update-pr-review
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes human comments from external GitHub PRs to update internal guidance, creating a surface for indirect prompt injection.
- Ingestion points:
scripts/aggregate_review_feedback.pyuses theghCLI to fetchhuman_review_comments,conversation_comments, and PR review bodies. - Boundary markers: The instructions do not specify structured delimiters for the ingested JSON feedback, relying on the agent's internal parsing.
- Capability inventory: The skill can suggest changes to repo-local review skills (e.g.,
.github/skills/review-pr-repo/SKILL.md), which are subsequently applied byscripts/apply_guidance_output.pyand validated byscripts/validate_write_surface.pyusinggitcommands. - Sanitization:
SKILL.mdprovides natural language guardrails, instructing the agent to abstract patterns rather than copying raw text, keep wording concise, and avoid pasting raw JSON into skill files. - [COMMAND_EXECUTION]: The skill relies on local shell command execution for its data gathering and validation workflows.
scripts/aggregate_review_feedback.py: Executesgh api graphql,gh repo view, andgh pr listto retrieve repository data and PR feedback.scripts/validate_write_surface.py: Executesgit diffandgit ls-filesto ensure no unauthorized files were modified during the process.
Audit Metadata