The Agent Skills Directory

[SAFE]: The skill implements a secure handling pattern for untrusted external data (such as PR comments or tool diagnostics). By placing this data in a raw_input quarantine namespace and requiring a separate ingest pass for validation before promotion to top-level fields, the skill effectively prevents indirect prompt injection from influencing the agent's behavior.
[SAFE]: File operations are strictly scoped to the local plan directory (.plan/plans/{plan_id}/artifacts/findings/). No attempts to access sensitive system files or credentials were found.
[SAFE]: The skill uses standard Python libraries and local helper modules from the trusted vendor workspace. No external or untrusted dependencies are downloaded or executed at runtime.

manage-findings