{"entries":[{"lastCommit":{"author":{"email":"127238744+teknium1@users.noreply.github.com","name":"Teknium","when":"2026-04-27T06:22:25.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-04-27T06:22:25.000000-07:00"},"hash":"df3c9593f8822762b5b4bbe30b582b62578d04d1","message":"feat(plugins): google_meet \\u2014 join, transcribe, speak, follow up (#16364)\n\n* feat(plugins): google_meet — bundled plugin for join+transcribe Meet calls\n\nv1 shipping transcribe-only. Spawns headless Chromium via Playwright,\njoins an explicit https://meet.google.com/ URL, enables live captions,\nand scrapes them into a transcript file the agent can read across turns.\nThe agent then has the meeting content in context and can do followup\nwork (send recap, file issues, schedule followups) with its regular tools.\n\nSurface:\n - Tools: meet_join, meet_status, meet_transcript, meet_leave, meet_say\n (meet_say is a v1 stub — returns not-implemented; v2 will wire\n realtime duplex audio via OpenAI Realtime / Gemini Live +\n BlackHole / PulseAudio null-sink.)\n - CLI: hermes meet setup | auth | join | status | transcript | stop\n - Lifecycle: on_session_end auto-leaves any still-running bot.\n\nSafety:\n - URL regex rejects anything that isn't https://meet.google.com/...\n - No calendar scanning, no auto-dial, no auto-consent announcement.\n - Single active meeting per install; a second meet_join leaves the first.\n - Platform-gated to Linux + macOS (Windows audio routing for v2 untested).\n - Opt-in: standalone plugin, user must add 'google_meet' to\n plugins.enabled in config.yaml.\n\nZero core changes. Plugin uses existing register_tool /\nregister_cli_command / register_hook surfaces. 21 new unit tests cover the\nURL safety gate, transcript dedup + status round-trip, process-manager\nrefusals/start/stop paths, tool-handler JSON shape under each branch,\nsession-end cleanup, and platform-gated register().\n\n* feat(plugins/google_meet): v2 realtime audio + v3 remote node host\n\nv2 \\u2014 agent speaks in-meeting\n audio_bridge.py: PulseAudio null-sink (Linux) + BlackHole probe (macOS).\n On Linux we load pactl module-null-sink + module-virtual-source, track\n module ids for teardown; Chrome gets PULSE_SOURCE= env so its\n fake mic reads what we write to the sink. macOS just probes BlackHole\n 2ch and returns its device name \\u2014 the plugin refuses to switch the\n user's default audio input (that would surprise them).\n realtime/openai_client.py: sync WebSocket client for the OpenAI Realtime\n API. RealtimeSession.speak(text) sends conversation.item.create +\n response.create, accumulates response.audio.delta PCM bytes, appends\n them to a file. RealtimeSpeaker runs a JSONL-queue loop consuming\n meet_say calls. 'websockets' is an optional dep imported lazily.\n meet_bot.py: when HERMES_MEET_MODE=realtime, provisions AudioBridge,\n starts RealtimeSession + speaker thread, spawns paplay to pump PCM\n into the null-sink, then cleans everything up on SIGTERM. If any\n realtime setup step fails, falls back cleanly to transcribe mode\n with an error flagged in status.json.\n process_manager.enqueue_say(): writes a JSONL line to say_queue.jsonl;\n refuses when no active meeting or active meeting is transcribe-only.\n tools.meet_say: real implementation; requires active mode='realtime'.\n meet_join: adds mode='transcribe'|'realtime' param.\n\nv3 \\u2014 remote node host\n node/protocol.py: JSON envelope (type, id, token, payload) + validate.\n node/registry.py: $HERMES_HOME/workspace/meetings/nodes.json, with\n resolve() auto-selecting the sole registered node when name is None.\n node/server.py: NodeServer \\u2014 websockets.serve, bearer-token auth,\n dispatches start_bot/stop/status/transcript/say/ping onto the local\n process_manager. Token auto-generated + persisted on first run.\n node/client.py: NodeClient \\u2014 short-lived sync WS per RPC, raises\n RuntimeError on error envelopes, clean API matching the server.\n node/cli.py: 'hermes meet node {run,list,approve,remove,status,ping}'\n subtree; wired into the main meet CLI by cli.py so 'hermes meet node'\n Just Works.\n tools.py: every meet_* tool accepts node=''|'auto'; when set,\n routes through NodeClient to the remote bot instead of running\n locally. Unknown node \\u2192 clear 'no registered meet node matches ...'\n error.\n cli.py: 'hermes meet join --node my-mac --mode realtime' and\n 'hermes meet say \"...\" --node my-mac' route to the node; 'hermes\n meet node approve ' registers one.\n\nTests\n 21 v1 tests updated (meet_say is no longer a stub; active-record now\n carries mode).\n 20 new audio_bridge + realtime tests.\n 42 new node tests (protocol/registry/server/client/cli).\n 17 new v1/v2/v3 integration tests at the plugin level covering\n enqueue_say edge cases, env var passthrough, mode validation, node\n routing (known/unknown/auto/ambiguous), and argparse wiring for\n `hermes meet say` + `hermes meet node` + --mode/--node flags.\n Total: 100 plugin tests + 58 plugin-system tests = 158 passing.\n\nE2E verified on Linux with fresh HERMES_HOME: plugin loads, 5 tools\nregister, on_session_end hook wires, 'hermes meet' CLI tree wires\nincluding the node subtree, NodeRegistry round-trips, meet_join routes\ncorrectly to NodeClient under node='my-mac' with mode='realtime',\nenqueue_say accepts realtime/rejects transcribe, argparse parses every\nnew flag cleanly.\n\nZero changes to core. All new code lives under plugins/google_meet/.\n\n* feat(plugins/google_meet): auto-install, admission detect, mac PCM pump, barge-in, richer status\n\nReady-for-live-test follow-up on PR #16364. Five additions that matter for\nthe first live run on a real Meet, in priority order:\n\n1. hermes meet install [--realtime] [--yes]\n pip install playwright websockets + python -m playwright install chromium\n --realtime: installs platform audio deps (pulseaudio-utils on Linux via\n sudo apt, blackhole-2ch + ffmpeg on macOS via brew). Prompts before\n sudo/brew unless --yes. Refuses on Windows. Refuses to auto-flip the\n macOS default input — user still selects BlackHole in System Settings\n (deliberate; surprise audio rerouting is worse than a manual step).\n\n2. Admission detection\n _detect_admission(page): Leave-button visible OR caption region\n attached OR participants list present → we're in-call.\n _detect_denied(page): 'You can\\'t join this video call' / 'You were\n removed' / 'No one responded to your request' → bail out.\n HERMES_MEET_LOBBY_TIMEOUT (default 300s) caps how long we sit in\n the lobby before giving up. in_call stays False until admitted.\n Status surfaces leaveReason: duration_expired | lobby_timeout |\n denied | page_closed.\n\n3. macOS PCM pump\n ffmpeg reads speaker.pcm (24kHz s16le mono) and writes to the\n BlackHole AVFoundation output via -f audiotoolbox\n -audio_device_index . _mac_audio_device_index() probes\n ffmpeg -f avfoundation -list_devices true to resolve 'BlackHole 2ch'\n → numeric index. Falls back to index 0 on probe failure. Linux\n paplay pump unchanged.\n\n4. Richer status dict\n _BotState now tracks realtime, realtimeReady, realtimeDevice,\n audioBytesOut, lastAudioOutAt, lastBargeInAt, joinAttemptedAt,\n leaveReason. RealtimeSession.audio_bytes_out / last_audio_out_at\n counters fold into the status file once a second so meet_status()\n can show the agent's voice activity in near-real-time.\n\n5. Barge-in\n RealtimeSession.cancel_response() sends type='response.cancel' over\n the same WS (lock-guarded so it's safe to call from the caption\n thread while speak() is reading frames). Handles response.cancelled\n as a terminal frame type. _looks_like_human_speaker() gates triggers\n so the bot's own name, 'You', 'Unknown', and blanks don't self-cancel.\n Called from the caption drain loop: when a new caption arrives\n attributed to a real participant while rt.session exists, we fire\n cancel_response() and stamp lastBargeInAt.