feat(plip): add DELETE /game/file (SD/SPIFFS)

Removes a staged file with the same sd/ vs SPIFFS path routing as the
POST handler (mounts SD on demand). Lets the host clean stale clips
when regenerating the SD voice pack. max_uri_handlers 17->18.
Validated: probe WAVs removed from /sdcard/voice (200, then 404).

idf_zacus/components/game_endpoint/CMakeLists.txt

@@ -22,4 +22,5 @@ idf_component_register(
         p7_coffre
         p5_morse
         p6_nfc
+        npc_engine
 )

idf_zacus/components/game_endpoint/game_endpoint.c

@@ -27,6 +27,7 @@
 #include "nvs_flash.h"
 
 #include "hints_client.h"
+#include "npc_engine.h"
 #include "scenario_mesh.h"
 #include "puzzle_binding.h"
 #include "local_puzzles.h"
@@ -807,6 +808,16 @@ 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) {

idf_zacus/components/game_endpoint/include/puzzle_binding.h

@@ -55,6 +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)
@@ -65,6 +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

@@ -239,9 +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/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.

idf_zacus/components/npc_engine/npc_engine.c

@@ -440,6 +440,19 @@ esp_err_t npc_engine_set_step(uint8_t step_id, uint32_t expected_duration_ms) {
     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

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