agentwire-dev 1.29.1__tar.gz → 1.29.2__tar.gz

{agentwire_dev-1.29.1 → agentwire_dev-1.29.2}/CHANGELOG.md

@@ -10,6 +10,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Removed
 
 - **Workflow engine** — entire `agentwire/workflows/` package (DAG runner, definitions, storage, pi runner, CLI), the `agentwire workflow` CLI subcommand, 5 MCP tools (`workflow_list/validate/run/history/show`), 3 portal endpoints (`/api/workflows/list|runs|runs/{id}`), the Workflows sidebar section + WorkflowWindow + ~230 lines of CSS, the scheduler's `workflow:` dispatch branch (`SchedulerTask.workflow`/`inputs` fields, `_dispatch_workflow_task`, `_render_workflow_inputs`, `_parse_workflow_summary`, `_workflow_failure_state`, validation block), 9 workflow test files plus the workflow-specific cases in `test_scheduler.py`, the `agentwire-workflows` skill, `docs/wiki/scheduling/workflows.md`, and ~120 dirs of historical runs under `~/.agentwire/workflows/`. The user's actual recurring automations all live as scheduler ensure tasks (`task:`); the engine was unused in practice and the empty sidebar surface was just noise. Scheduler is now ensure-only.
+- **On-hold mission docs** — retroactive accounting for the seven on-hold files left behind by #153 and wholesale-deleted in commit `2425847` (2026-05-13) as part of the GitHub-issues-as-SSOT migration: `docs/missions/OBJECTIVE.md` (folded into README + website), `docs/missions/pi-harness-overview.md` (roadmap superseded — phases shipped or moved to research wiki), `docs/missions/pi-workflow-advanced.md` (on-hold-awaiting-usage stance has no scoped plan to track), `docs/missions/pi-workflow-ui.md` (visible portion shipped; speculative remainder unscoped), `docs/missions/later/multi-channel-bus.md` (premise actively rejected — wire surface is now outbound-only), `docs/missions/later/personal-replica.md` (speculative ideation, not a plan), `docs/missions/later/ralph-loop-use-cases.md` (brainstorm artifact). Closes [#154](https://github.com/dotdevdotdev/agentwire-dev/issues/154).
 - **Inbound channel surface** — entire Telegram bridge (`agentwire/bridges/telegram.py`), Discord channel (`agentwire/channels/discord.py`), Slack channel (`agentwire/channels/slack.py`), Twilio SMS (`agentwire/channels/sms.py`), Webhook (`agentwire/channels/webhook.py`), outbound-Telegram (`agentwire/channels/telegram.py`), channel scaffolding template (`agentwire/channels/_template.py`), all `agentwire {telegram,discord,slack,sms,webhook}` CLI command groups, the `agentwire reply` command, MCP tools `discord_status` / `slack_status` / `sms_send` / `webhook_send`, the portal sidebar "socials" section, the `ServiceChannel` / `MessageQueueManager` / `QueuedMessage` / `compose_session_config` / `inject_instructions` / session-helper plumbing in `channels/base.py`, the TTS/STT primitives on the channel base class, the auto-injected `discord-dm` / `slack-dm` / `channel-admin` roles, the `OutputConfig.notify` task field plus its `_handle_task_notification` dispatcher (voice / alert / webhook / email / command), the doctor's Telegram bridge check, and the `aiogram` optional dep. Wire surface is now outbound-only: **email** (Resend) and **quo** (OpenPhone SMS) remain. Inbound user input flows through the portal. Security review traced the cut: Telegram bridge was secure (refuse-to-start on empty allowlist + silent default-deny), Discord/Slack fail-open on empty allowlist + unauthenticated channel `@mention`; rather than fix the gaps, the whole inbound surface comes out to keep the wire posture outbound-only.
 - **Anthropic SDK surface** — entire `agentwire/sdk/` package, `agentwire/repl/` (Textual REPL), `agentwire/workflows/runners/{anthropic,human_gate}.py`, `sdk-bypass` / `sdk-prompted` / `sdk-restricted` session types, `sdk-watch` portal window + sidebar section, `--runner` workflow CLI override, and the `claude-agent-sdk` dependency. Anthropic moved Agent SDK + `claude -p` usage out of the monthly subscription tier on 2026-05-13 (now API-billed); the in-tree consumers were too expensive to keep running. The `pi` workflow runner is now the only runner. For Anthropic-quality scheduled work, use a `claude-bypass` tmux session via `.agentwire.yml` + scheduler. Mission [#184](https://github.com/dotdevdotdev/agentwire-dev/issues/184). Findings captured in `~/.agentwire/wiki/` (anthropic-agent-sdk, anthropic-runner-findings, textual-repl-retrospective, workflow-runner-choice, claude-code-print-mode).
 