\n\nTests: 20 new unit tests across _BotState telemetry, barge-in gating,\nadmission/denied probe error handling, cancel_response with and without\na connected WS, and `hermes meet install` CLI wiring (flag parsing +\nend-to-end subprocess.run verification + Linux-already-installed fast\npath). Total 171 passing across all google_meet test files + the\nplugin-system regression suite.\n\nE2E verified on Linux: plugin loads, all 5 tools register,\n`hermes meet install --realtime --yes` parses, fresh-bot status.json\nhas every new telemetry key, cancel_response on a disconnected session\nreturns False without raising, barge-in helper gates the bot's own\nname correctly.\n\nStill out of scope (for a future PR, not blocking live test):\nmic → Realtime duplex (the agent listening to meeting audio via\nWebRTC), node-host TLS/pairing UX, Windows audio, Meet create+Twilio.\n\nDocs updated: SKILL.md now lists the installer subcommand, lobby\ntimeout, barge-in caveat, and the full status-dict reference table.\nREADME.md quick-start uses hermes meet install.","tree":"5ff3d2af72d075acc723c1642d1c1f101d0ac301"},"mode":"0100644","name":"__init__.py","size":1707},{"lastCommit":{"author":{"email":"127238744+teknium1@users.noreply.github.com","name":"Teknium","when":"2026-04-27T06:22:25.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-04-27T06:22:25.000000-07:00"},"hash":"df3c9593f8822762b5b4bbe30b582b62578d04d1","message":"feat(plugins): google_meet \\u2014 join, transcribe, speak, follow up (#16364)\n\n* feat(plugins): google_meet — bundled plugin for join+transcribe Meet calls\n\nv1 shipping transcribe-only. Spawns headless Chromium via Playwright,\njoins an explicit https://meet.google.com/ URL, enables live captions,\nand scrapes them into a transcript file the agent can read across turns.\nThe agent then has the meeting content in context and can do followup\nwork (send recap, file issues, schedule followups) with its regular tools.\n\nSurface:\n - Tools: meet_join, meet_status, meet_transcript, meet_leave, meet_say\n (meet_say is a v1 stub — returns not-implemented; v2 will wire\n realtime duplex audio via OpenAI Realtime / Gemini Live +\n BlackHole / PulseAudio null-sink.)\n - CLI: hermes meet setup | auth | join | status | transcript | stop\n - Lifecycle: on_session_end auto-leaves any still-running bot.\n\nSafety:\n - URL regex rejects anything that isn't https://meet.google.com/...\n - No calendar scanning, no auto-dial, no auto-consent announcement.\n - Single active meeting per install; a second meet_join leaves the first.\n - Platform-gated to Linux + macOS (Windows audio routing for v2 untested).\n - Opt-in: standalone plugin, user must add 'google_meet' to\n plugins.enabled in config.yaml.\n\nZero core changes. Plugin uses existing register_tool /\nregister_cli_command / register_hook surfaces. 21 new unit tests cover the\nURL safety gate, transcript dedup + status round-trip, process-manager\nrefusals/start/stop paths, tool-handler JSON shape under each branch,\nsession-end cleanup, and platform-gated register().\n\n* feat(plugins/google_meet): v2 realtime audio + v3 remote node host\n\nv2 \\u2014 agent speaks in-meeting\n audio_bridge.py: PulseAudio null-sink (Linux) + BlackHole probe (macOS).\n On Linux we load pactl module-null-sink + module-virtual-source, track\n module ids for teardown; Chrome gets PULSE_SOURCE= env so its\n fake mic reads what we write to the sink. macOS just probes BlackHole\n 2ch and returns its device name \\u2014 the plugin refuses to switch the\n user's default audio input (that would surprise them).\n realtime/openai_client.py: sync WebSocket client for the OpenAI Realtime\n API. RealtimeSession.speak(text) sends conversation.item.create +\n response.create, accumulates response.audio.delta PCM bytes, appends\n them to a file. RealtimeSpeaker runs a JSONL-queue loop consuming\n meet_say calls. 'websockets' is an optional dep imported lazily.\n meet_bot.py: when HERMES_MEET_MODE=realtime, provisions AudioBridge,\n starts RealtimeSession + speaker thread, spawns paplay to pump PCM\n into the null-sink, then cleans everything up on SIGTERM. If any\n realtime setup step fails, falls back cleanly to transcribe mode\n with an error flagged in status.json.\n process_manager.enqueue_say(): writes a JSONL line to say_queue.jsonl;\n refuses when no active meeting or active meeting is transcribe-only.\n tools.meet_say: real implementation; requires active mode='realtime'.\n meet_join: adds mode='transcribe'|'realtime' param.\n\nv3 \\u2014 remote node host\n node/protocol.py: JSON envelope (type, id, token, payload) + validate.\n node/registry.py: $HERMES_HOME/workspace/meetings/nodes.json, with\n resolve() auto-selecting the sole registered node when name is None.\n node/server.py: NodeServer \\u2014 websockets.serve, bearer-token auth,\n dispatches start_bot/stop/status/transcript/say/ping onto the local\n process_manager. Token auto-generated + persisted on first run.\n node/client.py: NodeClient \\u2014 short-lived sync WS per RPC, raises\n RuntimeError on error envelopes, clean API matching the server.\n node/cli.py: 'hermes meet node {run,list,approve,remove,status,ping}'\n subtree; wired into the main meet CLI by cli.py so 'hermes meet node'\n Just Works.\n tools.py: every meet_* tool accepts node=''|'auto'; when set,\n routes through NodeClient to the remote bot instead of running\n locally. Unknown node \\u2192 clear 'no registered meet node matches ...'\n error.\n cli.py: 'hermes meet join --node my-mac --mode realtime' and\n 'hermes meet say \"...\" --node my-mac' route to the node; 'hermes\n meet node approve ' registers one.\n\nTests\n 21 v1 tests updated (meet_say is no longer a stub; active-record now\n carries mode).\n 20 new audio_bridge + realtime tests.\n 42 new node tests (protocol/registry/server/client/cli).\n 17 new v1/v2/v3 integration tests at the plugin level covering\n enqueue_say edge cases, env var passthrough, mode validation, node\n routing (known/unknown/auto/ambiguous), and argparse wiring for\n `hermes meet say` + `hermes meet node` + --mode/--node flags.\n Total: 100 plugin tests + 58 plugin-system tests = 158 passing.\n\nE2E verified on Linux with fresh HERMES_HOME: plugin loads, 5 tools\nregister, on_session_end hook wires, 'hermes meet' CLI tree wires\nincluding the node subtree, NodeRegistry round-trips, meet_join routes\ncorrectly to NodeClient under node='my-mac' with mode='realtime',\nenqueue_say accepts realtime/rejects transcribe, argparse parses every\nnew flag cleanly.\n\nZero changes to core. All new code lives under plugins/google_meet/.\n\n* feat(plugins/google_meet): auto-install, admission detect, mac PCM pump, barge-in, richer status\n\nReady-for-live-test follow-up on PR #16364. Five additions that matter for\nthe first live run on a real Meet, in priority order:\n\n1. hermes meet install [--realtime] [--yes]\n pip install playwright websockets + python -m playwright install chromium\n --realtime: installs platform audio deps (pulseaudio-utils on Linux via\n sudo apt, blackhole-2ch + ffmpeg on macOS via brew). Prompts before\n sudo/brew unless --yes. Refuses on Windows. Refuses to auto-flip the\n macOS default input — user still selects BlackHole in System Settings\n (deliberate; surprise audio rerouting is worse than a manual step).\n\n2. Admission detection\n _detect_admission(page): Leave-button visible OR caption region\n attached OR participants list present → we're in-call.\n _detect_denied(page): 'You can\\'t join this video call' / 'You were\n removed' / 'No one responded to your request' → bail out.\n HERMES_MEET_LOBBY_TIMEOUT (default 300s) caps how long we sit in\n the lobby before giving up. in_call stays False until admitted.\n Status surfaces leaveReason: duration_expired | lobby_timeout |\n denied | page_closed.