review-hog-resolution-criteria

Pass

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data from pull request review threads, creating a surface for malicious actors to influence agent behavior through crafted comments.
  • Ingestion points: The agent ingests and acts upon review thread conversations from the pull request as defined in the SKILL.md resolution criteria.
  • Boundary markers: Absent. The instructions do not define specific delimiters or "ignore instructions" wrappers for the content of review comments.
  • Capability inventory: The agent possesses capabilities to modify files, commit changes to the repository, and execute testing/linting commands for verification.
  • Sanitization: Absent. The skill defines high-precedence keyword overrides like "SAFE TO FIX" and "E2E REQUIRED" which the agent is instructed to "never ignore," allowing a commenter to potentially override the agent's internal safety evaluations.
  • Mitigations: The skill includes internal logic for "trust-weighting" based on author association and restricts fixes to "contained and provable" changes, which provides some protection but does not eliminate the injection risk.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 12, 2026, 07:08 PM
Security Audit — agent-trust-hub — review-hog-resolution-criteria