skill-creator
Pass
Audited by Gen Agent Trust Hub on Mar 12, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the
claudeCLI and its own Python scripts to automate testing and benchmarking workflows. It also manages local network ports using standard system commands.\n- [EXTERNAL_DOWNLOADS]: The visualization viewer loads theSheetJSlibrary fromcdn.sheetjs.comto provide spreadsheet rendering functionality for evaluation data.\n- [REMOTE_CODE_EXECUTION]: The skill uses the officialanthropicclient to communicate with the Anthropic API for description optimization, which is a core feature of the iterative improvement loop.\n- [DATA_EXFILTRATION]: A local HTTP server is hosted on127.0.0.1to display evaluation reports. This allows for local review of subagent outputs and does not involve external data transmission.\n- [PROMPT_INJECTION]: The skill processes subagent transcripts and user feedback. It employs XML-style tags and escaping to protect against the misinterpretation of data as system instructions during its optimization processes.
Audit Metadata