\n\n3. macOS PCM pump\n ffmpeg reads speaker.pcm (24kHz s16le mono) and writes to the\n BlackHole AVFoundation output via -f audiotoolbox\n -audio_device_index . _mac_audio_device_index() probes\n ffmpeg -f avfoundation -list_devices true to resolve 'BlackHole 2ch'\n → numeric index. Falls back to index 0 on probe failure. Linux\n paplay pump unchanged.\n\n4. Richer status dict\n _BotState now tracks realtime, realtimeReady, realtimeDevice,\n audioBytesOut, lastAudioOutAt, lastBargeInAt, joinAttemptedAt,\n leaveReason. RealtimeSession.audio_bytes_out / last_audio_out_at\n counters fold into the status file once a second so meet_status()\n can show the agent's voice activity in near-real-time.\n\n5. Barge-in\n RealtimeSession.cancel_response() sends type='response.cancel' over\n the same WS (lock-guarded so it's safe to call from the caption\n thread while speak() is reading frames). Handles response.cancelled\n as a terminal frame type. _looks_like_human_speaker() gates triggers\n so the bot's own name, 'You', 'Unknown', and blanks don't self-cancel.\n Called from the caption drain loop: when a new caption arrives\n attributed to a real participant while rt.session exists, we fire\n cancel_response() and stamp lastBargeInAt.\n\nTests: 20 new unit tests across _BotState telemetry, barge-in gating,\nadmission/denied probe error handling, cancel_response with and without\na connected WS, and `hermes meet install` CLI wiring (flag parsing +\nend-to-end subprocess.run verification + Linux-already-installed fast\npath). Total 171 passing across all google_meet test files + the\nplugin-system regression suite.\n\nE2E verified on Linux: plugin loads, all 5 tools register,\n`hermes meet install --realtime --yes` parses, fresh-bot status.json\nhas every new telemetry key, cancel_response on a disconnected session\nreturns False without raising, barge-in helper gates the bot's own\nname correctly.\n\nStill out of scope (for a future PR, not blocking live test):\nmic → Realtime duplex (the agent listening to meeting audio via\nWebRTC), node-host TLS/pairing UX, Windows audio, Meet create+Twilio.\n\nDocs updated: SKILL.md now lists the installer subcommand, lobby\ntimeout, barge-in caveat, and the full status-dict reference table.\nREADME.md quick-start uses hermes meet install.","tree":"5ff3d2af72d075acc723c1642d1c1f101d0ac301"},"mode":"0100644","name":"cli.py","size":4205},{"lastCommit":{"author":{"email":"127238744+teknium1@users.noreply.github.com","name":"Teknium","when":"2026-04-27T06:22:25.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-04-27T06:22:25.000000-07:00"},"hash":"df3c9593f8822762b5b4bbe30b582b62578d04d1","message":"feat(plugins): google_meet \\u2014 join, transcribe, speak, follow up (#16364)\n\n* feat(plugins): google_meet — bundled plugin for join+transcribe Meet calls\n\nv1 shipping transcribe-only. Spawns headless Chromium via Playwright,\njoins an explicit https://meet.google.com/ URL, enables live captions,\nand scrapes them into a transcript file the agent can read across turns.\nThe agent then has the meeting content in context and can do followup\nwork (send recap, file issues, schedule followups) with its regular tools.\n\nSurface:\n - Tools: meet_join, meet_status, meet_transcript, meet_leave, meet_say\n (meet_say is a v1 stub — returns not-implemented; v2 will wire\n realtime duplex audio via OpenAI Realtime / Gemini Live +\n BlackHole / PulseAudio null-sink.)\n - CLI: hermes meet setup | auth | join | status | transcript | stop\n - Lifecycle: on_session_end auto-leaves any still-running bot.\n\nSafety:\n - URL regex rejects anything that isn't https://meet.google.com/...\n - No calendar scanning, no auto-dial, no auto-consent announcement.\n - Single active meeting per install; a second meet_join leaves the first.\n - Platform-gated to Linux + macOS (Windows audio routing for v2 untested).\n - Opt-in: standalone plugin, user must add 'google_meet' to\n plugins.enabled in config.yaml.\n\nZero core changes. Plugin uses existing register_tool /\nregister_cli_command / register_hook surfaces. 21 new unit tests cover the\nURL safety gate, transcript dedup + status round-trip, process-manager\nrefusals/start/stop paths, tool-handler JSON shape under each branch,\nsession-end cleanup, and platform-gated register().\n\n* feat(plugins/google_meet): v2 realtime audio + v3 remote node host\n\nv2 \\u2014 agent speaks in-meeting\n audio_bridge.py: PulseAudio null-sink (Linux) + BlackHole probe (macOS).\n On Linux we load pactl module-null-sink + module-virtual-source, track\n module ids for teardown; Chrome gets PULSE_SOURCE= env so its\n fake mic reads what we write to the sink. macOS just probes BlackHole\n 2ch and returns its device name \\u2014 the plugin refuses to switch the\n user's default audio input (that would surprise them).\n realtime/openai_client.py: sync WebSocket client for the OpenAI Realtime\n API. RealtimeSession.speak(text) sends conversation.item.create +\n response.create, accumulates response.audio.delta PCM bytes, appends\n them to a file. RealtimeSpeaker runs a JSONL-queue loop consuming\n meet_say calls. 'websockets' is an optional dep imported lazily.\n meet_bot.py: when HERMES_MEET_MODE=realtime, provisions AudioBridge,\n starts RealtimeSession + speaker thread, spawns paplay to pump PCM\n into the null-sink, then cleans everything up on SIGTERM. If any\n realtime setup step fails, falls back cleanly to transcribe mode\n with an error flagged in status.json.\n process_manager.enqueue_say(): writes a JSONL line to say_queue.jsonl;\n refuses when no active meeting or active meeting is transcribe-only.\n tools.meet_say: real implementation; requires active mode='realtime'.\n meet_join: adds mode='transcribe'|'realtime' param.\n\nv3 \\u2014 remote node host\n node/protocol.py: JSON envelope (type, id, token, payload) + validate.\n node/registry.py: $HERMES_HOME/workspace/meetings/nodes.json, with\n resolve() auto-selecting the sole registered node when name is None.\n node/server.py: NodeServer \\u2014 websockets.serve, bearer-token auth,\n dispatches start_bot/stop/status/transcript/say/ping onto the local\n process_manager. Token auto-generated + persisted on first run.\n node/client.py: NodeClient \\u2014 short-lived sync WS per RPC, raises\n RuntimeError on error envelopes, clean API matching the server.\n node/cli.py: 'hermes meet node {run,list,approve,remove,status,ping}'\n subtree; wired into the main meet CLI by cli.py so 'hermes meet node'\n Just Works.\n tools.py: every meet_* tool accepts node=''|'auto'; when set,\n routes through NodeClient to the remote bot instead of running\n locally. Unknown node \\u2192 clear 'no registered meet node matches ...'\n error.\n cli.py: 'hermes meet join --node my-mac --mode realtime' and\n 'hermes meet say \"...\" --node my-mac' route to the node; 'hermes\n meet node approve ' registers one.\n\nTests\n 21 v1 tests updated (meet_say is no longer a stub; active-record now\n carries mode).\n 20 new audio_bridge + realtime tests.\n 42 new node tests (protocol/registry/server/client/cli).\n 17 new v1/v2/v3 integration tests at the plugin level covering\n enqueue_say edge cases, env var passthrough, mode validation, node\n routing (known/unknown/auto/ambiguous), and argparse wiring for\n `hermes meet say` + `hermes meet node` + --mode/--node flags.\n Total: 100 plugin tests + 58 plugin-system tests = 158 passing.\n\nE2E verified on Linux with fresh HERMES_HOME: plugin loads, 5 tools\nregister, on_session_end hook wires, 'hermes meet' CLI tree wires\nincluding the node subtree, NodeRegistry round-trips, meet_join routes\ncorrectly to NodeClient under node='my-mac' with mode='realtime',\nenqueue_say accepts realtime/rejects transcribe, argparse parses every\nnew flag cleanly.\n\nZero changes to core. All new code lives under plugins/google_meet/.