AndrewSmigaj
AndrewSmigaj/OpenLLMRI
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/AndrewSmigaj/OpenLLMRI --skill server命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Uncertainty assessment before implementation — identify what you know, what you're guessing, and what to verify
Generate 2-3 genuinely different approaches to a problem and compare them
Challenge a design or approach — find real weaknesses, not performative objections
基于 SOC 职业分类
正在显示 SKILL.md
| name | server |
| description | Start, stop, and check status of backend and frontend servers |
Manage the Concept MRI backend (FastAPI + model) and frontend (Vite) servers.
| Constant | Value |
|---|---|
| Project root | /mnt/c/Users/emily/OpenAIHackathon-ConceptMRI |
| Python | /mnt/c/Users/emily/OpenAIHackathon-ConceptMRI/.venv/bin/python |
| Backend working dir | backend/src (relative to project root) |
| Backend URL | http://localhost:8000 |
| Frontend URL | http://localhost:5173 |
| Health endpoint | http://localhost:8000/health |
| Host binding | 0.0.0.0 (required for WSL2) |
NEVER use bare python3 — always use the full venv path above.
Each operation below is a single self-contained block. Copy the EXACT block — do not improvise or compose steps from multiple blocks.
Run this FIRST before any other operation to understand current state.
echo "=== Processes ===" && ps aux | grep -E "uvicorn|vite" | grep -v grep || echo "(none running)" && echo "=== Ports ===" && (fuser 8000/tcp 2>/dev/null && echo "8000: IN USE" || echo "8000: free") && (fuser 5173/tcp 2>/dev/null && echo "5173: IN USE" || echo "5173: free") && echo "=== Backend Health ===" && curl -s --max-time 3 http://localhost:8000/health 2>/dev/null | /mnt/c/Users/emily/OpenAIHackathon-ConceptMRI/.venv/bin/python -c "import json,sys; d=json.load(sys.stdin); s=d.get('loading',{}).get('stage','?'); e=d.get('loading',{}).get('elapsed_seconds'); print(f'Model loaded — ready' if d.get('model_loaded') else f'Stage: {s} ({e}s elapsed)' if e else f'Stage: {s}')" 2>/dev/null || echo "Not responding" && echo "=== Frontend ===" && curl -s -o /dev/null -w "HTTP %{http_code}" http://localhost:5173 2>/dev/null || echo "Not responding"
Use fuser -k (kills by port) — this is reliable on WSL2. pkill is NOT reliable here.
fuser -k 8000/tcp 2>/dev/null; fuser -k 5173/tcp 2>/dev/null; sleep 2 && echo "=== Verify ===" && (fuser 8000/tcp 2>/dev/null && echo "8000: STILL IN USE" || echo "8000: free") && (fuser 5173/tcp 2>/dev/null && echo "5173: STILL IN USE" || echo "5173: free")
If a port shows "STILL IN USE" after this, run fuser -k -9 <port>/tcp (SIGKILL).
Prerequisite: Port 8000 must be free (run OP-2 first if needed).
cd /mnt/c/Users/emily/OpenAIHackathon-ConceptMRI/backend/src && /mnt/c/Users/emily/OpenAIHackathon-ConceptMRI/.venv/bin/python -m uvicorn api.main:app --host 0.0.0.0 --port 8000 --reload
Run with run_in_background: true.
WSL2 note: On WSL2 with /mnt/c/ NTFS filesystem, inotify is unreliable. Existing endpoint edits usually reload fine, but new endpoints/routes may not be detected. If a new endpoint returns 404, do a full restart (OP-2 + OP-3). See TROUBLESHOOTING.md.
Run AFTER OP-3. Must use run_in_background: true — takes ~2-3 minutes.
The health endpoint now reports loading stage in real time (the API serves immediately while model loads in background). No need to read log files.
PY=/mnt/c/Users/emily/OpenAIHackathon-ConceptMRI/.venv/bin/python; for i in $(seq 1 60); do H=$(curl -s --max-time 3 http://localhost:8000/health 2>/dev/null); if [ -z "$H" ]; then echo "[$i] Waiting for API..."; sleep 5; continue; fi; STAGE=$(echo "$H" | $PY -c "import json,sys; print(json.load(sys.stdin).get('loading',{}).get('stage','unknown'))"); ELAPSED=$(echo "$H" | $PY -c "import json,sys; print(json.load(sys.stdin).get('loading',{}).get('elapsed_seconds','?'))"); if [ "$STAGE" = "ready" ]; then echo "READY — model loaded in ${ELAPSED}s"; exit 0; fi; if [ "$STAGE" = "failed" ]; then echo "FAILED — check backend logs"; exit 1; fi; echo "[$i] Stage: $STAGE (${ELAPSED}s elapsed)"; sleep 5; done; echo
Expected output:
[1] Waiting for API...
[2] Stage: initializing (3.2s elapsed)
[3] Stage: loading_model (8.1s elapsed)
...
[24] Stage: loading_model (118.5s elapsed)
[25] Stage: creating_service (121.0s elapsed)
[26] READY — model loaded in 123.4s
Stages: not_started → initializing → loading_model → creating_service → ready | failed
Prerequisite: Port 5173 must be free.
cd /mnt/c/Users/emily/OpenAIHackathon-ConceptMRI/frontend && npm run dev
Run with run_in_background: true. Vite uses strictPort: true — will error if 5173 is taken.
This is the typical workflow after code changes or when the backend needs a fresh start.
run_in_background: true)run_in_background: true) — can launch in parallel with OP-3run_in_background: true)Only needed if Vite HMR stops working (rare).
fuser -k 5173/tcp 2>/dev/null; sleep 1; cd /mnt/c/Users/emily/OpenAIHackathon-ConceptMRI/frontend && npm run dev
Run with run_in_background: true.
| Change made | Action needed |
|---|---|
Frontend .tsx/.ts edit | None — Vite HMR handles it |
Backend .py edit (existing endpoint) | Usually none (--reload). If stale, full restart |
| Backend new endpoint or route | ALWAYS full restart (WSL2 inotify unreliable) |
| Backend dependency added | Full restart |
Frontend dependency added (npm install) | Restart Vite only |
fuser -k to stop, not pkill — pkill is unreliable on WSL2model_loaded: true before making API calls that need the model0.0.0.0 (not 127.0.0.1) for WSL2 networkingpython3 — always /mnt/c/Users/emily/OpenAIHackathon-ConceptMRI/.venv/bin/python