adversarial-review

Pass

Audited by Gen Agent Trust Hub on Sep 13, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from pull request diffs, descriptions, and issue comments using tools like gh pr view and gh pr diff. A malicious contributor could embed instructions within these fields to influence the agent's behavior during the review process.
  • Ingestion points: Pull request metadata, diffs, and issue descriptions/comments as described in SKILL.md.
  • Boundary markers: None identified; the agent is instructed to read the full content of these external sources without specific delimiters.
  • Capability inventory: Executes shell commands via the gh CLI and runs local project test suites.
  • Sanitization: No explicit sanitization or filtering of the ingested external text is performed.
  • [COMMAND_EXECUTION]: The skill utilizes the GitHub CLI (gh) to perform actions such as viewing pull requests, viewing diffs, and posting verdict comments.
  • [DYNAMIC_EXECUTION]: The instructions require the agent to run the project's existing test suite to verify identified defects. While the skill specifies using a disposable worktree, this involves executing code that is currently under review and may be untrusted.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 13, 2026, 06:24 AM
Security Audit — agent-trust-hub — adversarial-review