vm-cleanup - SKILL.md Agent Skill

name: vm-cleanup description: This skill should be used when the user asks to "clean up", "free disk space", "clean the VM", "liberer de la memoire", "nettoyer", or when disk usage exceeds 80%. Also trigger proactively after heavy build/test sessions that generate large artifacts. allowed-tools: Bash, Read, Glob

VM Cleanup Skill

Clean up disk space on the development VM by removing build artifacts, caches, stale containers, and temporary files related to the VarunaPoC project. Designed for the self-hosted runner environment.

When to Trigger

User explicitly asks for cleanup
Disk usage on / exceeds 80% (check with df -h /)
After heavy Docker build sessions
After running E2E test suites (Playwright generates large artifacts)
After multiple cargo build cycles (target dirs grow fast)

Cleanup Procedure

Phase 1: Assess

Run the diagnostic script to see current disk state before taking action:

bash .claude/skills/vm-cleanup/scripts/disk-report.sh

Present the report to the user. Only proceed with cleanup if disk usage warrants it or if the user explicitly requested it.

Phase 2: Safe Cleanup (always safe to run)

These operations never break anything:

Python caches: find . -type d -name __pycache__ -exec rm -rf {} + 2>/dev/null
Pytest/Ruff caches: rm -rf .pytest_cache .ruff_cache
Temp files: rm -f /tmp/security-scan-log.txt /tmp/pr*-logs.txt
Old logs: find /var/log -name "*.gz" -o -name "*.old" -o -name "*.[0-9]" | xargs rm -f 2>/dev/null
Pip cache: rm -rf /root/.cache/pip
npm cache: npm cache clean --force 2>/dev/null
Playwright artifacts: rm -rf frontend/test-results frontend/e2e/test-results frontend/e2e/playwright-report
Build outputs: rm -rf frontend/dist

Phase 3: Medium Impact (ask user first)

These free significant space but may require rebuild time:

Docker prune: docker system prune -f (removes stopped containers, dangling images)
Cargo target clean: cd native/quiche-cloudflare && cargo clean (can be several GB)
Node modules: rm -rf frontend/node_modules (requires npm ci to restore)

Phase 4: High Impact (always ask user first)

These free the most space but have significant rebuild cost:

Docker full prune: docker system prune -a -f --volumes (removes ALL unused images/volumes)
Python venvs: rm -rf backend/venv backend/.venv (requires pip install -r requirements.txt)
Rust toolchain cache: rm -rf /root/.cargo/registry/cache (re-downloads on next build)

Post-Cleanup

After cleanup, run the diagnostic script again to confirm space was freed:

bash .claude/skills/vm-cleanup/scripts/disk-report.sh

Report the before/after comparison to the user.

Important Notes

Never delete Slides/ directory (60+ GB of medical imaging data)
Never delete .claude/ tracked files (project memory/config)
Never delete backend/data/ (SQLite databases, uploaded files)
The backend/venv and backend/.venv are gitignored and can always be recreated
Docker images can always be rebuilt from Dockerfiles