\n\n* feat(plugins/google_meet): auto-install, admission detect, mac PCM pump, barge-in, richer status\n\nReady-for-live-test follow-up on PR #16364. Five additions that matter for\nthe first live run on a real Meet, in priority order:\n\n1. hermes meet install [--realtime] [--yes]\n pip install playwright websockets + python -m playwright install chromium\n --realtime: installs platform audio deps (pulseaudio-utils on Linux via\n sudo apt, blackhole-2ch + ffmpeg on macOS via brew). Prompts before\n sudo/brew unless --yes. Refuses on Windows. Refuses to auto-flip the\n macOS default input — user still selects BlackHole in System Settings\n (deliberate; surprise audio rerouting is worse than a manual step).\n\n2. Admission detection\n _detect_admission(page): Leave-button visible OR caption region\n attached OR participants list present → we're in-call.\n _detect_denied(page): 'You can\\'t join this video call' / 'You were\n removed' / 'No one responded to your request' → bail out.\n HERMES_MEET_LOBBY_TIMEOUT (default 300s) caps how long we sit in\n the lobby before giving up. in_call stays False until admitted.\n Status surfaces leaveReason: duration_expired | lobby_timeout |\n denied | page_closed.\n\n3. macOS PCM pump\n ffmpeg reads speaker.pcm (24kHz s16le mono) and writes to the\n BlackHole AVFoundation output via -f audiotoolbox\n -audio_device_index . _mac_audio_device_index() probes\n ffmpeg -f avfoundation -list_devices true to resolve 'BlackHole 2ch'\n → numeric index. Falls back to index 0 on probe failure. Linux\n paplay pump unchanged.\n\n4. Richer status dict\n _BotState now tracks realtime, realtimeReady, realtimeDevice,\n audioBytesOut, lastAudioOutAt, lastBargeInAt, joinAttemptedAt,\n leaveReason. RealtimeSession.audio_bytes_out / last_audio_out_at\n counters fold into the status file once a second so meet_status()\n can show the agent's voice activity in near-real-time.\n\n5. Barge-in\n RealtimeSession.cancel_response() sends type='response.cancel' over\n the same WS (lock-guarded so it's safe to call from the caption\n thread while speak() is reading frames). Handles response.cancelled\n as a terminal frame type. _looks_like_human_speaker() gates triggers\n so the bot's own name, 'You', 'Unknown', and blanks don't self-cancel.\n Called from the caption drain loop: when a new caption arrives\n attributed to a real participant while rt.session exists, we fire\n cancel_response() and stamp lastBargeInAt.\n\nTests: 20 new unit tests across _BotState telemetry, barge-in gating,\nadmission/denied probe error handling, cancel_response with and without\na connected WS, and `hermes meet install` CLI wiring (flag parsing +\nend-to-end subprocess.run verification + Linux-already-installed fast\npath). Total 171 passing across all google_meet test files + the\nplugin-system regression suite.\n\nE2E verified on Linux: plugin loads, all 5 tools register,\n`hermes meet install --realtime --yes` parses, fresh-bot status.json\nhas every new telemetry key, cancel_response on a disconnected session\nreturns False without raising, barge-in helper gates the bot's own\nname correctly.\n\nStill out of scope (for a future PR, not blocking live test):\nmic → Realtime duplex (the agent listening to meeting audio via\nWebRTC), node-host TLS/pairing UX, Windows audio, Meet create+Twilio.\n\nDocs updated: SKILL.md now lists the installer subcommand, lobby\ntimeout, barge-in caveat, and the full status-dict reference table.\nREADME.md quick-start uses hermes meet install.","tree":"5ff3d2af72d075acc723c1642d1c1f101d0ac301"},"mode":"0100644","name":"client.py","size":3819},{"lastCommit":{"author":{"email":"127238744+teknium1@users.noreply.github.com","name":"Teknium","when":"2026-04-27T06:22:25.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-04-27T06:22:25.000000-07:00"},"hash":"df3c9593f8822762b5b4bbe30b582b62578d04d1","message":"feat(plugins): google_meet \\u2014 join, transcribe, speak, follow up (#16364)\n\n* feat(plugins): google_meet — bundled plugin for join+transcribe Meet calls\n\nv1 shipping transcribe-only. Spawns headless Chromium via Playwright,\njoins an explicit https://meet.google.com/ URL, enables live captions,\nand scrapes them into a transcript file the agent can read across turns.\nThe agent then has the meeting content in context and can do followup\nwork (send recap, file issues, schedule followups) with its regular tools.\n\nSurface:\n - Tools: meet_join, meet_status, meet_transcript, meet_leave, meet_say\n (meet_say is a v1 stub — returns not-implemented; v2 will wire\n realtime duplex audio via OpenAI Realtime / Gemini Live +\n BlackHole / PulseAudio null-sink.)\n - CLI: hermes meet setup | auth | join | status | transcript | stop\n - Lifecycle: on_session_end auto-leaves any still-running bot.\n\nSafety:\n - URL regex rejects anything that isn't https://meet.google.com/...\n - No calendar scanning, no auto-dial, no auto-consent announcement.\n - Single active meeting per install; a second meet_join leaves the first.\n - Platform-gated to Linux + macOS (Windows audio routing for v2 untested).\n - Opt-in: standalone plugin, user must add 'google_meet' to\n plugins.enabled in config.yaml.\n\nZero core changes. Plugin uses existing register_tool /\nregister_cli_command / register_hook surfaces. 21 new unit tests cover the\nURL safety gate, transcript dedup + status round-trip, process-manager\nrefusals/start/stop paths, tool-handler JSON shape under each branch,\nsession-end cleanup, and platform-gated register().\n\n* feat(plugins/google_meet): v2 realtime audio + v3 remote node host\n\nv2 \\u2014 agent speaks in-meeting\n audio_bridge.py: PulseAudio null-sink (Linux) + BlackHole probe (macOS).\n On Linux we load pactl module-null-sink + module-virtual-source, track\n module ids for teardown; Chrome gets PULSE_SOURCE= env so its\n fake mic reads what we write to the sink. macOS just probes BlackHole\n 2ch and returns its device name \\u2014 the plugin refuses to switch the\n user's default audio input (that would surprise them).\n realtime/openai_client.py: sync WebSocket client for the OpenAI Realtime\n API. RealtimeSession.speak(text) sends conversation.item.create +\n response.create, accumulates response.audio.delta PCM bytes, appends\n them to a file. RealtimeSpeaker runs a JSONL-queue loop consuming\n meet_say calls. 'websockets' is an optional dep imported lazily.\n meet_bot.py: when HERMES_MEET_MODE=realtime, provisions AudioBridge,\n starts RealtimeSession + speaker thread, spawns paplay to pump PCM\n into the null-sink, then cleans everything up on SIGTERM. If any\n realtime setup step fails, falls back cleanly to transcribe mode\n with an error flagged in status.json.\n process_manager.enqueue_say(): writes a JSONL line to say_queue.jsonl;\n refuses when no active meeting or active meeting is transcribe-only.\n tools.meet_say: real implementation; requires active mode='realtime'.\n meet_join: adds mode='transcribe'|'realtime' param.\n\nv3 \\u2014 remote node host\n node/protocol.py: JSON envelope (type, id, token, payload) + validate.\n node/registry.py: $HERMES_HOME/workspace/meetings/nodes.json, with\n resolve() auto-selecting the sole registered node when name is None.\n node/server.py: NodeServer \\u2014 websockets.serve, bearer-token auth,\n dispatches start_bot/stop/status/transcript/say/ping onto the local\n process_manager. Token auto-generated + persisted on first run.\n node/client.py: NodeClient \\u2014 short-lived sync WS per RPC, raises\n RuntimeError on error envelopes, clean API matching the server.\n node/cli.py: 'hermes meet node {run,list,approve,remove,status,ping}'\n subtree; wired into the main meet CLI by cli.py so 'hermes meet node'\n Just Works.\n tools.py: every meet_* tool accepts node=''|'auto'; when set,\n routes through NodeClient to the remote bot instead of running\n locally. Unknown node \\u2192 clear 'no registered meet node matches ...'\n error.\n cli.py: 'hermes meet join --node my-mac --mode realtime' and\n 'hermes meet say \"...\" --node my-mac' route to the node; 'hermes\n meet node approve ' registers one.