visuals
Pass
Audited by Gen Agent Trust Hub on Oct 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill uses browser automation to interact with external URLs, introducing a surface for indirect prompt injection from malicious web content. \n
- Ingestion points: Opening external URLs via playwright-cli (SKILL.md, screenshots.md). \n
- Boundary markers: None identified for external web content. \n
- Capability inventory: Access to git, gh-attach, sed, and playwright-cli. \n
- Sanitization: No sanitization of external web content is performed. \n- [DYNAMIC_EXECUTION]: The skill generates a JavaScript file by injecting agent-supplied text into a template using sed, followed by execution via playwright-cli eval (screenshots.md). This involves dynamic script generation and runtime execution. \n- [EXTERNAL_DOWNLOADS]: The skill downloads Mermaid and Vega-Lite libraries from the jsDelivr CDN (render.js). This is a neutral finding involving a well-known service for visualization purposes. \n- [COMMAND_EXECUTION]: The skill runs a local web server using python3 -m http.server to serve visualization components (screenshots.md). This bypasses browser restrictions on local files by exposing the working directory to the local network.
Audit Metadata