The Agent Skills Directory

[SAFE]: The skill utilizes permissionMode: plan when calling the audit tool, which restricts the underlying agent to a read-only state. This ensures that the audit process cannot modify files or project configuration.
[SAFE]: The configuration in agents/openai.yaml disables allow_implicit_invocation. This is a security best practice that prevents the skill from being triggered automatically by the agent, ensuring audits only occur upon explicit user request and avoiding potential self-review loops.
[SAFE]: The prompt instructions include a provenance note that explicitly directs the auditor to ignore external claims and apply independent judgment. This provides a layer of defense against indirect prompt injection where comments in the audited code might attempt to influence the audit outcome.

audit