fix: honor audio_as_voice streaming + parse tests (#490) (thanks @jarvis-medmatic)

CHANGELOG.md

@@ -42,6 +42,7 @@
 - Commands: harden slash command registry and list text-only commands in `/commands`.
 - Models/Auth: show per-agent auth candidates in `/model status`, and add `clawdbot models auth order {get,set,clear}` (per-agent auth rotation overrides). — thanks @steipete
 - Telegram: keep streamMode draft-only; avoid forcing block streaming. (#619) — thanks @rubyrunsstuff
+- Telegram: add `[[audio_as_voice]]` tag support for voice notes with streaming-safe delivery. (#490) — thanks @jarvis-medmatic
 - Debugging: add raw model stream logging flags and document gateway watch mode.
 - Gateway: decode dns-sd escaped UTF-8 in discovery output and show scan progress immediately. — thanks @steipete
 - Agent: add claude-cli/opus-4.5 runner via Claude CLI with resume support (tools disabled).

src/agents/pi-embedded-runner.ts

@@ -776,6 +776,7 @@ export async function compactEmbeddedPiSession(params: {
   const enqueueGlobal =
     params.enqueue ??
     ((task, opts) => enqueueCommandInLane(globalLane, task, opts));
+  const runAbortController = new AbortController();
   return enqueueCommandInLane(sessionLane, () =>
     enqueueGlobal(async () => {
       const resolvedWorkspace = resolveUserPath(params.workspaceDir);

src/agents/pi-tools.ts

@@ -651,7 +651,9 @@ export function createClawdbotCodingTools(options?: {
   // Without this, some providers (notably OpenAI) will reject root-level union schemas.
   const normalized = subagentFiltered.map(normalizeToolParameters);
   const withAbort = options?.abortSignal
-    ? normalized.map((tool) => wrapToolWithAbortSignal(tool, options.abortSignal))
+    ? normalized.map((tool) =>
+        wrapToolWithAbortSignal(tool, options.abortSignal),
+      )
     : normalized;
 
   // Anthropic blocks specific lowercase tool names (bash, read, write, edit) with OAuth tokens.

src/auto-reply/reply/abort.ts

@@ -3,6 +3,7 @@ import type { ClawdbotConfig } from "../../config/config.js";
 import {
   loadSessionStore,
   resolveStorePath,
+  type SessionEntry,
   saveSessionStore,
 } from "../../config/sessions.js";
 import {
@@ -35,7 +36,7 @@ export function setAbortMemory(key: string, value: boolean): void {
 }
 
 function resolveSessionEntryForKey(
-  store: Record<string, { sessionId: string; updatedAt: number }> | undefined,
+  store: Record<string, SessionEntry> | undefined,
   sessionKey: string | undefined,
 ) {
   if (!store || !sessionKey) return {};

src/auto-reply/reply/dispatch-from-config.test.ts

@@ -7,7 +7,10 @@ import type { ReplyDispatcher } from "./reply-dispatcher.js";
 
 const mocks = vi.hoisted(() => ({
   routeReply: vi.fn(async () => ({ ok: true, messageId: "mock" })),
-  tryFastAbortFromMessage: vi.fn(async () => ({ handled: false, aborted: false })),
+  tryFastAbortFromMessage: vi.fn(async () => ({
+    handled: false,
+    aborted: false,
+  })),
 }));
 
 vi.mock("./route-reply.js", () => ({

src/media/parse.test.ts

@@ -0,0 +1,19 @@
+import { describe, expect, it } from "vitest";
+
+import { splitMediaFromOutput } from "./parse.js";
+
+describe("splitMediaFromOutput", () => {
+  it("detects audio_as_voice tag and strips it", () => {
+    const result = splitMediaFromOutput("Hello [[audio_as_voice]] world");
+    expect(result.audioAsVoice).toBe(true);
+    expect(result.text).toBe("Hello world");
+  });
+
+  it("keeps audio_as_voice detection stable across calls", () => {
+    const input = "Hello [[audio_as_voice]]";
+    const first = splitMediaFromOutput(input);
+    const second = splitMediaFromOutput(input);
+    expect(first.audioAsVoice).toBe(true);
+    expect(second.audioAsVoice).toBe(true);
+  });
+});

src/media/parse.ts

@@ -34,6 +34,7 @@ function isInsideFence(
 
 // Regex to detect [[audio_as_voice]] tag
 const AUDIO_AS_VOICE_RE = /\[\[audio_as_voice\]\]/gi;
+const AUDIO_AS_VOICE_TEST_RE = /\[\[audio_as_voice\]\]/i;
 
 export function splitMediaFromOutput(raw: string): {
   text: string;
@@ -123,7 +124,7 @@ export function splitMediaFromOutput(raw: string): {
     .trim();
 
   // Detect and strip [[audio_as_voice]] tag
-  const hasAudioAsVoice = AUDIO_AS_VOICE_RE.test(cleanedText);
+  const hasAudioAsVoice = AUDIO_AS_VOICE_TEST_RE.test(cleanedText);
   if (hasAudioAsVoice) {
     cleanedText = cleanedText
       .replace(AUDIO_AS_VOICE_RE, "")