skill-creator
Pass
Audited by Gen Agent Trust Hub on Jun 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: Several core automation scripts (run_eval.py, improve_description.py, run_loop.py) use the subprocess module to execute the claude CLI. This is required for spawning subagent tasks to test skills and for retrieving description improvements.
- [COMMAND_EXECUTION]: The evaluation viewer utility (generate_review.py) invokes the lsof command through subprocess to identify and manage local network ports, ensuring the visualization server can start on its dedicated port.
- [EXTERNAL_DOWNLOADS]: The browser-based review interface (viewer.html) references the SheetJS library from a public CDN (cdn.sheetjs.com). This is used to enable the local rendering of Excel spreadsheets generated during testing.
- [INDIRECT_PROMPT_INJECTION]: The skill architecture creates an indirect injection surface by ingesting and processing user-supplied test prompts and feedback from JSON files. This is inherent to the skill's purpose as a testing tool, and the documentation includes explicit safety guidelines for authors.
- [SAFE]: No evidence of credential harvesting, unauthorized network exfiltration, or persistence was detected. The skill operates within the expected scope of a developer-oriented productivity tool.
Audit Metadata