brainstorming
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from user terminal descriptions and browser-based interaction events to generate project mockups and design specifications. Ingestion points: User requests in the terminal; interaction events (clicks/selections) from the visual companion interface stored in state_dir/events. Boundary markers: The skill employs a in SKILL.md to prevent implementation without explicit human approval. Capability inventory: Writes HTML files to the local filesystem; starts a Node.js WebSocket server; executes commands to open a browser; dispatches subagents via builder-check-prompt.md. Sanitization: Includes basic HTML escaping for metadata in server.cjs, though the AI generates mockup content dynamically from user input.
- [COMMAND_EXECUTION]: The skill invokes local system commands to manage a visual companion server and launch a web browser. These operations are protected by a random session token generated at startup.
- [EXTERNAL_DOWNLOADS]: The visual companion server attempts to download a branding logo from https://primeradiant.com/brand/superpowers-visual-brainstorming-logo.png when the interface is loaded in a browser.
Audit Metadata