The Agent Skills Directory

[SAFE]: The skill provides a robust safety framework for the AI agent, specifically instructing it to treat all analyzed code as untrusted evidence and to ignore instructions embedded within reviewed files, effectively mitigating prompt injection risks.\n- [SAFE]: The skill includes explicit policies for redacting sensitive information like API keys and credentials, preventing accidental data exposure during the review process.\n- [SAFE]: Evaluation fixtures located in the evals/files/ directory contain intentionally vulnerable code patterns and synthetic placeholders for credentials. These are used strictly for quality assurance to verify the skill's detection logic and are clearly marked as non-production, inert text.\n- [SAFE]: No evidence of malicious command execution, network exfiltration, persistence mechanisms, or obfuscation was found within the skill's instructions or reference materials.

code-review