fix: restore Discord voice replies

CHANGELOG.md

@@ -18,6 +18,7 @@ Docs: https://docs.openclaw.ai
 
 ### Fixes
 
+- Discord/voice: run voice-channel turns under a voice-output policy that hides the agent `tts` tool and asks for spoken reply text, so `/vc join` sessions synthesize and play agent replies instead of ending with `NO_REPLY`. Fixes #61536. Thanks @aounakram.
 - Plugins/runtime-deps: prune legacy version-scoped plugin runtime-deps roots during bundled dependency repair and cover the path in Package Acceptance's upgrade-survivor matrix, so upgrades from 2026.4.x no longer leave stale per-plugin runtime trees after doctor runs. Thanks @vincentkoc.
 - Plugins/runtime-deps: keep Gateway startup plugin imports and runtime plugin fallback loads verify-only after startup/config repair planning, so packaged installs no longer spawn package-manager repair from hot paths after readiness. Refs #75283 and #75069. Thanks @brokemac79 and @xiaohuaxi.
 - Voice Call/realtime: add default-off fast memory/session context for `openclaw_agent_consult`, giving live calls a bounded answer-or-miss path before the full agent consult. Fixes #71849. Thanks @amzzzzzzz.

docs/channels/discord.md

@@ -1075,7 +1075,7 @@ Voice channel pipeline:
 
 - Discord PCM capture is converted to a WAV temp file.
 - `tools.media.audio` handles STT, for example `openai/gpt-4o-mini-transcribe`.
-- The transcript is sent through normal Discord ingress and routing.
+- The transcript is sent through Discord ingress and routing while the response LLM runs with a voice-output policy that hides the agent `tts` tool and asks for returned text, because Discord voice owns final TTS playback.
 - `voice.model`, when set, overrides only the response LLM for this voice-channel turn.
 - `voice.tts` is merged over `messages.tts`; the resulting audio is played in the joined channel.
 

extensions/discord/src/voice/manager.e2e.test.ts

@@ -539,6 +539,41 @@ describe("DiscordVoiceManager", () => {
     expect(commandArgs?.model).toBe("openai/gpt-5.4-mini");
   });
 
+  it("runs voice replies under Discord voice output policy", async () => {
+    agentCommandMock.mockResolvedValueOnce({
+      payloads: [{ text: "hello back" }],
+    } as never);
+
+    const client = createClient();
+    client.fetchMember.mockResolvedValue({
+      nickname: "Guest Nick",
+      user: {
+        id: "u-guest",
+        username: "guest",
+        globalName: "Guest",
+        discriminator: "4321",
+      },
+    });
+    const manager = createManager({ groupPolicy: "open" }, client, {
+      commands: { useAccessGroups: false },
+    });
+    await processVoiceSegment(manager, "u-guest");
+
+    const commandArgs = agentCommandMock.mock.calls.at(-1)?.[0] as
+      | { message?: string; messageChannel?: string; messageProvider?: string }
+      | undefined;
+
+    expect(commandArgs?.messageChannel).toBe("discord");
+    expect(commandArgs?.messageProvider).toBe("discord-voice");
+    expect(commandArgs?.message).toContain("Do not call the tts tool");
+    expect(textToSpeechMock).toHaveBeenCalledWith(
+      expect.objectContaining({
+        channel: "discord",
+        text: "hello back",
+      }),
+    );
+  });
+
   it("reuses speaker context cache for repeated segments from the same speaker", async () => {
     const client = createClient();
     client.fetchMember.mockResolvedValue({

extensions/discord/src/voice/prompt.test.ts

@@ -1,14 +1,16 @@
 import { describe, expect, it } from "vitest";
-import { formatVoiceIngressPrompt } from "./prompt.js";
+import { DISCORD_VOICE_SPOKEN_OUTPUT_CONTRACT, formatVoiceIngressPrompt } from "./prompt.js";
 
 describe("formatVoiceIngressPrompt", () => {
-  it("formats speaker-labeled voice input without imperative-looking prefixes", () => {
+  it("formats speaker-labeled voice input with the spoken-output contract", () => {
     expect(formatVoiceIngressPrompt("hello there", "speaker-1")).toBe(
-      'Voice transcript from speaker "speaker-1":\nhello there',
+      `${DISCORD_VOICE_SPOKEN_OUTPUT_CONTRACT}\n\nVoice transcript from speaker "speaker-1":\nhello there`,
     );
   });
 
