evidence-backed-review
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute build, lint, and test commands defined within the repository being reviewed (rules/execution-and-pipeline.md). This is a core functionality intended to verify the impact of changes through automated tooling.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content from git diffs, pull request descriptions, and external documentation. While this presents an attack surface, the instructions emphasize read-only inspection, verification through actual code paths, and mandatory evidence citing (file:line), which serve as process-level guardrails against accidental instruction following.
- [SECURITY_AND_ABUSE_PATHS]: The skill explicitly includes a specialized rule for detecting security vulnerabilities such as path traversal, insecure authorization, and IDOR (rules/security-and-abuse-paths.md). The provided evaluation fixtures (e.g., invoices.js) contain examples of these vulnerabilities to test and ensure the skill's effectiveness in detection.
- [DYNAMIC_EXECUTION]: The test harness (evals/mutations.mjs) uses Node.js subprocesses to execute a validation script (evals/invariants.mjs) against the skill's own rules. This is a local testing mechanism and does not involve remote code loading.
Audit Metadata