@tarfa/cli 0.1.7 → 0.1.8

package/dist/{chunk-4M2ZRGQR.js → chunk-BHXNKH3U.js}

@@ -5526,6 +5526,9 @@ var clientMessageSchema = z5.discriminatedUnion("type", [
   z5.object({ type: z5.literal("voiceStart"), workspaceId: z5.string(), id: z5.string() }),
   z5.object({ type: z5.literal("voiceChunk"), id: z5.string(), data: z5.string() }),
   z5.object({ type: z5.literal("voiceEnd"), id: z5.string() }),
+  // Throw the dictation away: Claude writes nothing. Sent only to an agent whose authResult names
+  // the `voiceCancel` feature (an older one would drop it, and Claude would type what it heard).
+  z5.object({ type: z5.literal("voiceCancel"), id: z5.string() }),
   // The text of Claude's last reply in a workspace, read aloud on the phone. Sent only to an agent
   // whose authResult names the `lastReply` feature.
   z5.object({ type: z5.literal("lastReply"), workspaceId: z5.string() }),
@@ -5554,7 +5557,8 @@ var AGENT_FEATURES = {
   lastReply: "lastReply",
   dictationKeys: "dictationKeys",
   setDictationKey: "setDictationKey",
-  addresses: "addresses"
+  addresses: "addresses",
+  voiceCancel: "voiceCancel"
 };
 var agentMessageSchema = z5.discriminatedUnion("type", [
   // `credential` is what authenticated the session (`token` only from an agent from before token

package/dist/cli.js

@@ -73,7 +73,7 @@ import {
   withEnrolmentMode,
   withLegacyEnv,
   workspaceLink
-} from "./chunk-4M2ZRGQR.js";
+} from "./chunk-BHXNKH3U.js";
 
 // src/cli.ts
 import { randomBytes } from "node:crypto";

package/dist/serve.js

@@ -78,7 +78,7 @@ import {
   systemPathResolver,
   withLegacyEnv,
   writeFileAtomic
-} from "./chunk-4M2ZRGQR.js";
+} from "./chunk-BHXNKH3U.js";
 
 // src/serve.ts
 import { homedir as homedir2 } from "node:os";
@@ -91,6 +91,21 @@ var TAIL_MS = 500;
 var IDLE_MS = 5e3;
 var MAX_MS = 5 * 6e4;
 var DRAIN_MS = LEAD_MS + 1e4;
+var CANCEL_POLL_MS = 50;
+var CANCEL_WAIT_MS = 1500;
+var NOT_LISTENING = "Dictation not cancelled: Claude was not listening";
+var CANCEL_KEY = "";
+function claudeVoiceListening(screen) {
+  const lines = screen.split("\n");
+  let prompt = -1;
+  for (let i = lines.length - 1; i >= 0; i--) {
+    if (lines[i].startsWith("\u276F")) {
+      prompt = i;
+      break;
+    }
+  }
+  return prompt >= 2 && lines[prompt - 2].trim() === "listening\u2026";
+}
 function voiceSettingsProblem(settings) {
   const s = typeof settings === "object" && settings !== null ? settings : {};
   const enabled = s.voice?.enabled ?? s.voiceEnabled;
@@ -102,10 +117,11 @@ function voiceSettingsProblem(settings) {
 }
 var errorText = (e) => e instanceof Error ? e.message : String(e);
 var Dictation = class {
-  constructor(deps, pressSpace, release) {
+  constructor(deps, pressSpace, release, pane) {
     this.deps = deps;
     this.pressSpace = pressSpace;
     this.release = release;
+    this.pane = pane;
     this.done = new Promise((resolve) => this.resolve = resolve);
     this.max = setTimeout(() => this.abort("Dictation stopped: it ran past 5 minutes"), MAX_MS);
     this.touch();
@@ -114,11 +130,13 @@ var Dictation = class {
   deps;
   pressSpace;
   release;
+  pane;
   done;
   resolve;
   queue = [];
   stream = null;
   ended = false;
+  cancelling = false;
   finished = false;
   hold;
   lead;
@@ -132,7 +150,7 @@ var Dictation = class {
     else this.queue.push(pcm);
   }
   end() {
-    if (this.finished || this.ended) return;
+    if (this.finished || this.ended || this.cancelling) return;
     this.ended = true;
     clearTimeout(this.idle);
     if (this.stream) this.drain(this.stream);
@@ -141,9 +159,39 @@ var Dictation = class {
     this.stream?.abort();
     this.finish(reason);
   }
+  cancel() {
+    if (this.finished || this.ended || this.cancelling) return;
+    this.cancelling = true;
+    clearTimeout(this.lead);
+    clearTimeout(this.idle);
+    this.queue = [];
+    this.stream?.abort();
+    this.stream = null;
+    if (!this.hold) return this.finish(null);
+    void this.cancelOnceListening(Date.now());
+  }
+  // The hold goes on until Claude shows it records, so the key lands while it still listens.
+  async cancelOnceListening(since) {
+    const pane = this.pane;
+    if (!pane) return this.finish(NOT_LISTENING);
+    while (!this.finished) {
+      const listening = await pane.listening().catch(() => false);
+      if (this.finished) return;
+      if (listening) {
+        clearInterval(this.hold);
+        try {
+          pane.pressCancel();
+        } catch {
+        }
+        return this.finish(null);
+      }
+      if (Date.now() - since >= CANCEL_WAIT_MS) return this.finish(NOT_LISTENING);
+      await new Promise((resolve) => setTimeout(resolve, CANCEL_POLL_MS));
+    }
+  }
   async begin() {
     const problem = voiceSettingsProblem(await this.deps.settings.read().catch(() => null));
-    if (this.finished) return;
+    if (this.finished || this.cancelling) return;
     if (problem) return this.finish(problem);
     const press = () => {
       try {
@@ -156,7 +204,7 @@ var Dictation = class {
     this.lead = setTimeout(() => this.play(), LEAD_MS);
   }
   play() {
-    if (this.finished) return;
+    if (this.finished || this.cancelling) return;
     let stream;
     try {
       stream = this.deps.player.open();
@@ -164,7 +212,9 @@ var Dictation = class {
       return this.finish(errorText(e));
     }
     this.stream = stream;
-    stream.played.catch((e) => this.finish(errorText(e)));
+    stream.played.catch((e) => {
+      if (!this.cancelling) this.finish(errorText(e));
+    });
     for (const pcm of this.queue.splice(0)) stream.write(pcm);
     if (this.ended) this.drain(stream);
   }
@@ -201,13 +251,13 @@ var ClaudeVoice = class {
   available() {
     return this.deps.player.available();
   }
-  start(pressSpace) {
+  start(pressSpace, pane) {
     if (this.busy) {
       const done = Promise.resolve("Claude voice is already in use on this machine");
-      return { write: () => void 0, end: () => void 0, abort: () => void 0, done };
+      return { write: () => void 0, end: () => void 0, abort: () => void 0, cancel: () => void 0, done };
     }
     this.busy = true;
-    return new Dictation(this.deps, pressSpace, () => this.busy = false);
+    return new Dictation(this.deps, pressSpace, () => this.busy = false, pane);
   }
 };
 