-  it("returns the bare transcript when no speaker label exists", () => {
-    expect(formatVoiceIngressPrompt("hello there")).toBe("hello there");
+  it("keeps unlabeled transcripts under the spoken-output contract", () => {
+    expect(formatVoiceIngressPrompt("hello there")).toBe(
+      `${DISCORD_VOICE_SPOKEN_OUTPUT_CONTRACT}\n\nhello there`,
+    );
   });
 });

extensions/discord/src/voice/prompt.ts

@@ -1,8 +1,17 @@
+export const DISCORD_VOICE_SPOKEN_OUTPUT_CONTRACT = [
+  "Discord voice reply requirements:",
+  "- Return only the concise text that should be spoken aloud in the voice channel.",
+  "- Do not call the tts tool; Discord voice will synthesize and play the returned text.",
+  "- Do not reply with NO_REPLY unless no spoken response is appropriate.",
+  "- Keep the response brief and conversational.",
+].join("\n");
+
 export function formatVoiceIngressPrompt(transcript: string, speakerLabel?: string): string {
   const cleanedTranscript = transcript.trim();
   const cleanedLabel = speakerLabel?.trim();
-  if (!cleanedLabel) {
-    return cleanedTranscript;
-  }
-  return [`Voice transcript from speaker "${cleanedLabel}":`, cleanedTranscript].join("\n");
+  const voiceInput = cleanedLabel
+    ? [`Voice transcript from speaker "${cleanedLabel}":`, cleanedTranscript].join("\n")
+    : cleanedTranscript;
+
+  return [DISCORD_VOICE_SPOKEN_OUTPUT_CONTRACT, voiceInput].join("\n\n");
 }

extensions/discord/src/voice/segment.ts

@@ -18,6 +18,7 @@ import {
 import type { DiscordVoiceSpeakerContextResolver } from "./speaker-context.js";
 import { synthesizeVoiceReplyAudio, transcribeVoiceAudio } from "./tts.js";
 
+const DISCORD_VOICE_MESSAGE_PROVIDER = "discord-voice";
 const logger = createSubsystemLogger("discord/voice");
 
 export async function processDiscordVoiceSegment(params: {
@@ -89,6 +90,7 @@ export async function processDiscordVoiceSegment(params: {
       sessionKey: entry.route.sessionKey,
       agentId: entry.route.agentId,
       messageChannel: "discord",
+      messageProvider: DISCORD_VOICE_MESSAGE_PROVIDER,
       senderIsOwner: speaker.senderIsOwner,
       allowModelOverride: Boolean(modelOverride),
       model: modelOverride,

src/agents/command/attempt-execution.cli.test.ts

@@ -424,7 +424,7 @@ describe("CLI attempt execution", () => {
     });
   });
 
-  it("forwards user trigger and channel context to CLI runs", async () => {
+  it("forwards separate user trigger, channel, and provider context to CLI runs", async () => {
     const sessionKey = "agent:main:direct:claude-channel-context";
     const sessionEntry: SessionEntry = {
       sessionId: "openclaw-session-channel",
@@ -450,10 +450,13 @@ describe("CLI attempt execution", () => {
       resolvedThinkLevel: "medium",
       timeoutMs: 1_000,
       runId: "run-cli-channel-context",
-      opts: { senderIsOwner: false } as Parameters<typeof runAgentAttempt>[0]["opts"],
+      opts: {
+        senderIsOwner: false,
+        messageProvider: "discord-voice",
+      } as Parameters<typeof runAgentAttempt>[0]["opts"],
       runContext: {} as Parameters<typeof runAgentAttempt>[0]["runContext"],
       spawnedBy: undefined,
-      messageChannel: "telegram",
+      messageChannel: "discord",
       skillsSnapshot: undefined,
       resolvedVerboseLevel: undefined,
       agentDir: tmpDir,
@@ -468,8 +471,8 @@ describe("CLI attempt execution", () => {
     expect(runCliAgentMock).toHaveBeenCalledWith(
       expect.objectContaining({
         trigger: "user",
-        messageChannel: "telegram",
-        messageProvider: "telegram",
+        messageChannel: "discord",
+        messageProvider: "discord-voice",
       }),
     );
   });
@@ -567,6 +570,7 @@ describe("CLI attempt execution", () => {
         senderIsOwner: false,
         modelRun: true,
         promptMode: "none",
+        messageProvider: "discord-voice",
         inputProvenance: {
           kind: "inter_session",
           sourceSessionKey: "agent:main:discord:source",
@@ -575,7 +579,7 @@ describe("CLI attempt execution", () => {
       } as Parameters<typeof runAgentAttempt>[0]["opts"],
       runContext: {} as Parameters<typeof runAgentAttempt>[0]["runContext"],
       spawnedBy: undefined,
-      messageChannel: "telegram",
+      messageChannel: "discord",
       skillsSnapshot: undefined,
       resolvedVerboseLevel: undefined,
       agentDir: tmpDir,
@@ -593,6 +597,8 @@ describe("CLI attempt execution", () => {
         model: "claude-opus-4-7",
         agentHarnessId: "pi",
         prompt: "raw prompt",
+        messageChannel: "discord",
+        messageProvider: "discord-voice",
         modelRun: true,
         promptMode: "none",
         disableTools: true,

src/agents/command/attempt-execution.ts

@@ -481,7 +481,7 @@ export function runAgentAttempt(params: {
         skillsSnapshot: params.skillsSnapshot,
         messageChannel: params.messageChannel,
         streamParams: params.opts.streamParams,
-        messageProvider: params.messageChannel,
+        messageProvider: params.opts.messageProvider ?? params.messageChannel,
         agentAccountId: params.runContext.accountId,
         senderIsOwner: params.opts.senderIsOwner,
         cleanupBundleMcpOnRunEnd: params.opts.cleanupBundleMcpOnRunEnd,
@@ -550,6 +550,7 @@ export function runAgentAttempt(params: {
     agentId: params.sessionAgentId,
     trigger: "user",
     messageChannel: params.messageChannel,
+    messageProvider: params.opts.messageProvider ?? params.messageChannel,
     agentAccountId: params.runContext.accountId,
     messageTo: params.opts.replyTo ?? params.opts.to,
     messageThreadId: params.opts.threadId,