\n\nTests\n 21 v1 tests updated (meet_say is no longer a stub; active-record now\n carries mode).\n 20 new audio_bridge + realtime tests.\n 42 new node tests (protocol/registry/server/client/cli).\n 17 new v1/v2/v3 integration tests at the plugin level covering\n enqueue_say edge cases, env var passthrough, mode validation, node\n routing (known/unknown/auto/ambiguous), and argparse wiring for\n `hermes meet say` + `hermes meet node` + --mode/--node flags.\n Total: 100 plugin tests + 58 plugin-system tests = 158 passing.\n\nE2E verified on Linux with fresh HERMES_HOME: plugin loads, 5 tools\nregister, on_session_end hook wires, 'hermes meet' CLI tree wires\nincluding the node subtree, NodeRegistry round-trips, meet_join routes\ncorrectly to NodeClient under node='my-mac' with mode='realtime',\nenqueue_say accepts realtime/rejects transcribe, argparse parses every\nnew flag cleanly.\n\nZero changes to core. All new code lives under plugins/google_meet/.\n\n* feat(plugins/google_meet): auto-install, admission detect, mac PCM pump, barge-in, richer status\n\nReady-for-live-test follow-up on PR #16364. Five additions that matter for\nthe first live run on a real Meet, in priority order:\n\n1. hermes meet install [--realtime] [--yes]\n pip install playwright websockets + python -m playwright install chromium\n --realtime: installs platform audio deps (pulseaudio-utils on Linux via\n sudo apt, blackhole-2ch + ffmpeg on macOS via brew). Prompts before\n sudo/brew unless --yes. Refuses on Windows. Refuses to auto-flip the\n macOS default input — user still selects BlackHole in System Settings\n (deliberate; surprise audio rerouting is worse than a manual step).\n\n2. Admission detection\n _detect_admission(page): Leave-button visible OR caption region\n attached OR participants list present → we're in-call.\n _detect_denied(page): 'You can\\'t join this video call' / 'You were\n removed' / 'No one responded to your request' → bail out.\n HERMES_MEET_LOBBY_TIMEOUT (default 300s) caps how long we sit in\n the lobby before giving up. in_call stays False until admitted.\n Status surfaces leaveReason: duration_expired | lobby_timeout |\n denied | page_closed.\n\n3. macOS PCM pump\n ffmpeg reads speaker.pcm (24kHz s16le mono) and writes to the\n BlackHole AVFoundation output via -f audiotoolbox\n -audio_device_index . _mac_audio_device_index() probes\n ffmpeg -f avfoundation -list_devices true to resolve 'BlackHole 2ch'\n → numeric index. Falls back to index 0 on probe failure. Linux\n paplay pump unchanged.\n\n4. Richer status dict\n _BotState now tracks realtime, realtimeReady, realtimeDevice,\n audioBytesOut, lastAudioOutAt, lastBargeInAt, joinAttemptedAt,\n leaveReason. RealtimeSession.audio_bytes_out / last_audio_out_at\n counters fold into the status file once a second so meet_status()\n can show the agent's voice activity in near-real-time.\n\n5. Barge-in\n RealtimeSession.cancel_response() sends type='response.cancel' over\n the same WS (lock-guarded so it's safe to call from the caption\n thread while speak() is reading frames). Handles response.cancelled\n as a terminal frame type. _looks_like_human_speaker() gates triggers\n so the bot's own name, 'You', 'Unknown', and blanks don't self-cancel.\n Called from the caption drain loop: when a new caption arrives\n attributed to a real participant while rt.session exists, we fire\n cancel_response() and stamp lastBargeInAt.\n\nTests: 20 new unit tests across _BotState telemetry, barge-in gating,\nadmission/denied probe error handling, cancel_response with and without\na connected WS, and `hermes meet install` CLI wiring (flag parsing +\nend-to-end subprocess.run verification + Linux-already-installed fast\npath). Total 171 passing across all google_meet test files + the\nplugin-system regression suite.\n\nE2E verified on Linux: plugin loads, all 5 tools register,\n`hermes meet install --realtime --yes` parses, fresh-bot status.json\nhas every new telemetry key, cancel_response on a disconnected session\nreturns False without raising, barge-in helper gates the bot's own\nname correctly.\n\nStill out of scope (for a future PR, not blocking live test):\nmic → Realtime duplex (the agent listening to meeting audio via\nWebRTC), node-host TLS/pairing UX, Windows audio, Meet create+Twilio.\n\nDocs updated: SKILL.md now lists the installer subcommand, lobby\ntimeout, barge-in caveat, and the full status-dict reference table.\nREADME.md quick-start uses hermes meet install.","tree":"5ff3d2af72d075acc723c1642d1c1f101d0ac301"},"mode":"0100644","name":"protocol.py","size":4249},{"lastCommit":{"author":{"email":"127238744+teknium1@users.noreply.github.com","name":"Teknium","when":"2026-04-27T06:22:25.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-04-27T06:22:25.000000-07:00"},"hash":"df3c9593f8822762b5b4bbe30b582b62578d04d1","message":"feat(plugins): google_meet \\u2014 join, transcribe, speak, follow up (#16364)\n\n* feat(plugins): google_meet — bundled plugin for join+transcribe Meet calls\n\nv1 shipping transcribe-only. Spawns headless Chromium via Playwright,\njoins an explicit https://meet.google.com/ URL, enables live captions,\nand scrapes them into a transcript file the agent can read across turns.\nThe agent then has the meeting content in context and can do followup\nwork (send recap, file issues, schedule followups) with its regular tools.\n\nSurface:\n - Tools: meet_join, meet_status, meet_transcript, meet_leave, meet_say\n (meet_say is a v1 stub — returns not-implemented; v2 will wire\n realtime duplex audio via OpenAI Realtime / Gemini Live +\n BlackHole / PulseAudio null-sink.)\n - CLI: hermes meet setup | auth | join | status | transcript | stop\n - Lifecycle: on_session_end auto-leaves any still-running bot.\n\nSafety:\n - URL regex rejects anything that isn't https://meet.google.com/...\n - No calendar scanning, no auto-dial, no auto-consent announcement.\n - Single active meeting per install; a second meet_join leaves the first.\n - Platform-gated to Linux + macOS (Windows audio routing for v2 untested).\n - Opt-in: standalone plugin, user must add 'google_meet' to\n plugins.enabled in config.yaml.\n\nZero core changes. Plugin uses existing register_tool /\nregister_cli_command / register_hook surfaces. 21 new unit tests cover the\nURL safety gate, transcript dedup + status round-trip, process-manager\nrefusals/start/stop paths, tool-handler JSON shape under each branch,\nsession-end cleanup, and platform-gated register().\n\n* feat(plugins/google_meet): v2 realtime audio + v3 remote node host\n\nv2 \\u2014 agent speaks in-meeting\n audio_bridge.py: PulseAudio null-sink (Linux) + BlackHole probe (macOS).\n On Linux we load pactl module-null-sink + module-virtual-source, track\n module ids for teardown; Chrome gets PULSE_SOURCE= env so its\n fake mic reads what we write to the sink. macOS just probes BlackHole\n 2ch and returns its device name \\u2014 the plugin refuses to switch the\n user's default audio input (that would surprise them).\n realtime/openai_client.py: sync WebSocket client for the OpenAI Realtime\n API. RealtimeSession.speak(text) sends conversation.item.create +\n response.create, accumulates response.audio.delta PCM bytes, appends\n them to a file. RealtimeSpeaker runs a JSONL-queue loop consuming\n meet_say calls. 'websockets' is an optional dep imported lazily.\n meet_bot.py: when HERMES_MEET_MODE=realtime, provisions AudioBridge,\n starts RealtimeSession + speaker thread, spawns paplay to pump PCM\n into the null-sink, then cleans everything up on SIGTERM. If any\n realtime setup step fails, falls back cleanly to transcribe mode\n with an error flagged in status.json.\n process_manager.enqueue_say(): writes a JSONL line to say_queue.jsonl;\n refuses when no active meeting or active meeting is transcribe-only.\n tools.meet_say: real implementation; requires active mode='realtime'.\n meet_join: adds mode='transcribe'|'realtime' param.