test-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted material provided by the user, which could contain instructions intended to influence the agent's behavior.
- Ingestion points: Material includes test code, fixtures, assertions, run history, failure logs, change diffs, and CI configuration (as specified in
SKILL.mdandprompts/test-code-review.md). - Boundary markers: The skill enforces a rigorous "six-part input audit" (known, missing, conflicting, stale, out_of_scope, and assumptions) which requires the agent to verify the state of input data before generating findings.
- Capability inventory: The skill explicitly forbids the agent from performing dangerous actions such as executing tests, editing code, or modifying target systems ("Do not use it to execute tests... it does not execute tests or turn a design inventory into coverage... Do not edit requirements, code, test assets, or target systems"). There are no scripts or executable tools bundled with the skill.
- Sanitization: The prompt instructions require the agent to separate facts from inferences and recommendations, and to explicitly mark missing or unverified evidence, reducing the risk of the agent adopting instructions found within the reviewed material.
- [NO_CODE]: The skill consists exclusively of Markdown prompts, YAML configuration, and CSV/JSON metadata. It does not include any executable scripts (Python, Node.js, Shell, etc.), which significantly reduces its attack surface.
Audit Metadata