{agentwire_dev-1.29.1 → agentwire_dev-1.29.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentwire-dev
-Version: 1.29.1
+Version: 1.29.2
 Summary: Multi-session voice web interface for AI coding agents
 Project-URL: Homepage, https://agentwire.dev
 Project-URL: Repository, https://github.com/dotdevdotdev/agentwire-dev
@@ -686,6 +686,7 @@ Classifier: Topic :: Software Development :: User Interfaces
 Requires-Python: >=3.10
 Requires-Dist: aiohttp-jinja2>=1.6
 Requires-Dist: aiohttp>=3.9.0
+Requires-Dist: av>=12.0.0
 Requires-Dist: jinja2>=3.1.0
 Requires-Dist: markdown>=3.5
 Requires-Dist: mcp>=1.2.0

{agentwire_dev-1.29.1 → agentwire_dev-1.29.2}/agentwire/__init__.py

@@ -1,3 +1,3 @@
 """AgentWire - Multi-session voice web interface for AI coding agents."""
 
-__version__ = "1.29.1"
+__version__ = "1.29.2"

{agentwire_dev-1.29.1 → agentwire_dev-1.29.2}/agentwire/config.py

@@ -109,6 +109,10 @@ class STTConfig:
 
     url: str | None = None  # STT server URL (e.g., http://localhost:8100)
     timeout: int = 30
+    # Milliseconds of silence to prepend to the decoded audio before sending
+    # to the STT backend. Default 0 — moonshine doesn't need it. Set to ~300
+    # if your backend (e.g. older faster-whisper builds) clips the first syllable.
+    silence_prepend_ms: int = 0
 
 
 @dataclass
@@ -406,6 +410,7 @@ def _dict_to_config(data: dict) -> Config:
     stt = STTConfig(
         url=stt_data.get("url"),
         timeout=stt_data.get("timeout", 30),
+        silence_prepend_ms=int(stt_data.get("silence_prepend_ms", 0)),
     )
 
     # Agent

{agentwire_dev-1.29.1 → agentwire_dev-1.29.2}/agentwire/server.py

@@ -3283,7 +3283,13 @@ projects:
             return web.json_response({"error": str(e)}, status=500)
 
     async def handle_transcribe(self, request: web.Request) -> web.Response:
-        """Transcribe audio to text."""
+        """Transcribe audio to text.
+
+        Decodes WebM/Opus uploads in-process via PyAV (no ffmpeg subprocess
+        startup) and resamples to 16 kHz mono PCM16 — the canonical input
+        shape for Whisper- and Moonshine-class models. Optionally prepends a
+        configurable amount of silence (``stt.silence_prepend_ms``, default 0).
+        """
         try:
             reader = await request.multipart()
             audio_field = await reader.next()
@@ -3291,46 +3297,88 @@ projects:
             if audio_field is None:
                 return web.json_response({"error": "No audio data"})
 
-            # Read audio data
             audio_data = await audio_field.read()
-
             if not audio_data:
                 return web.json_response({"error": "Empty audio data"})
 
-            # Save webm to temp file
-            with tempfile.NamedTemporaryFile(suffix=".webm", delete=False) as f:
-                f.write(audio_data)
-                webm_path = f.name
+            silence_ms = int(getattr(self.config.stt, "silence_prepend_ms", 0) or 0)
 
