melech-code-review-clone
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the GitHub CLI (
gh) andgitto index pull request metadata and repository history for training. These tools are used correctly with list-based arguments in thesubprocess.runcalls, which prevents command injection risks. - [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests PR comments from GitHub which could contain adversarial instructions. This is mitigated by the 'Trainer' logic, which distills behavioral patterns (e.g., tone and focus areas) into a structured model rather than executing raw input, and requires the user to manually review and 'Publish' the model before the secondary 'Clone' agent is active.
- [DYNAMIC_EXECUTION]: The skill dynamically generates a secondary agent skill (
cr-clone-<login>) based on a template. This generation is restricted to a safe, predefined local format and serves the core functional purpose of the tool.
Audit Metadata