code-degunker
Pass
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill defines a 'senior code reviewer' persona to provide context for the agent's behavior. These instructions are standard for defining task parameters and do not include malicious directives to bypass safety filters or disregard system instructions.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill contains several examples of hardcoded credentials and insecure configurations (e.g., 'super-secret-key-change-me', 'postgresql://admin:password123@localhost/mydb'). These are explicitly presented as 'before' examples of anti-patterns to be corrected and do not function as actual secrets or exfiltration code.\n- [INDIRECT_PROMPT_INJECTION]: The skill operates on user-provided code, which is an untrusted ingestion point. However, its instructions are focused on defensive security and it does not grant the agent capabilities to perform unauthorized actions such as network requests or file system operations.\n
- Ingestion points: User-provided source code via 'Full Review Mode' or 'Branch Review Mode' (SKILL.md).\n
- Boundary markers: None explicitly defined for the input data, but the skill follows a structured analysis process.\n
- Capability inventory: The skill is limited to textual analysis and providing rewritten code suggestions.\n
- Sanitization: The skill's primary purpose is to instruct the agent to find and fix missing validation and sanitization in the target codebase.
Audit Metadata