-            # Convert webm to wav (16kHz mono for Whisper)
-            wav_path = webm_path.replace(".webm", ".wav")
             try:
-                logger.info("Converting webm to wav: %s -> %s", webm_path, wav_path)
-                proc = await asyncio.create_subprocess_exec(
-                    "ffmpeg", "-i", webm_path,
-                    "-ar", "16000", "-ac", "1", "-y", wav_path,
-                    stdout=asyncio.subprocess.DEVNULL,
-                    stderr=asyncio.subprocess.DEVNULL,
+                wav_data = await asyncio.get_event_loop().run_in_executor(
+                    None,
+                    self._decode_audio_to_wav,
+                    audio_data,
+                    silence_ms,
                 )
-                await proc.wait()
+            except Exception as e:
+                logger.error("Failed to decode audio: %s", e)
+                return web.json_response({"error": "Audio conversion failed"})
 
-                if proc.returncode != 0 or not Path(wav_path).exists():
-                    logger.error("Failed to convert webm to wav (ffmpeg returned %d)", proc.returncode)
-                    return web.json_response({"error": "Audio conversion failed"})
+            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
+                f.write(wav_data)
+                wav_path = f.name
 
-                # Transcribe the wav file
+            try:
                 logger.info("Transcribing %s via %s backend", wav_path, type(self.stt).__name__)
                 text = await self.stt.transcribe(Path(wav_path))
                 logger.info("Transcription result: %s", text)
                 return web.json_response({"text": text})
             finally:
-                Path(webm_path).unlink(missing_ok=True)
                 Path(wav_path).unlink(missing_ok=True)
 
         except Exception as e:
             logger.error(f"Transcription failed: {e}")
             return web.json_response({"error": str(e)})
 
+    @staticmethod
+    def _decode_audio_to_wav(audio_data: bytes, silence_prepend_ms: int = 0) -> bytes:
+        """Decode arbitrary input audio (WebM/Opus, MP3, M4A, …) to 16 kHz mono PCM16 WAV.
+
+        Replaces the previous ``ffmpeg -i in.webm out.wav`` subprocess. Subprocess
+        cold-start was 100–300 ms before any actual decoding; PyAV uses libav
+        bindings in-process so the only cost is the decoding itself.
+        """
+        import io
+        import wave
+
+        import av  # PyAV — declared in pyproject.toml `dependencies`
+
+        target_rate = 16000
+
+        with av.open(io.BytesIO(audio_data), mode="r") as container:
+            if not container.streams.audio:
+                raise RuntimeError("Input contains no audio stream")
+
+            resampler = av.AudioResampler(format="s16", layout="mono", rate=target_rate)
+            pcm_chunks: list[bytes] = []
+
+            def _frame_bytes(f) -> bytes:
+                # AudioFrame.planes[0] may include SIMD alignment padding; slice
+                # to the exact PCM length (samples × channels × bytes_per_sample).
+                bytes_per_sample = f.format.bytes  # 2 for s16
+                channels = len(f.layout.channels)  # 1 for mono
+                size = f.samples * channels * bytes_per_sample
+                return bytes(f.planes[0])[:size]
+
+            for frame in container.decode(audio=0):
+                for resampled in resampler.resample(frame):
+                    pcm_chunks.append(_frame_bytes(resampled))
+            for resampled in resampler.resample(None):
+                pcm_chunks.append(_frame_bytes(resampled))
+
+        if silence_prepend_ms > 0:
+            silence_samples = int(target_rate * silence_prepend_ms / 1000)
+            pcm_chunks.insert(0, b"\x00\x00" * silence_samples)
+
+        pcm_data = b"".join(pcm_chunks)
+        buf = io.BytesIO()
+        with wave.open(buf, "wb") as wav:
+            wav.setnchannels(1)
+            wav.setsampwidth(2)  # 16-bit
+            wav.setframerate(target_rate)
+            wav.writeframes(pcm_data)
+        return buf.getvalue()
+
     async def handle_upload(self, request: web.Request) -> web.Response:
         """Upload an image file for attachment to messages."""
         try:

