Guardrails

Link: https://github.com/adewale/guardrails-skill

Code quality verification gates wired into the agent lifecycle. Not guidance the agent might follow — checks that block progress until they pass.

What problem it solves

Coding agents can declare success without running tests, weaken their own test configs to make checks pass, and thrash on the same failed fix repeatedly. This skill enforces lifecycle hooks that prevent all three.

Key behaviors

• Lifecycle hooks — fast checks on every stop, full suite on every commit
• Circuit breaker — after 2 failed fix attempts, forces diagnostic tool-building instead of retrying
• Config protection — agent cannot weaken its own guardrails
• Integration check — verifies new code is reachable, not just correct
• Diagnostic escalation — catalog of tools and notations for when the agent is stuck

Who uses this

Developers using Claude Code (or other AI coding agents) who want automated quality enforcement rather than prose instructions the agent might ignore.

Example

Install with npx skills add adewale/guardrails-skill. The agent then runs format/lint/type/test checks on every stop, full suite on every commit, and escalates to diagnostic tools after 2 failed fix attempts.

• Based on a real use case
• No duplicates in existing skills
• Follows skill structure template
• Repository is public and skill is tested