メニュー
Use after any change to server/, app/, or voice_forge/ to run the repo's only verification paths. There is no CI and no server test suite — these commands ARE the gates.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/iamudesharma/clinicguard --skill verify-stackコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Use when adding, changing, or debugging data-channel messages between the client and the agent — topic agent.events. The contract lives in THREE places that must stay in sync: server/agent.py, app/lib/state/call_state.dart, and voice_forge (packages/voice_forge/lib/src/session/agent_session.dart + examples/poc_server).
Use when adding or switching an STT, TTS, or LLM backend (STT_BACKEND, TTS_BACKEND, LLM_BACKEND) in the ClinicGuard server, or adding provider-specific settings to server/config.py. Follow this checklist in order — each step is cross-coupled with the next.
Use when working with Supabase in this repo — schema.sql, store.py queries, realtime subscriptions, RLS policies, or writing SQL queries against ClinicGuard data. Covers which key to use where, what's demo-only, and the store.py fallback behavior.
SKILL.md を表示中
| name | verify-stack |
| description | Use after any change to server/, app/, or voice_forge/ to run the repo's only verification paths. There is no CI and no server test suite — these commands ARE the gates. |
Run only the components you touched; run all three if unsure. Each component has exactly one smoke path — do not invent others (e.g. there is no pytest).
server/)cd server && uv sync # after touching pyproject.toml
uv run python agent.py console # fastest: full voice loop in terminal
uv run uvicorn api.main:app --reload --host 0.0.0.0 --port 8000 &
curl localhost:8000/health # -> healthy JSON
Cold start is 5-40s (Whisper/Piper models prewarm once per process — JobExecutorType.THREAD). First turn is the slow one.
app/)cd app && flutter test && flutter analyze
Tests are only test/barge_in_detector_test.dart + test/widget_test.dart. flutter_lints is the only linting. Device runs: simulator has no mic; physical iPhone needs --dart-define=API_BASE_URL=http://<mac-lan-ip>:8000.
voice_forge/examples/poc_server)dart run bin/agent_self_test.dart # multi-turn agent loop -> must end "RESULT: PASS"
dart run bin/speech_check.dart # offline VAD/STT/TTS check -> "RESULT: PASS"
dart run bin/self_test.dart # WebRTC transport loopback -> "RESULT: PASS"
Plus, in examples/poc_client AND packages/voice_forge_flutter:
flutter test && flutter analyze
Gates: every self-test must end with RESULT: PASS; any failure means the change is not done. The agent needs models/ + third_party/native/ (fetched by scripts/fetch_models.sh / fetch_native.sh).