\n\nv3 \\u2014 remote node host\n node/protocol.py: JSON envelope (type, id, token, payload) + validate.\n node/registry.py: $HERMES_HOME/workspace/meetings/nodes.json, with\n resolve() auto-selecting the sole registered node when name is None.\n node/server.py: NodeServer \\u2014 websockets.serve, bearer-token auth,\n dispatches start_bot/stop/status/transcript/say/ping onto the local\n process_manager. Token auto-generated + persisted on first run.\n node/client.py: NodeClient \\u2014 short-lived sync WS per RPC, raises\n RuntimeError on error envelopes, clean API matching the server.\n node/cli.py: 'hermes meet node {run,list,approve,remove,status,ping}'\n subtree; wired into the main meet CLI by cli.py so 'hermes meet node'\n Just Works.\n tools.py: every meet_* tool accepts node=''|'auto'; when set,\n routes through NodeClient to the remote bot instead of running\n locally. Unknown node \\u2192 clear 'no registered meet node matches ...'\n error.\n cli.py: 'hermes meet join --node my-mac --mode realtime' and\n 'hermes meet say \"...\" --node my-mac' route to the node; 'hermes\n meet node approve ' registers one.\n\nTests\n 21 v1 tests updated (meet_say is no longer a stub; active-record now\n carries mode).\n 20 new audio_bridge + realtime tests.\n 42 new node tests (protocol/registry/server/client/cli).\n 17 new v1/v2/v3 integration tests at the plugin level covering\n enqueue_say edge cases, env var passthrough, mode validation, node\n routing (known/unknown/auto/ambiguous), and argparse wiring for\n `hermes meet say` + `hermes meet node` + --mode/--node flags.\n Total: 100 plugin tests + 58 plugin-system tests = 158 passing.\n\nE2E verified on Linux with fresh HERMES_HOME: plugin loads, 5 tools\nregister, on_session_end hook wires, 'hermes meet' CLI tree wires\nincluding the node subtree, NodeRegistry round-trips, meet_join routes\ncorrectly to NodeClient under node='my-mac' with mode='realtime',\nenqueue_say accepts realtime/rejects transcribe, argparse parses every\nnew flag cleanly.\n\nZero changes to core. All new code lives under plugins/google_meet/.\n\n* feat(plugins/google_meet): auto-install, admission detect, mac PCM pump, barge-in, richer status\n\nReady-for-live-test follow-up on PR #16364. Five additions that matter for\nthe first live run on a real Meet, in priority order:\n\n1. hermes meet install [--realtime] [--yes]\n pip install playwright websockets + python -m playwright install chromium\n --realtime: installs platform audio deps (pulseaudio-utils on Linux via\n sudo apt, blackhole-2ch + ffmpeg on macOS via brew). Prompts before\n sudo/brew unless --yes. Refuses on Windows. Refuses to auto-flip the\n macOS default input — user still selects BlackHole in System Settings\n (deliberate; surprise audio rerouting is worse than a manual step).\n\n2. Admission detection\n _detect_admission(page): Leave-button visible OR caption region\n attached OR participants list present → we're in-call.\n _detect_denied(page): 'You can\\'t join this video call' / 'You were\n removed' / 'No one responded to your request' → bail out.\n HERMES_MEET_LOBBY_TIMEOUT (default 300s) caps how long we sit in\n the lobby before giving up. in_call stays False until admitted.\n Status surfaces leaveReason: duration_expired | lobby_timeout |\n denied | page_closed.\n\n3. macOS PCM pump\n ffmpeg reads speaker.pcm (24kHz s16le mono) and writes to the\n BlackHole AVFoundation output via -f audiotoolbox\n -audio_device_index . _mac_audio_device_index() probes\n ffmpeg -f avfoundation -list_devices true to resolve 'BlackHole 2ch'\n → numeric index. Falls back to index 0 on probe failure. Linux\n paplay pump unchanged.\n\n4. Richer status dict\n _BotState now tracks realtime, realtimeReady, realtimeDevice,\n audioBytesOut, lastAudioOutAt, lastBargeInAt, joinAttemptedAt,\n leaveReason. RealtimeSession.audio_bytes_out / last_audio_out_at\n counters fold into the status file once a second so meet_status()\n can show the agent's voice activity in near-real-time.\n\n5. Barge-in\n RealtimeSession.cancel_response() sends type='response.cancel' over\n the same WS (lock-guarded so it's safe to call from the caption\n thread while speak() is reading frames). Handles response.cancelled\n as a terminal frame type. _looks_like_human_speaker() gates triggers\n so the bot's own name, 'You', 'Unknown', and blanks don't self-cancel.\n Called from the caption drain loop: when a new caption arrives\n attributed to a real participant while rt.session exists, we fire\n cancel_response() and stamp lastBargeInAt.\n\nTests: 20 new unit tests across _BotState telemetry, barge-in gating,\nadmission/denied probe error handling, cancel_response with and without\na connected WS, and `hermes meet install` CLI wiring (flag parsing +\nend-to-end subprocess.run verification + Linux-already-installed fast\npath). Total 171 passing across all google_meet test files + the\nplugin-system regression suite.\n\nE2E verified on Linux: plugin loads, all 5 tools register,\n`hermes meet install --realtime --yes` parses, fresh-bot status.json\nhas every new telemetry key, cancel_response on a disconnected session\nreturns False without raising, barge-in helper gates the bot's own\nname correctly.\n\nStill out of scope (for a future PR, not blocking live test):\nmic → Realtime duplex (the agent listening to meeting audio via\nWebRTC), node-host TLS/pairing UX, Windows audio, Meet create+Twilio.\n\nDocs updated: SKILL.md now lists the installer subcommand, lobby\ntimeout, barge-in caveat, and the full status-dict reference table.\nREADME.md quick-start uses hermes meet install.","tree":"5ff3d2af72d075acc723c1642d1c1f101d0ac301"},"mode":"0100644","name":"registry.py","size":3627},{"lastCommit":{"author":{"email":"127238744+teknium1@users.noreply.github.com","name":"Teknium","when":"2026-04-27T06:22:25.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-04-27T06:22:25.000000-07:00"},"hash":"df3c9593f8822762b5b4bbe30b582b62578d04d1","message":"feat(plugins): google_meet \\u2014 join, transcribe, speak, follow up (#16364)\n\n* feat(plugins): google_meet — bundled plugin for join+transcribe Meet calls\n\nv1 shipping transcribe-only. Spawns headless Chromium via Playwright,\njoins an explicit https://meet.google.com/ URL, enables live captions,\nand scrapes them into a transcript file the agent can read across turns.\nThe agent then has the meeting content in context and can do followup\nwork (send recap, file issues, schedule followups) with its regular tools.\n\nSurface:\n - Tools: meet_join, meet_status, meet_transcript, meet_leave, meet_say\n (meet_say is a v1 stub — returns not-implemented; v2 will wire\n realtime duplex audio via OpenAI Realtime / Gemini Live +\n BlackHole / PulseAudio null-sink.)\n - CLI: hermes meet setup | auth | join | status | transcript | stop\n - Lifecycle: on_session_end auto-leaves any still-running bot.\n\nSafety:\n - URL regex rejects anything that isn't https://meet.google.com/...\n - No calendar scanning, no auto-dial, no auto-consent announcement.\n - Single active meeting per install; a second meet_join leaves the first.\n - Platform-gated to Linux + macOS (Windows audio routing for v2 untested).\n - Opt-in: standalone plugin, user must add 'google_meet' to\n plugins.enabled in config.yaml.\n\nZero core changes. Plugin uses existing register_tool /\nregister_cli_command / register_hook surfaces. 21 new unit tests cover the\nURL safety gate, transcript dedup + status round-trip, process-manager\nrefusals/start/stop paths, tool-handler JSON shape under each branch,\nsession-end cleanup, and platform-gated register().\n\n* feat(plugins/google_meet): v2 realtime audio + v3 remote node host\n\nv2 \\u2014 agent speaks in-meeting\n audio_bridge.py: PulseAudio null-sink (Linux) + BlackHole probe (macOS).