review-loop
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a review-repair-review loop without any malicious functionality. It does not perform unauthorized network requests, access sensitive files, or attempt to persist across sessions.\n- [INDIRECT_PROMPT_INJECTION]: The skill handles external data in the form of review targets, which is an inherent surface for indirect prompt injection. The skill includes specific defensive measures for this.\n
- Ingestion points: Untrusted data enters via the 'Target' and 'Requirements' fields as specified in
reviewer-contract.md.\n - Boundary markers: The
reviewer-contract.mdfile contains explicit instructions to ignore any prompt text within the target that attempts to manipulate the reviewer's behavior or outcomes.\n - Capability inventory: The skill coordinates other tools (e.g.,
generic-review) and manages a repair cycle through the agent's standard tool-calling interface.\n - Sanitization: Input is not programmatically sanitized, but the skill requires strict adherence to the
finding-schema.mdfor all outputs, which limits the impact of malicious data.
Audit Metadata