brainstorming

Pass

Audited by Gen Agent Trust Hub on Sep 11, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a 'Visual Companion' feature that processes user interaction data from a browser-based interface.
  • Ingestion points: The agent is instructed in visual-companion.md to read user interaction events (clicks, selections) from $STATE_DIR/events. These events are generated by the browser-side helper (scripts/helper.js) and recorded by the companion server (scripts/server.cjs).
  • Boundary markers: The skill's instructions state that terminal-based conversation is the primary feedback source, and browser events are used to supplement the context.
  • Capability inventory: The skill can write design documents to the filesystem, invoke implementation planning skills, and manage a local background server.
  • Sanitization: Browser events are structured as JSON and parsed by the server; however, the agent directly reads this logged interaction data to influence the design process.
  • [COMMAND_EXECUTION]: The skill manages a local background Node.js server using shell scripts for brainstorming sessions.
  • Server Management: Scripts scripts/start-server.sh and scripts/stop-server.sh use shell utilities like ps, kill, and nohup to manage the lifecycle of the companion server.
  • Security Best Practices: The Node.js server (scripts/server.cjs) uses path.basename when serving files from the content directory, preventing directory traversal attacks. It also includes an automated shutdown mechanism (lifecycleCheck) that exits the server if the parent process dies or after 30 minutes of idle time.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 11, 2026, 03:43 PM
Security Audit — agent-trust-hub — brainstorming