\n On Linux we load pactl module-null-sink + module-virtual-source, track\n module ids for teardown; Chrome gets PULSE_SOURCE= env so its\n fake mic reads what we write to the sink. macOS just probes BlackHole\n 2ch and returns its device name \\u2014 the plugin refuses to switch the\n user's default audio input (that would surprise them).\n realtime/openai_client.py: sync WebSocket client for the OpenAI Realtime\n API. RealtimeSession.speak(text) sends conversation.item.create +\n response.create, accumulates response.audio.delta PCM bytes, appends\n them to a file. RealtimeSpeaker runs a JSONL-queue loop consuming\n meet_say calls. 'websockets' is an optional dep imported lazily.\n meet_bot.py: when HERMES_MEET_MODE=realtime, provisions AudioBridge,\n starts RealtimeSession + speaker thread, spawns paplay to pump PCM\n into the null-sink, then cleans everything up on SIGTERM. If any\n realtime setup step fails, falls back cleanly to transcribe mode\n with an error flagged in status.json.\n process_manager.enqueue_say(): writes a JSONL line to say_queue.jsonl;\n refuses when no active meeting or active meeting is transcribe-only.\n tools.meet_say: real implementation; requires active mode='realtime'.\n meet_join: adds mode='transcribe'|'realtime' param.\n\nv3 \\u2014 remote node host\n node/protocol.py: JSON envelope (type, id, token, payload) + validate.\n node/registry.py: $HERMES_HOME/workspace/meetings/nodes.json, with\n resolve() auto-selecting the sole registered node when name is None.\n node/server.py: NodeServer \\u2014 websockets.serve, bearer-token auth,\n dispatches start_bot/stop/status/transcript/say/ping onto the local\n process_manager. Token auto-generated + persisted on first run.\n node/client.py: NodeClient \\u2014 short-lived sync WS per RPC, raises\n RuntimeError on error envelopes, clean API matching the server.\n node/cli.py: 'hermes meet node {run,list,approve,remove,status,ping}'\n subtree; wired into the main meet CLI by cli.py so 'hermes meet node'\n Just Works.\n tools.py: every meet_* tool accepts node=''|'auto'; when set,\n routes through NodeClient to the remote bot instead of running\n locally. Unknown node \\u2192 clear 'no registered meet node matches ...'\n error.\n cli.py: 'hermes meet join --node my-mac --mode realtime' and\n 'hermes meet say \"...\" --node my-mac' route to the node; 'hermes\n meet node approve ' registers one.\n\nTests\n 21 v1 tests updated (meet_say is no longer a stub; active-record now\n carries mode).\n 20 new audio_bridge + realtime tests.\n 42 new node tests (protocol/registry/server/client/cli).\n 17 new v1/v2/v3 integration tests at the plugin level covering\n enqueue_say edge cases, env var passthrough, mode validation, node\n routing (known/unknown/auto/ambiguous), and argparse wiring for\n `hermes meet say` + `hermes meet node` + --mode/--node flags.\n Total: 100 plugin tests + 58 plugin-system tests = 158 passing.\n\nE2E verified on Linux with fresh HERMES_HOME: plugin loads, 5 tools\nregister, on_session_end hook wires, 'hermes meet' CLI tree wires\nincluding the node subtree, NodeRegistry round-trips, meet_join routes\ncorrectly to NodeClient under node='my-mac' with mode='realtime',\nenqueue_say accepts realtime/rejects transcribe, argparse parses every\nnew flag cleanly.\n\nZero changes to core. All new code lives under plugins/google_meet/.\n\n* feat(plugins/google_meet): auto-install, admission detect, mac PCM pump, barge-in, richer status\n\nReady-for-live-test follow-up on PR #16364. Five additions that matter for\nthe first live run on a real Meet, in priority order:\n\n1. hermes meet install [--realtime] [--yes]\n pip install playwright websockets + python -m playwright install chromium\n --realtime: installs platform audio deps (pulseaudio-utils on Linux via\n sudo apt, blackhole-2ch + ffmpeg on macOS via brew). Prompts before\n sudo/brew unless --yes. Refuses on Windows. Refuses to auto-flip the\n macOS default input — user still selects BlackHole in System Settings\n (deliberate; surprise audio rerouting is worse than a manual step).\n\n2. Admission detection\n _detect_admission(page): Leave-button visible OR caption region\n attached OR participants list present → we're in-call.\n _detect_denied(page): 'You can\\'t join this video call' / 'You were\n removed' / 'No one responded to your request' → bail out.\n HERMES_MEET_LOBBY_TIMEOUT (default 300s) caps how long we sit in\n the lobby before giving up. in_call stays False until admitted.\n Status surfaces leaveReason: duration_expired | lobby_timeout |\n denied | page_closed.\n\n3. macOS PCM pump\n ffmpeg reads speaker.pcm (24kHz s16le mono) and writes to the\n BlackHole AVFoundation output via -f audiotoolbox\n -audio_device_index . _mac_audio_device_index() probes\n ffmpeg -f avfoundation -list_devices true to resolve 'BlackHole 2ch'\n → numeric index. Falls back to index 0 on probe failure. Linux\n paplay pump unchanged.\n\n4. Richer status dict\n _BotState now tracks realtime, realtimeReady, realtimeDevice,\n audioBytesOut, lastAudioOutAt, lastBargeInAt, joinAttemptedAt,\n leaveReason. RealtimeSession.audio_bytes_out / last_audio_out_at\n counters fold into the status file once a second so meet_status()\n can show the agent's voice activity in near-real-time.\n\n5. Barge-in\n RealtimeSession.cancel_response() sends type='response.cancel' over\n the same WS (lock-guarded so it's safe to call from the caption\n thread while speak() is reading frames). Handles response.cancelled\n as a terminal frame type. _looks_like_human_speaker() gates triggers\n so the bot's own name, 'You', 'Unknown', and blanks don't self-cancel.\n Called from the caption drain loop: when a new caption arrives\n attributed to a real participant while rt.session exists, we fire\n cancel_response() and stamp lastBargeInAt.\n\nTests: 20 new unit tests across _BotState telemetry, barge-in gating,\nadmission/denied probe error handling, cancel_response with and without\na connected WS, and `hermes meet install` CLI wiring (flag parsing +\nend-to-end subprocess.run verification + Linux-already-installed fast\npath). Total 171 passing across all google_meet test files + the\nplugin-system regression suite.\n\nE2E verified on Linux: plugin loads, all 5 tools register,\n`hermes meet install --realtime --yes` parses, fresh-bot status.json\nhas every new telemetry key, cancel_response on a disconnected session\nreturns False without raising, barge-in helper gates the bot's own\nname correctly.\n\nStill out of scope (for a future PR, not blocking live test):\nmic → Realtime duplex (the agent listening to meeting audio via\nWebRTC), node-host TLS/pairing UX, Windows audio, Meet create+Twilio.\n\nDocs updated: SKILL.md now lists the installer subcommand, lobby\ntimeout, barge-in caveat, and the full status-dict reference table.\nREADME.md quick-start uses hermes meet install.","tree":"5ff3d2af72d075acc723c1642d1c1f101d0ac301"},"mode":"0100644","name":"server.py","size":7634}],"lastCommit":{"author":{"email":"127238744+teknium1@users.noreply.github.com","name":"Teknium","when":"2026-04-27T06:22:25.000000-07:00"},"committer":{"email":"noreply@github.com","name":"GitHub","when":"2026-04-27T06:22:25.000000-07:00"},"hash":"df3c9593f8822762b5b4bbe30b582b62578d04d1","message":"feat(plugins): google_meet \\u2014 join, transcribe, speak, follow up (#16364)\n\n* feat(plugins): google_meet — bundled plugin for join+transcribe Meet calls\n\nv1 shipping transcribe-only. Spawns headless Chromium via Playwright,\njoins an explicit https://meet.google.com/ URL, enables live captions,\nand scrapes them into a transcript file the agent can read across turns.\nThe agent then has the meeting content in context and can do followup\nwork (send recap, file issues, schedule followups) with its regular tools.\n\nSurface:\n - Tools: meet_join, meet_status, meet_transcript, meet_leave, meet_say\n (meet_say is a v1 stub — returns not-implemented; v2 will wire\n realtime duplex audio via OpenAI Realtime / Gemini Live +\n BlackHole / PulseAudio null-sink.)