src/agents/command/types.ts

@@ -71,6 +71,8 @@ export type AgentCommandOpts = {
   threadId?: string | number;
   /** Message channel context. */
   messageChannel?: string;
+  /** Tool-policy/output surface context. Defaults to messageChannel. */
+  messageProvider?: string;
   /** Delivery channel. */
   channel?: string;
   /** Account ID for multi-account channel routing. */

src/agents/pi-embedded-runner/run/attempt.test.ts

@@ -22,6 +22,7 @@ import {
   resolveEmbeddedAgentStreamFn,
   resolveUnknownToolGuardThreshold,
   shouldCreateBundleMcpRuntimeForAttempt,
+  resolveAttemptToolPolicyMessageProvider,
   resolvePromptBuildHookResult,
   resolvePromptModeForSession,
   shouldStripBootstrapFromEmbeddedContext,
@@ -195,6 +196,21 @@ describe("shouldCreateBundleMcpRuntimeForAttempt", () => {
   });
 });
 
+describe("resolveAttemptToolPolicyMessageProvider", () => {
+  it("prefers explicit tool-policy provider over transport channel", () => {
+    expect(
+      resolveAttemptToolPolicyMessageProvider({
+        messageChannel: "discord",
+        messageProvider: "discord-voice",
+      }),
+    ).toBe("discord-voice");
+  });
+
+  it("falls back to message channel when provider is omitted", () => {
+    expect(resolveAttemptToolPolicyMessageProvider({ messageChannel: "discord" })).toBe("discord");
+  });
+});
+
 describe("resolvePromptBuildHookResult", () => {
   function createLegacyOnlyHookRunner() {
     return {

src/agents/pi-embedded-runner/run/attempt.ts

@@ -566,6 +566,13 @@ export function shouldCreateBundleMcpRuntimeForAttempt(params: {
   );
 }
 
+export function resolveAttemptToolPolicyMessageProvider(params: {
+  messageProvider?: string;
+  messageChannel?: string;
+}): string | undefined {
+  return params.messageProvider ?? params.messageChannel;
+}
+
 function collectAttemptExplicitToolAllowlistSources(params: {
   config?: EmbeddedRunAttemptParams["config"];
   sessionKey?: string;
@@ -784,7 +791,7 @@ export async function runEmbeddedAttempt(
                 elevated: params.bashElevated,
               },
               sandbox,
-              messageProvider: params.messageChannel ?? params.messageProvider,
+              messageProvider: resolveAttemptToolPolicyMessageProvider(params),
               agentAccountId: params.agentAccountId,
               messageTo: params.messageTo,
               messageThreadId: params.messageThreadId,
@@ -1003,7 +1010,7 @@ export async function runEmbeddedAttempt(
       agentId: sessionAgentId,
       modelProvider: params.provider,
       modelId: params.modelId,
-      messageProvider: params.messageChannel ?? params.messageProvider,
+      messageProvider: resolveAttemptToolPolicyMessageProvider(params),
       agentAccountId: params.agentAccountId,
       groupId: params.groupId,
       groupChannel: params.groupChannel,
@@ -1030,7 +1037,7 @@ export async function runEmbeddedAttempt(
       agentId: sessionAgentId,
       modelProvider: params.provider,
       modelId: params.modelId,
-      messageProvider: params.messageChannel ?? params.messageProvider,
+      messageProvider: resolveAttemptToolPolicyMessageProvider(params),
       agentAccountId: params.agentAccountId,
       groupId: params.groupId,
       groupChannel: params.groupChannel,

src/agents/pi-tools.message-provider-policy.test.ts

@@ -4,7 +4,7 @@ import { filterToolNamesByMessageProvider } from "./pi-tools.message-provider-po
 const DEFAULT_TOOL_NAMES = ["read", "write", "tts", "web_search"];
 
 describe("createOpenClawCodingTools message provider policy", () => {
-  it.each(["voice", "VOICE", " Voice "])(
+  it.each(["voice", "VOICE", " Voice ", "discord-voice", "DISCORD-VOICE", " Discord-Voice "])(
     "does not expose tts tool for normalized voice provider: %s",
     (messageProvider) => {
       const names = new Set(filterToolNamesByMessageProvider(DEFAULT_TOOL_NAMES, messageProvider));

src/agents/pi-tools.message-provider-policy.ts

@@ -1,6 +1,7 @@
 import { normalizeOptionalLowercaseString } from "../shared/string-coerce.js";
 
 const TOOL_DENY_BY_MESSAGE_PROVIDER: Readonly<Record<string, readonly string[]>> = {
+  "discord-voice": ["tts"],
   voice: ["tts"],
 };