agent-safety
by compass-soul
Coding Agents & IDEsBrief
Outbound safety for autonomous AI agents — scans YOUR output before it leaves the machine.
Description
What This Skill Does
Automated outbound safety for AI agents and developers. Scans files and staged git content for API keys, tokens, PII, and secrets before they leave the machine. Enforces security at the git level via pre-commit hooks, not prompts.
Unlike prompt-based reminders or inbound scanners, this enforces rules at the commit level so secrets are blocked automatically, not caught after the fact.
When to Use It
- Blocking accidental API key commits before they hit GitHub
- Installing a one-time safety hook on a new repo
- Scanning a directory before publishing a package
- Checking for hardcoded emails or SSNs in generated files
- Running a system health check during periodic agent maintenance
Example Workflow
User asks: install the pre-commit hook on my repo at ~/projects/myapp
- 1
Resolve scripts/install-hook.sh path relative to the skill file
- 2
Run: bash scripts/install-hook.sh ~/projects/myapp
- 3
Confirm the hook is written to ~/projects/myapp/.git/hooks/pre-commit
- 4
Report that future commits in that repo will be scanned automatically
Pre-commit hook installed. Any staged secrets or SSNs will block the commit; PII patterns will flag for review.
Installation
clawhub install compass-soul/agent-safetyRun the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.