The Agent Skills Directory

[DATA_EXPOSURE_&_EXFILTRATION]: The skill contains explicit safety instructions in SKILL.md to never expose credentials, personal data, or production write operations in tests, logs, or examples. No sensitive file paths or network exfiltration patterns were detected in the provided scripts or prompts.
[REMOTE_CODE_EXECUTION]: There are no patterns of remote code execution. The evaluation scripts (check_domain_decisions.py and check-domain-decisions.sh) use standard libraries and are intended for deterministic validation of model outputs in a development environment, not for runtime execution of untrusted code.
[INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted data (AI-generated test code and requirements). While this presents an attack surface where malicious code could attempt to influence the auditor, the skill includes robust procedural instructions to report only evidence-backed findings and stick to specific domain rules, mitigating the risk of following embedded instructions.
[PROMPT_INJECTION]: No malicious prompt injection or bypass patterns were found. The use of strong instructional language (e.g., "Never recommend deleting assertions") is focused on maintaining testing quality and safety rather than overriding agent constraints.
[DYNAMIC_EXECUTION]: No evidence of unsafe dynamic execution (such as eval() or exec()) on user-provided input was found. The Python evaluation script uses safe parsing methods for JSON and regex.

ai-generated-test-review