chaos-code-risk-scan
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the system git CLI and internal Python scripts to perform repository analysis. These commands are executed via subprocess.run with argument lists rather than shell strings, which mitigations command injection vulnerabilities. The repository path is validated to ensure it is a valid git work tree.
- [DATA_EXFILTRATION]: Source code snippets and repository metadata are extracted and transmitted to the Harness platform using the harness-mcp-v2 server. This data movement is the primary intended function of the skill and is directed toward the vendor's own infrastructure.
- [DYNAMIC_EXECUTION]: The skill uses a multi-phase workflow that involves generating and executing local Python scripts (inventory.py, validate_evidence.py, coverage.py) at runtime to process source code and validate findings before submission.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code from the target repository, which represents a potential attack surface. However, the risk is effectively mitigated by mandatory verification scripts that ensure all code locations and snippets cited in findings appear verbatim in the repository's git history, preventing the agent from being misled into fabricating or misrepresenting findings.
Audit Metadata