pr-self-review
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources that could contain instructions designed to manipulate the agent's behavior.
- Ingestion points: The skill reads untrusted data from Pull Request titles, bodies, diffs, comments, and issues using
gh pr view,gh pr diff, andgh apicommands as described inSKILL.md. - Boundary markers: No explicit delimiters are used for data, but the skill provides strict stylistic guidelines and context-gathering steps that help the agent focus on specific data attributes.
- Capability inventory: The skill possesses the capability to post content back to GitHub as PR reviews via
gh api --method POSTas defined inreferences/github-posting.md. - Sanitization: The skill includes robust mitigations, most notably a requirement for explicit human approval before any data is posted. It also mandates the use of JSON files for API inputs to prevent command injection from untrusted PR content into shell arguments.
Audit Metadata