Safeclaw Integration

SafeClaw Integration

Run isolated Claude Code sessions inside Docker containers for YOLO mode, overnight builds, and untrusted code execution.

Why SafeClaw

• Isolation: Code runs in a container, not on your host
• Resource limits: CPU, memory, and disk caps prevent runaway sessions
• Reproducibility: Same container image every time
• Safety: YOLO mode (--dangerously-skip-permissions) is safe inside a container
• Cleanup: Container is destroyed after session ends

Single Container Session

docker run --rm -it \
  --name safeclaw-session \
  --cpus="2" \
  --memory="4g" \
  --pids-limit=256 \
  -v "$(pwd):/workspace:rw" \
  -v "$HOME/.claude:/home/coder/.claude:ro" \
  -e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
  -w /workspace \
  ghcr.io/anthropics/claude-code:latest \
  claude -p "Execute the plan in tasks/plan.md" \
  --dangerously-skip-permissions \
  --max-turns 100

Parallel Containers

Run multiple isolated sessions simultaneously:

# Session 1: Frontend
docker run --rm -d \
  --name safeclaw-frontend \
  --cpus="1.5" --memory="3g" \
  -v "$(pwd):/workspace:rw" \
  -v "$HOME/.claude:/home/coder/.claude:ro" \
  -e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
  -w /workspace \
  ghcr.io/anthropics/claude-code:latest \
  claude -p "Build the frontend components" --dangerously-skip-permissions

# Session 2: Backend
docker run --rm -d \
  --name safeclaw-backend \
  --cpus="1.5" --memory="3g" \
  -v "$(pwd):/workspace:rw" \
  -v "$HOME/.claude:/home/coder/.claude:ro" \
  -e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
  -w /workspace \
  ghcr.io/anthropics/claude-code:latest \
  claude -p "Build the API endpoints" --dangerously-skip-permissions

# Monitor
docker logs -f safeclaw-frontend &
docker logs -f safeclaw-backend &

Overnight Build with Resource Limits

docker run --rm -d \
  --name safeclaw-overnight \
  --cpus="2" \
  --memory="6g" \
  --pids-limit=512 \
  --storage-opt size=10G \
  --restart=on-failure:3 \
  -v "$(pwd):/workspace:rw" \
  -v "$HOME/.claude:/home/coder/.claude:ro" \
  -e ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
  -w /workspace \
  ghcr.io/anthropics/claude-code:latest \
  claude -p "Full implementation plan. Checkpoint after each phase." \
  --dangerously-skip-permissions \
  --max-turns 200

# Set a timeout kill (8 hours)
(sleep 28800 && docker kill safeclaw-overnight 2>/dev/null) &

Resource Limit Reference

Safety Checklist

• Never mount host / or sensitive directories
• Mount .claude as read-only (:ro)
• Set --pids-limit to prevent fork bombs
• Set memory limits to prevent OOM on host
• Use --rm to auto-cleanup containers
• Set --restart=on-failure:N with a retry cap
• Use named containers for easy monitoring
• Kill containers with timeout for overnight runs