{agentwire_dev-1.29.1 → agentwire_dev-1.29.2}/agentwire/static/css/desktop.css

@@ -1162,6 +1162,111 @@ body {
     text-align: center;
 }
 
+/* =========================================================================
+   Command palette (Cmd/Ctrl+K)
+   ========================================================================= */
+
+.cmdk-overlay {
+    align-items: flex-start;
+    padding-top: 12vh;
+}
+
+.command-palette {
+    width: 560px;
+    max-width: 92vw;
+    display: flex;
+    flex-direction: column;
+    overflow: hidden;
+}
+
+.cmdk-search {
+    display: flex;
+    align-items: center;
+    gap: 10px;
+    padding: 12px 16px;
+    border-bottom: 1px solid var(--chrome-border);
+}
+
+.cmdk-search-icon {
+    color: var(--text-muted);
+    font-size: 16px;
+}
+
+.cmdk-input {
+    flex: 1;
+    background: transparent;
+    border: none;
+    outline: none;
+    color: var(--text);
+    font-size: 16px;
+    font-family: inherit;
+}
+
+.cmdk-input::placeholder {
+    color: var(--text-muted);
+}
+
+.cmdk-body {
+    padding: 8px;
+    max-height: 50vh;
+    overflow-y: auto;
+}
+
+/* Form drill-ins reuse the quicktask form styling. */
+.cmdk-body .quicktask-form {
+    padding: 8px 8px 4px;
+}
+
+.cmdk-item {
+    display: flex;
+    align-items: center;
+    gap: 12px;
+    padding: 10px 12px;
+    border-radius: 6px;
+    cursor: pointer;
+    font-size: 14px;
+    color: var(--text);
+}
+
+.cmdk-item-selected {
+    background: var(--accent);
+    color: var(--background);
+}
+
+.cmdk-item-icon {
+    width: 18px;
+    text-align: center;
+    opacity: 0.85;
+}
+
+.cmdk-item-label {
+    flex: 1;
+}
+
+.cmdk-item-sub {
+    font-size: 12px;
+    color: var(--text-muted);
+}
+
+.cmdk-item-selected .cmdk-item-sub {
+    color: var(--background);
+    opacity: 0.8;
+}
+
+.cmdk-empty {
+    padding: 20px 12px;
+    text-align: center;
+    color: var(--text-muted);
+    font-size: 14px;
+}
+
+.cmdk-footer {
+    padding: 8px 16px;
+    border-top: 1px solid var(--chrome-border);
+    font-size: 12px;
+    color: var(--text-muted);
+}
+
 .sidebar-quicktask {
     background: transparent;
     border: none;
@@ -4601,3 +4706,171 @@ body.sidebar-pinned .sidebar-tab {
 }
 