\n - CLI: hermes meet setup | auth | join | status | transcript | stop\n - Lifecycle: on_session_end auto-leaves any still-running bot.\n\nSafety:\n - URL regex rejects anything that isn't https://meet.google.com/...\n - No calendar scanning, no auto-dial, no auto-consent announcement.\n - Single active meeting per install; a second meet_join leaves the first.\n - Platform-gated to Linux + macOS (Windows audio routing for v2 untested).\n - Opt-in: standalone plugin, user must add 'google_meet' to\n plugins.enabled in config.yaml.\n\nZero core changes. Plugin uses existing register_tool /\nregister_cli_command / register_hook surfaces. 21 new unit tests cover the\nURL safety gate, transcript dedup + status round-trip, process-manager\nrefusals/start/stop paths, tool-handler JSON shape under each branch,\nsession-end cleanup, and platform-gated register().\n\n* feat(plugins/google_meet): v2 realtime audio + v3 remote node host\n\nv2 \\u2014 agent speaks in-meeting\n audio_bridge.py: PulseAudio null-sink (Linux) + BlackHole probe (macOS).\n On Linux we load pactl module-null-sink + module-virtual-source, track\n module ids for teardown; Chrome gets PULSE_SOURCE= env so its\n fake mic reads what we write to the sink. macOS just probes BlackHole\n 2ch and returns its device name \\u2014 the plugin refuses to switch the\n user's default audio input (that would surprise them).\n realtime/openai_client.py: sync WebSocket client for the OpenAI Realtime\n API. RealtimeSession.speak(text) sends conversation.item.create +\n response.create, accumulates response.audio.delta PCM bytes, appends\n them to a file. RealtimeSpeaker runs a JSONL-queue loop consuming\n meet_say calls. 'websockets' is an optional dep imported lazily.\n meet_bot.py: when HERMES_MEET_MODE=realtime, provisions AudioBridge,\n starts RealtimeSession + speaker thread, spawns paplay to pump PCM\n into the null-sink, then cleans everything up on SIGTERM. If any\n realtime setup step fails, falls back cleanly to transcribe mode\n with an error flagged in status.json.\n process_manager.enqueue_say(): writes a JSONL line to say_queue.jsonl;\n refuses when no active meeting or active meeting is transcribe-only.\n tools.meet_say: real implementation; requires active mode='realtime'.\n meet_join: adds mode='transcribe'|'realtime' param.\n\nv3 \\u2014 remote node host\n node/protocol.py: JSON envelope (type, id, token, payload) + validate.\n node/registry.py: $HERMES_HOME/workspace/meetings/nodes.json, with\n resolve() auto-selecting the sole registered node when name is None.\n node/server.py: NodeServer \\u2014 websockets.serve, bearer-token auth,\n dispatches start_bot/stop/status/transcript/say/ping onto the local\n process_manager. Token auto-generated + persisted on first run.\n node/client.py: NodeClient \\u2014 short-lived sync WS per RPC, raises\n RuntimeError on error envelopes, clean API matching the server.\n node/cli.py: 'hermes meet node {run,list,approve,remove,status,ping}'\n subtree; wired into the main meet CLI by cli.py so 'hermes meet node'\n Just Works.\n tools.py: every meet_* tool accepts node=''|'auto'; when set,\n routes through NodeClient to the remote bot instead of running\n locally. Unknown node \\u2192 clear 'no registered meet node matches ...'\n error.\n cli.py: 'hermes meet join --node my-mac --mode realtime' and\n 'hermes meet say \"...\" --node my-mac' route to the node; 'hermes\n meet node approve ' registers one.\n\nTests\n 21 v1 tests updated (meet_say is no longer a stub; active-record now\n carries mode).\n 20 new audio_bridge + realtime tests.\n 42 new node tests (protocol/registry/server/client/cli).\n 17 new v1/v2/v3 integration tests at the plugin level covering\n enqueue_say edge cases, env var passthrough, mode validation, node\n routing (known/unknown/auto/ambiguous), and argparse wiring for\n `hermes meet say` + `hermes meet node` + --mode/--node flags.\n Total: 100 plugin tests + 58 plugin-system tests = 158 passing.\n\nE2E verified on Linux with fresh HERMES_HOME: plugin loads, 5 tools\nregister, on_session_end hook wires, 'hermes meet' CLI tree wires\nincluding the node subtree, NodeRegistry round-trips, meet_join routes\ncorrectly to NodeClient under node='my-mac' with mode='realtime',\nenqueue_say accepts realtime/rejects transcribe, argparse parses every\nnew flag cleanly.\n\nZero changes to core. All new code lives under plugins/google_meet/.\n\n* feat(plugins/google_meet): auto-install, admission detect, mac PCM pump, barge-in, richer status\n\nReady-for-live-test follow-up on PR #16364. Five additions that matter for\nthe first live run on a real Meet, in priority order:\n\n1. hermes meet install [--realtime] [--yes]\n pip install playwright websockets + python -m playwright install chromium\n --realtime: installs platform audio deps (pulseaudio-utils on Linux via\n sudo apt, blackhole-2ch + ffmpeg on macOS via brew). Prompts before\n sudo/brew unless --yes. Refuses on Windows. Refuses to auto-flip the\n macOS default input — user still selects BlackHole in System Settings\n (deliberate; surprise audio rerouting is worse than a manual step).\n\n2. Admission detection\n _detect_admission(page): Leave-button visible OR caption region\n attached OR participants list present → we're in-call.\n _detect_denied(page): 'You can\\'t join this video call' / 'You were\n removed' / 'No one responded to your request' → bail out.\n HERMES_MEET_LOBBY_TIMEOUT (default 300s) caps how long we sit in\n the lobby before giving up. in_call stays False until admitted.\n Status surfaces leaveReason: duration_expired | lobby_timeout |\n denied | page_closed.\n\n3. macOS PCM pump\n ffmpeg reads speaker.pcm (24kHz s16le mono) and writes to the\n BlackHole AVFoundation output via -f audiotoolbox\n -audio_device_index . _mac_audio_device_index() probes\n ffmpeg -f avfoundation -list_devices true to resolve 'BlackHole 2ch'\n → numeric index. Falls back to index 0 on probe failure. Linux\n paplay pump unchanged.\n\n4. Richer status dict\n _BotState now tracks realtime, realtimeReady, realtimeDevice,\n audioBytesOut, lastAudioOutAt, lastBargeInAt, joinAttemptedAt,\n leaveReason. RealtimeSession.audio_bytes_out / last_audio_out_at\n counters fold into the status file once a second so meet_status()\n can show the agent's voice activity in near-real-time.\n\n5. Barge-in\n RealtimeSession.cancel_response() sends type='response.cancel' over\n the same WS (lock-guarded so it's safe to call from the caption\n thread while speak() is reading frames). Handles response.cancelled\n as a terminal frame type. _looks_like_human_speaker() gates triggers\n so the bot's own name, 'You', 'Unknown', and blanks don't self-cancel.\n Called from the caption drain loop: when a new caption arrives\n attributed to a real participant while rt.session exists, we fire\n cancel_response() and stamp lastBargeInAt.\n\nTests: 20 new unit tests across _BotState telemetry, barge-in gating,\nadmission/denied probe error handling, cancel_response with and without\na connected WS, and `hermes meet install` CLI wiring (flag parsing +\nend-to-end subprocess.run verification + Linux-already-installed fast\npath). Total 171 passing across all google_meet test files + the\nplugin-system regression suite.\n\nE2E verified on Linux: plugin loads, all 5 tools register,\n`hermes meet install --realtime --yes` parses, fresh-bot status.json\nhas every new telemetry key, cancel_response on a disconnected session\nreturns False without raising, barge-in helper gates the bot's own\nname correctly.\n\nStill out of scope (for a future PR, not blocking live test):\nmic → Realtime duplex (the agent listening to meeting audio via\nWebRTC), node-host TLS/pairing UX, Windows audio, Meet create+Twilio.\n\nDocs updated: SKILL.md now lists the installer subcommand, lobby\ntimeout, barge-in caveat, and the full status-dict reference table.\nREADME.md quick-start uses hermes meet install.","tree":"5ff3d2af72d075acc723c1642d1c1f101d0ac301"}}