audit-agent-skills

name	audit-agent-skills
description	Audit local or third-party agent skills for trigger quality, workflow precision, token efficiency, compatibility, provenance, and execution safety. Use before installing, publishing, trusting, or substantially revising SKILL.md packages and their scripts, references, assets, or tool dependencies.

Audit Agent Skills

1. Inventory every file, declared dependency, executable path, external call, permission, and source. Establish publisher, license, version, maintenance state, and update mechanism for third-party material.
2. Validate the platform schema and folder naming with the available official validator. Treat product-specific metadata as non-portable unless the target supports it.
3. Test the description against direct, indirect, incomplete, negative, and neighboring requests. Require a focused user goal, distinct triggers, and clear boundaries with other skills.
4. Inspect the body for explicit inputs, ordered actions, outputs, stop or question conditions, prohibited inference, and checkable completion criteria. Match instruction freedom to operational risk.
5. Remove duplicated meaning, stale material, default-behavior instructions, and unrelated exposition. Move branch-specific detail to directly linked references and deterministic repeated work to tested scripts.
6. Review scripts and dependencies without executing untrusted code. Check secret handling, destructive actions, network access, sandbox assumptions, approvals, and auditability.
7. Report findings by severity with evidence and a minimal remediation. Distinguish structural validation from behavioral confidence.

Finish when trigger behavior, execution behavior, context cost, and safety have each been evaluated with no unaudited executable path.