untangle-codebase
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process and refactor entire codebases. This creates a surface for indirect prompt injection if the codebase being analyzed contains malicious instructions disguised as code or comments. The skill instructions attempt to mitigate this by providing a highly structured logic for analysis (identifying hidden links, needless links, and surplus) rather than simply 'following' instructions found in the code.
- [COMMAND_EXECUTION]: The skill advises the agent to generate and execute local scripts for tasks such as creating stripped-down copies of the codebase or generating dependency graphs. While this involves command execution, these are agent-generated tools for internal analysis of the local environment and do not involve remote code or untrusted sources.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses the local filesystem and Git history to perform its function. However, there are no patterns suggesting data exfiltration or unauthorized access to sensitive files (e.g., credentials). It explicitly suggests writing to temporary directories to prevent accidental corruption of the working tree.
Audit Metadata