@@ -1288,7 +1338,7 @@ var PRE_AUTH_MAX_MESSAGE_BYTES = 8 * 1024;
 var MAX_AUTH_ATTEMPTS = 3;
 var MAX_HELD_MESSAGES = 16;
 var PAIRING_SECRET = /^[A-Za-z0-9_-]{43}$/;
-var FEATURES = [AGENT_FEATURES.lastReply, AGENT_FEATURES.dictationKeys, AGENT_FEATURES.setDictationKey];
+var FEATURES = [AGENT_FEATURES.lastReply, AGENT_FEATURES.dictationKeys, AGENT_FEATURES.setDictationKey, AGENT_FEATURES.voiceCancel];
 var NoiseClientConnection = class {
   constructor(socket, staticPrivate, options) {
     this.socket = socket;
@@ -2517,7 +2567,7 @@ var ConnectionHandler = class {
       case "voiceStart": {
         const { workspaceId, id } = message;
         this.cancelVoice("Superseded by a newer dictation");
-        const voice = { id, workspaceId, session: null, queue: [], ended: false, cancelled: null };
+        const voice = { id, workspaceId, session: null, queue: [], ended: false, cancelled: null, discarded: false };
         this.voice = voice;
         const reply = (error) => {
           if (this.voice === voice) this.voice = null;
@@ -2528,9 +2578,15 @@ var ConnectionHandler = class {
         const tmuxSession = (await this.deps.tmux.listSessions(this.deps.config.sessionPrefix)).find((s) => s.name === workspaceId);
         if (!tmuxSession || !runsClaude(tmuxSession.command)) return reply("Claude voice works only in a Claude Code workspace");
         if (tmuxSession.program?.state === "blocked") return reply("Answer Claude\u2019s question first, then dictate");
+        if (voice.discarded) return reply(null);
         if (voice.cancelled) return reply(voice.cancelled);
-        const session = this.deps.claudeVoice.start(() => {
-          if (this.attachment?.workspaceId === workspaceId) this.attachment.session.write(" ");
+        const typeInto = (data) => {
+          if (this.attachment?.workspaceId === workspaceId) this.attachment.session.write(data);
+        };
+        const tmux2 = this.deps.tmux;
+        const session = this.deps.claudeVoice.start(() => typeInto(" "), {
+          listening: async () => !!tmux2.capturePane && claudeVoiceListening(await tmux2.capturePane(workspaceId, 0)),
+          pressCancel: () => typeInto(CANCEL_KEY)
         });
         voice.session = session;
         for (const pcm of voice.queue.splice(0)) session.write(pcm);
@@ -2546,9 +2602,17 @@ var ConnectionHandler = class {
         else voice.queue.push(pcm);
         return;
       }
+      case "voiceCancel": {
+        const voice = this.voice;
+        if (voice?.id !== message.id || voice.ended) return;
+        voice.ended = true;
+        voice.discarded = true;
+        voice.session?.cancel();
+        return;
+      }
       case "voiceEnd": {
         const voice = this.voice;
-        if (voice?.id !== message.id) return;
+        if (voice?.id !== message.id || voice.discarded) return;
         voice.ended = true;
         voice.session?.end();
         return;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tarfa/cli",
-  "version": "0.1.7",
+  "version": "0.1.8",
   "description": "Tarfa agent and CLI: serves your Claude Code terminal sessions (tmux workspaces) to the Tarfa phone app.",
   "repository": {
     "type": "git",