guard
npx machina-cli add skill Ruya-AI/cozempic/guard --openclawStart the cozempic guard daemon for continuous session protection.
Default (recommended)
cozempic guard --daemon --threshold 50 -rx standard --interval 30
This runs in the background and:
- Checkpoints team state every 30 seconds
- At 60% of threshold (30MB): applies gentle prune, no reload
- At threshold (50MB): applies full prescription + auto-reload with team state preserved
For agent teams
Guard mode is essential for sessions running agent teams. Without it, auto-compaction triggers and the lead agent loses team state (TeamCreate, SendMessage, tasks are discarded).
Options
--threshold N— hard threshold in MB (default: 50)--soft-threshold N— soft threshold in MB (default: 60% of hard)--threshold-tokens N— hard threshold in tokens (fires whichever hits first)--no-reload— prune without restarting Claude--no-reactive— disable kqueue/polling file watcher-rx NAME— prescription at hard threshold (default: standard)
Check if running
The daemon writes to /tmp/cozempic_guard_*.log. Check with:
ls /tmp/cozempic_guard_*.pid 2>/dev/null
Source
git clone https://github.com/Ruya-AI/cozempic/blob/main/plugin/skills/guard/SKILL.mdView on GitHub Overview
The guard skill starts a background daemon that monitors session size and prunes data before compaction triggers. It helps protect team state, especially for agent teams, by applying a gentle prune at 60% of the threshold and a full prescription with auto-reload at the hard threshold.
How This Skill Works
The guard runs as a daemon (cozempic guard --daemon) with a configured hard threshold and interval. It checkpoints team state periodically, triggers a gentle prune at 60% of the threshold without a reload, and performs a full prune with auto-reload at the hard threshold while preserving team state, logging activity to /tmp/cozempic_guard_*.log and pid files.
When to Use It
- When running agent teams to prevent loss of TeamCreate, SendMessage, and tasks during auto-compaction
- Before compaction triggers to avoid unexpected state loss
- When session sizes approach the hard threshold (default 50 MB) or the soft threshold
- In environments requiring continuous background protection without manual intervention
- When you need regular checkpoints of team state and automated pruning
Quick Start
- Step 1: Start the daemon with: cozempic guard --daemon --threshold 50 -rx standard --interval 30
- Step 2: Confirm it's running by checking the PID files: ls /tmp/cozempic_guard_*.pid
- Step 3: Monitor status and logs: tail -f /tmp/cozempic_guard_*.log
Best Practices
- Run the guard as a background daemon to provide continuous session protection
- Tune --threshold and --soft-threshold to match your data size and risk tolerance
- Use --no-reload only for prune-without-restart testing or special workflows
- Enable guard for agent teams to preserve state during pruning and reloads
- regularly verify /tmp/cozempic_guard_*.log and /tmp/cozempic_guard_*.pid to confirm operation
Example Use Cases
- Start the daemon with default hard threshold of 50MB for an agent team: cozempic guard --daemon --threshold 50 -rx standard --interval 30
- Guard triggers a gentle prune at 60% of the hard threshold (30MB) without reload
- At the hard threshold (50MB), guard performs full pruning and auto-reload while preserving team state
- Verify guard is running by listing pid files: ls /tmp/cozempic_guard_*.pid
- Disable automatic reload for testing using --no-reload while pruning
