If you're running agent teams in Claude Code, this is essential. Without it, auto-compaction hits and your lead agent loses all team state (TeamCreate, SendMessage, the whole coordination layer). Guard runs a background daemon that monitors session size and auto-prunes at configurable thresholds before compaction triggers. Default setup checkpoints every 30 seconds, does a gentle prune at 30MB, and a full prune with auto-reload at 50MB while preserving team state. It's the difference between your agent team running continuously for hours versus getting wiped mid-task. For single-agent sessions it's optional, but for multi-agent work it prevents a specific, painful failure mode.

GitHub stars
332
last updated
Jun 11, 2026

Install

$ npx -y skills add ruya-ai/cozempic --skill guard --agent claude-code

Installs into .claude/skills of the current project.

Files

Start 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:

  1. Checkpoints team state every 30 seconds
  2. At 60% of threshold (30MB): applies gentle prune, no reload
  3. 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 status and stop

Sponsors

Details

Repository
github.com/ruya-ai/cozempic
License
MIT
First seen
Jun 11, 2026

More from ruya-ai/cozempic

All 5

Recommended

1juliusbrussee avatarcavemanjuliusbrussee/cavemanUltra-compressed communication mode. Cuts output tokens 65% (measured) by speaking like caveman while keeping full technical accuracy. Supports intensity levels618.3Kskills.sh installs
2mattpocock avatargrill-memattpocock/skillsA relentless interview to sharpen a plan or design.1.5Mskills.sh installs
3shadcn avatarimproveshadcn/improveSurvey any codebase as a senior advisor and produce prioritized, self-contained implementation plans for OTHER models/agents to execute. Strictly read-only on s40Kskills.sh installs
4obra avatarsystematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes326.7Kskills.sh installs
5forrestchang avatarkarpathy-guidelinesforrestchang/andrej-karpathy-skillsBehavioral guidelines to reduce common LLM coding mistakes through explicit assumptions, simplicity, and verifiable success criteria.17.9Kskills.sh installs
6vercel-labs avatarfind-skillsvercel-labs/skillsHelps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express int4.3Mskills.sh installs