diff --git a/.moss_ws/apps/sensors/speech_guard/.gitignore b/.moss_ws/apps/sensors/speech_guard/.gitignore new file mode 100644 index 00000000..cf94f816 --- /dev/null +++ b/.moss_ws/apps/sensors/speech_guard/.gitignore @@ -0,0 +1,6 @@ +*.log +venv/ +.venv/ +uv.lock +.idea +.env \ No newline at end of file diff --git a/.moss_ws/apps/sensors/speech_guard/APP.md b/.moss_ws/apps/sensors/speech_guard/APP.md new file mode 100644 index 00000000..7d1af6ab --- /dev/null +++ b/.moss_ws/apps/sensors/speech_guard/APP.md @@ -0,0 +1,8 @@ +--- +arguments: '' +description: '' +executable: uv +respawn: false +script: main.py +workers: 1 +--- \ No newline at end of file diff --git a/.moss_ws/apps/sensors/speech_guard/CLAUDE.md b/.moss_ws/apps/sensors/speech_guard/CLAUDE.md new file mode 100644 index 00000000..bea3d329 --- /dev/null +++ b/.moss_ws/apps/sensors/speech_guard/CLAUDE.md @@ -0,0 +1,75 @@ +# MOSS App Development + +You are working inside a MOSS App — an independent, process-isolated unit deliverable +through the MOSS app system. Think of this as its own small project. + +## Project mindset + +- This app is a standalone unit. It may be downloaded from a hub or shared independently. +- Tests live in `tests/` under this directory, NOT in the main project's `tests/`. +- Dependencies go in `pyproject.toml`; the app gets its own `.venv` via `uv run`. + +## Directory layout + +``` +apps/// +├── APP.md # metadata — MOSS discovers the app through this +├── CLAUDE.md # this file — AI developer context +├── main.py # entry point +├── pyproject.toml # optional — independent dependencies +├── src/ # optional — multi-module app code +├── tests/ # tests live here, NOT in the main project +└── runtime/ # runtime data (auto-created by MOSS) + ├── assets/ + ├── configs/ + └── logs/ +``` + +When you have a `pyproject.toml`, treat this as a proper project — source in `src/`, +tests in `tests/`, managed by `uv run`. The `runtime/` directory is a MOSS convention, +auto-managed by the framework. + +## Dependency management + +App dependencies are managed by `uv run` — it creates the app's own venv automatically. +`moss apps start` and `moss apps test` both use `uv run` under the hood. + +For the full decision framework (three isolation levels, when to use pyproject.toml vs +PEP 723 vs shared runtime), see `moss howtos read app-dev/build-an-app`. + +## Testing + +Put tests in `tests/` under this app directory. Use `test_channel()` for the common case:: + +```python +import pytest +from ghoshell_moss.core.blueprint.channel_builder import new_channel, test_channel + +@pytest.mark.asyncio +async def test_my_command(): + chan = new_channel(name="my_app") + + @chan.build.command() + async def ping() -> str: + return "pong" + + tasks = await test_channel(chan, ctml='') + assert await tasks[0] == "pong" +``` + +Run with: `uv run pytest tests/ -v` + +This is the baseline. For scopes, observe, cancel, nested channels, or complex CTML +scenarios, `test_channel` is not enough — read `moss howtos read app-dev/test-an-app`. + +## Common mistakes + +- **`uv sync --active` in this directory** — pollutes the main project's venv. + App dependencies are isolated. Use `uv run` or `moss apps test` instead. +- **Creating apps by hand (`mkdir apps/xxx`)** — skips this template and the + runtime directory scaffold. Always use `moss apps create`. +- **Writing tests under the main project's `tests/`** — apps are independent + projects. Tests live here, in this app's own `tests/` directory. + +--- +*Generated by moss apps create. Last updated 2026-06-07.* diff --git a/.moss_ws/apps/sensors/speech_guard/main.py b/.moss_ws/apps/sensors/speech_guard/main.py new file mode 100644 index 00000000..ae03a98a --- /dev/null +++ b/.moss_ws/apps/sensors/speech_guard/main.py @@ -0,0 +1,45 @@ +"""Speech Streaming Monitor — 订阅 speech/streaming Topic，终端逐句显示。 + +验证 SpeechStreamingTopic 跨进程发布链路是否正常。 +运行方式：moss apps test speech/streaming_monitor + +启动后等待 Ghost TTS 输出（另一终端运行 moss-run-ghost echo --mode show）， +本终端实时显示每句文本及批次边界。 +""" + +import asyncio +import time + +from ghoshell_moss.core.blueprint.matrix import Matrix +from ghoshell_moss.topics.audio import SpeechTopic + + +async def main(matrix: Matrix) -> None: + print("Speech Streaming Monitor") + print(f" topic : {SpeechTopic.default_topic_name()}") + print(f" type : {SpeechTopic.topic_type()}") + print(" 等待 Ghost TTS 输出...\n") + + window = matrix.session.topics.create_window_for( + SpeechTopic, max_size=200 + ) + + t0 = time.monotonic() + seen = 0 + batch_count = 0 + + try: + while True: + await asyncio.sleep(0.12) + items = window.values() + new_items = items[seen:] + for item in new_items: + elapsed = time.monotonic() - t0 + print(f" [{elapsed:6.1f}s] │ {item.text}") + seen = len(items) + except KeyboardInterrupt: + print(f"\n 共接收 {seen} 条事件，{batch_count} 个批次") + + +if __name__ == "__main__": + Matrix.discover().run(main) \ No newline at end of file diff --git a/.moss_ws/src/MOSS/modes/show/MODE.md b/.moss_ws/src/MOSS/modes/show/MODE.md index 4fefa57f..de8bf388 100644 --- a/.moss_ws/src/MOSS/modes/show/MODE.md +++ b/.moss_ws/src/MOSS/modes/show/MODE.md @@ -1,10 +1,7 @@ --- apps: ["browsers/*", "games/*", "tools/*", "ui/*", "sensors/*"] bringup_apps: [ - "tools/screen_capture", - "ui/reflex", - "sensors/audio_capture", "sensors/listener", - "games/ai_eye", "sensors/vision", +# "sensors/speech_guard" ] ctml_version: '' description: '' diff --git a/.moss_ws/src/MOSS/modes/show/channels.py b/.moss_ws/src/MOSS/modes/show/channels.py index 6603afc8..98078261 100644 --- a/.moss_ws/src/MOSS/modes/show/channels.py +++ b/.moss_ws/src/MOSS/modes/show/channels.py @@ -18,9 +18,12 @@ from ghoshell_moss.channels.terminal_channel import new_terminal_channel from ghoshell_moss.channels.mermaid_draw import new_mermaid_channel from ghoshell_moss.channels.web_bookmark import new_web_bookmark_channel +from ghoshell_moss.core import SpeechChannelModule main = new_default_shell_main_channel() +main.with_module(SpeechChannelModule()) + main.import_channels( AppStoreChannel(name='apps'), new_mac_control_channel(name='mac'), diff --git a/src/ghoshell_moss/core/concepts/interpreter.py b/src/ghoshell_moss/core/concepts/interpreter.py index 1099bcc2..7d81a06e 100644 --- a/src/ghoshell_moss/core/concepts/interpreter.py +++ b/src/ghoshell_moss/core/concepts/interpreter.py @@ -727,7 +727,6 @@ def empty_stopped(): tasks = parser.on_token(item) if tasks is not None: for task in tasks: - task.on_compiled() task_callback(task) await asyncio.sleep(0.0) except asyncio.CancelledError: diff --git a/src/ghoshell_moss/core/runtime/_base_channel_runtime.py b/src/ghoshell_moss/core/runtime/_base_channel_runtime.py index 9251b284..f3e480eb 100644 --- a/src/ghoshell_moss/core/runtime/_base_channel_runtime.py +++ b/src/ghoshell_moss/core/runtime/_base_channel_runtime.py @@ -520,7 +520,9 @@ async def _compiled_task_loop(self) -> None: continue if not task.is_bare_task(): # 只有非 bare 才执行 on compiled. - task.on_compiled() + # 需要注入 runtime context，因为 partial 函数可能使用 CommandUtil。 + with ChannelCtx(self, task).in_ctx(): + task.on_compiled() # prepare to send await self._consume_compiled_task_with_paths(paths, task) diff --git a/src/ghoshell_moss/core/speech/speech_module.py b/src/ghoshell_moss/core/speech/speech_module.py index 176b60ff..aab2a88b 100644 --- a/src/ghoshell_moss/core/speech/speech_module.py +++ b/src/ghoshell_moss/core/speech/speech_module.py @@ -6,7 +6,7 @@ from ghoshell_moss.core.concepts.command import Command, PyCommand from ghoshell_moss.contracts.speech import Speech, SpeechStream, TTSSpeech from ghoshell_moss.core.speech.null import NullSpeech -from ghoshell_moss.topics import SpeechTopic, Publisher +from ghoshell_moss.topics import SpeechTopic def build_content_command(speech: Speech) -> Command: @@ -26,38 +26,36 @@ def __init__( *, role: str = '', name: str = '', - publisher: Publisher[SpeechTopic] | None = None, ): self._speech = speech self._role = role self._name = name - self._publisher = publisher def build_content_command(self) -> Command: speech = self._speech async def _feed_stream(stream: SpeechStream, deltas): - content = '' - try: - if not speech.is_running(): - return - has_first_chunk = False - async for chunk in deltas: - if not has_first_chunk and chunk.strip(): - has_first_chunk = True - await stream.start_synthesis() - stream.feed(chunk) - content += chunk - stream.commit() - except asyncio.CancelledError: - await stream.close() - finally: - if self._publisher: - self._publisher.pub(SpeechTopic( - speaker_name=self._name, - role=self._role, - text=content, - )) + publisher = CommandUtil.topic_publisher(SpeechTopic) + async with publisher: + content = '' + try: + if not speech.is_running(): + return + has_first_chunk = False + async for chunk in deltas: + if not has_first_chunk and chunk.strip(): + has_first_chunk = True + await stream.start_synthesis() + stream.feed(chunk) + content += chunk + publisher.pub(SpeechTopic( + speaker_name=self._name, + role=self._role, + text=content, + )) + stream.commit() + except asyncio.CancelledError: + await stream.close() async def _content_partial(chunks__): if not speech.is_running(): @@ -124,17 +122,26 @@ async def say_partial( stream = tts_speech.new_tts_stream(batch) async def run_tts_batch() -> None: - try: - nonlocal chunks__ - await stream.start_synthesis() - async for chunk in chunks__: - if stream.is_closed(): - return - stream.feed(chunk) - except Exception as e: - await stream.fail(e) - finally: - stream.commit() + publisher = CommandUtil.topic_publisher(SpeechTopic) + async with publisher: + try: + nonlocal chunks__ + await stream.start_synthesis() + content = "" + async for chunk in chunks__: + if stream.is_closed(): + return + stream.feed(chunk) + content += chunk + publisher.pub(SpeechTopic( + speaker_name=self._name, + role=self._role, + text=content, + )) + except Exception as e: + await stream.fail(e) + finally: + stream.commit() _ = asyncio.create_task(run_tts_batch()) return [], dict(voice=voice, chunks__=stream, as_default=as_default) @@ -166,15 +173,12 @@ def __init__( register_content: bool = False, role: str = 'ghost', name: str = '', - publishing: bool = False, ): self._speech: Speech | None = None self._own_commands = {} self._register_content = register_content self._role = role self._name = name - self._publishing = publishing - self._publisher: Publisher[SpeechTopic] | None = None def name(self) -> str: return "speech" @@ -185,15 +189,11 @@ def own_commands(self) -> dict[str, Command]: async def on_startup(self) -> None: if CommandUtil.enabled(): self._speech = CommandUtil.get_contract(Speech) - if self._publishing: - self._publisher = CommandUtil.topic_publisher(SpeechTopic) - await self._publisher.__aenter__() self._speech = self._speech or NullSpeech() factory = _SpeechCommandFactory( self._speech, role=self._role, name=self._name, - publisher=self._publisher, ) commands = {} if self._register_content: @@ -206,5 +206,3 @@ async def on_startup(self) -> None: async def on_close(self) -> None: self._speech = None - if self._publisher: - await self._publisher.__aexit__(None, None, None)