     config.stack_size      = 8192;

idf_zacus/main/CMakeLists.txt

@@ -23,6 +23,7 @@ idf_component_register(
         esp_event
         espressif__mdns
         display_ui
+        gamebook
         puzzle_state
         p7_coffre
         p5_morse

idf_zacus/main/main.c

@@ -55,6 +55,7 @@
 #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"
@@ -436,6 +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();

plip_voice/main/audio.c

@@ -137,6 +137,12 @@ static void ensure_sd_mounted(void)
     }
 }
 
+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)

plip_voice/main/audio.h

@@ -39,6 +39,12 @@ 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). */

plip_voice/main/conversation.c

@@ -84,12 +84,16 @@ static char s_sid[32] = {0};
  * 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 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). */
+ * 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;
 
 /* Known numbers: ringback when dialed */
@@ -105,6 +109,20 @@ static bool is_known(const char *num)
     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)
 {
     tones_stop();
@@ -114,6 +132,8 @@ static void go_idle(void)
 #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");
 }
@@ -134,6 +154,7 @@ static void enter_incoming_greet(void)
     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;
@@ -257,18 +278,57 @@ static void conv_task(void *arg)
                 go_idle();
                 break;
             }
-            /* Fetch greeting WAV from gateway and enqueue playback */
-            if (turn_client_greeting(s_sid, s_number,
-                                     "/spiffs/turn.wav")) {
-                audio_play_async("/spiffs/turn.wav");
-            } else {
-                ESP_LOGW(TAG, "turn_client_greeting failed — proceeding silent");
+            /* 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.
@@ -404,12 +464,14 @@ void conversation_on_hook_change(bool offhook)
     s_hook_changed = true;
 }
 
-void conversation_arm_incoming(const char *number)
+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) — ringing until pickup",
-             s_incoming_number);
+    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

@@ -6,5 +6,8 @@ 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. */
-void conversation_arm_incoming(const char *number);
+ * 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/net.c

@@ -22,6 +22,7 @@
 #include <string.h>
 #include <stdio.h>
 #include <errno.h>
+#include <sys/stat.h>
 
 #include "audio.h"
 #include "tones.h"
@@ -53,7 +54,8 @@
 
 #define SPIFFS_LABEL  "storage"
 #define SPIFFS_BASE   "/spiffs"
-#define MAX_BODY      (256 * 1024)
+#define MAX_BODY      (512 * 1024)   /* streamed in 2 KB chunks → RAM-safe; raised
+                                        so full-length say() hint WAVs fit on SD */
 
 static volatile bool    s_connected  = false;
 static httpd_handle_t   s_httpd      = NULL;
@@ -209,14 +211,39 @@ static esp_err_t handle_file_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();
+    /* 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; }
 
-    /* Build full SPIFFS path. */
-    char full[160];
-    if (fpath[0] == '/') {
-        snprintf(full, sizeof(full), "%s%s", SPIFFS_BASE, fpath);
+    if (to_sd) {
+        if (!audio_ensure_sd())
+            return send_json(req, "503 Service Unavailable", "{\"error\":\"no sd card\"}");
+        base = PLIP_SD_MOUNT;   /* "/sdcard" */
     } else {
-        snprintf(full, sizeof(full), "%s/%s", SPIFFS_BASE, fpath);
+        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");
@@ -254,6 +281,44 @@ static esp_err_t handle_file_post(httpd_req_t *req)
     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
@@ -483,10 +548,13 @@ 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[48] = {0}, number[16] = {0};
+    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));
-    conversation_arm_incoming(number);  /* defaults to "17" if empty; starts ringing */
+    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");
@@ -812,7 +880,7 @@ esp_err_t net_init(void)
     /* Start HTTP server. */
     httpd_config_t hcfg = HTTPD_DEFAULT_CONFIG();
     hcfg.server_port      = 80;
-    hcfg.max_uri_handlers = 17;
+    hcfg.max_uri_handlers = 18;
     hcfg.stack_size       = 8192;
 
     esp_err_t ret = httpd_start(&s_httpd, &hcfg);
@@ -833,6 +901,10 @@ esp_err_t net_init(void)
         .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,
@@ -888,6 +960,7 @@ esp_err_t net_init(void)
     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);

plip_voice/sdkconfig.defaults

@@ -52,3 +52,7 @@ CONFIG_PLIP_HOOK_ACTIVE_HIGH=y
 # 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