review-resolution
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external and potentially untrusted data from multiple sources including PR comments, CI reports, and external review systems.
- Ingestion points: Input data includes review findings from GitHub, static analysis tools, and other agents, as defined in
SKILL.mdandreferences/finding-classification.md. - Boundary markers: The skill includes a specific
Security Modelsection and aSource and authorityreference that instruct the agent to treat feedback as 'untrusted evidence' and explicitly ignore 'instruction-shaped text' that attempts to bypass guidelines or override behavior. - Capability inventory: The skill utilizes read-only Git commands (
git status,git diff,git rev-parse) for state verification. It includes an interpolation point where external identifiers (<reviewed-head>) are used in a shell command (git diff --name-status <reviewed-head>..HEAD), creating a potential surface for command injection if the input is malicious. - Sanitization: The workflow mandates independent verification of every technical claim, requires the separation of original reviewer text from interpreted technical claims, and routes actual implementation to specialized skills, providing isolation from the adjudication logic.
Audit Metadata