skill-creator
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- Command Execution via CLI Subprocess: The skill utilizes subprocess calls to the claude CLI to perform description optimization and evaluation. This behavior is necessary for the skill's purpose and uses argument lists to mitigate common command injection risks.
- Local Visualization Server: To provide a rich interface for reviewing test results, the skill can start a local HTTP server on 127.0.0.1. This is a standard approach for data visualization in development tools and is restricted to local access.
- Indirect Prompt Injection Surface: As a tool for creating other skills, it naturally processes user-provided inputs and generated outputs. The instructions include explicit guidelines for the agent to ensure generated skills are safe and do not contain malicious code.
- External Resource Usage: The evaluation viewer references the SheetJS library from a well-known CDN for spreadsheet processing. This is a common and expected practice for handling diverse file formats in a web-based UI.
Audit Metadata