pr-review

Pass

Audited by Gen Agent Trust Hub on Oct 29, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves processing external pull request diffs and source code, which are untrusted data sources that could contain malicious instructions.
  • Ingestion points: The skill uses the read_file tool to ingest diffs and surrounding source files as defined in SKILL.md.
  • Boundary markers: While the instructions emphasize scoping to the diff, no specific technical delimiters are defined to isolate untrusted code content from agent instructions.
  • Capability inventory: The skill is scoped to reading files and executing a local Python validation script (scripts/validate-output.py) for format checking.
  • Sanitization: The skill implements robust mitigations, including a requirement to manually verify all findings against the code and a mandatory human-in-the-loop gate before posting results to external systems or approving sensitive code paths (e.g., authentication, payments).
  • [COMMAND_EXECUTION]: The skill includes a test script tests/run-validation.py that executes a local helper script using the subprocess module.
  • Evidence: The script tests/run-validation.py calls subprocess.run([sys.executable, str(validator)], ...) to run the local validator. This is a safe execution of a known local script using the current Python interpreter with shell=False, and it is only intended to be run upon explicit user request.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 29, 2026, 03:18 PM