 
+/* =========================================================================
+   Pinned Message Panel — floating, draggable, resizable
+   ========================================================================= */
+
+/* Pin affordance on chat messages — visible on hover, top-right corner. */
+.chat-message {
+    position: relative;
+    padding-right: 32px;
+}
+
+.chat-message-pin-btn {
+    position: absolute;
+    top: 4px;
+    right: 4px;
+    width: 22px;
+    height: 22px;
+    padding: 0;
+    border: 1px solid var(--chrome-border);
+    border-radius: 4px;
+    background: var(--chrome);
+    color: var(--text-muted);
+    font-size: 12px;
+    line-height: 1;
+    cursor: pointer;
+    opacity: 0;
+    transition: opacity 120ms ease, color 120ms ease, background 120ms ease;
+}
+
+.chat-message:hover .chat-message-pin-btn,
+.chat-message:focus-within .chat-message-pin-btn,
+.chat-message-pin-btn:focus-visible {
+    opacity: 0.9;
+}
+
+.chat-message-pin-btn:hover {
+    background: var(--hover);
+    color: var(--accent);
+    opacity: 1;
+}
+
+.chat-message-pin-btn[data-pinned="true"] {
+    opacity: 1;
+    color: var(--accent);
+    border-color: var(--accent);
+}
+
+/* The floating panel itself. position: fixed keeps it above all winboxes
+   regardless of scroll, typing, or which session window is active. */
+.pinned-message-panel {
+    position: fixed;
+    display: flex;
+    flex-direction: column;
+    background: var(--chrome);
+    color: var(--text);
+    border: 1px solid var(--chrome-border);
+    border-radius: 8px;
+    box-shadow: 0 10px 30px rgba(0, 0, 0, 0.55), 0 0 0 1px var(--primary-subtle);
+    z-index: 99998;
+    min-width: 200px;
+    min-height: 140px;
+    overflow: hidden;
+    user-select: text;
+}
+
+.pinned-message-panel.dragging,
+.pinned-message-panel.resizing {
+    user-select: none;
+}
+
+.pinned-message-header {
+    display: flex;
+    align-items: center;
+    gap: 8px;
+    padding: 6px 10px;
+    background: var(--background);
+    border-bottom: 1px solid var(--chrome-border);
+    cursor: grab;
+    flex-shrink: 0;
+}
+
+.pinned-message-panel.dragging .pinned-message-header {
+    cursor: grabbing;
+}
+
+.pinned-message-role {
+    font-size: 11px;
+    text-transform: uppercase;
+    letter-spacing: 0.06em;
+    padding: 2px 6px;
+    border-radius: 4px;
+    background: var(--chrome);
+    color: var(--text-muted);
+    border: 1px solid var(--chrome-border);
+    flex-shrink: 0;
+}
+
+.pinned-message-role[data-role="user"] {
+    color: var(--accent);
+    border-color: var(--primary-subtle);
+}
+
+.pinned-message-title {
+    flex: 1;
+    font-size: 12px;
+    color: var(--text-muted);
+    text-transform: uppercase;
+    letter-spacing: 0.08em;
+    overflow: hidden;
+    white-space: nowrap;
+    text-overflow: ellipsis;
+}
+
+.pinned-message-unpin {
+    width: 22px;
+    height: 22px;
+    padding: 0;
+    border: 1px solid var(--chrome-border);
+    border-radius: 4px;
+    background: transparent;
+    color: var(--text-muted);
+    font-size: 14px;
+    line-height: 1;
+    cursor: pointer;
+    transition: color 120ms ease, background 120ms ease, border-color 120ms ease;
+    flex-shrink: 0;
+}
+
+.pinned-message-unpin:hover {
+    color: var(--error);
+    border-color: var(--error);
+    background: var(--hover);
+}
+
+.pinned-message-body {
+    flex: 1;
+    overflow: auto;
+    padding: 10px 12px;
+    font-size: 14px;
+    line-height: 1.45;
+    color: var(--text);
+    white-space: pre-wrap;
+    word-break: break-word;
+}
+
+/* Corner resize handle — visible on hover. */
+.pinned-message-resize-handle {
+    position: absolute;
+    right: 0;
+    bottom: 0;
+    width: 16px;
+    height: 16px;
+    cursor: nwse-resize;
+    opacity: 0;
+    transition: opacity 120ms ease;
+    /* Diagonal grip lines drawn with two thin gradients. */
+    background:
+        linear-gradient(135deg, transparent 0 45%, var(--text-muted) 45% 55%, transparent 55% 100%) no-repeat,
+        linear-gradient(135deg, transparent 0 70%, var(--text-muted) 70% 80%, transparent 80% 100%) no-repeat;
+    background-size: 10px 10px, 6px 6px;
+    background-position: bottom 3px right 3px, bottom 3px right 3px;
+}
+
+.pinned-message-panel:hover .pinned-message-resize-handle,
+.pinned-message-panel.resizing .pinned-message-resize-handle {
+    opacity: 0.85;
+}
+
+