skills/openai/plugins/brainstorming/Gen Agent Trust Hub

brainstorming

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • Local Web Server Deployment: The skill utilizes start-server.sh and server.cjs to initiate a Node.js-based web server. This server hosts the "Visual Companion," allowing users to view mockups in a browser. While it defaults to 127.0.0.1, it can be configured to bind to 0.0.0.0, potentially exposing content to other devices on the network.\n- Dynamic Browser Scripting: To facilitate interactive mockups, the visual companion generates HTML and injects a helper script (helper.js) into the browser session. This script manages communication between the browser and the agent, representing a form of dynamic code execution in the client environment.\n- Indirect Prompt Injection Surface: The agent is designed to ingest project context from local files, documentation, and browser-based user events. This ingestion of external, untrusted content could potentially be used to influence the agent's behavior.\n
  • Ingestion points: Project source files, git history, documentation, and user interaction logs (state_dir/events).\n
  • Boundary markers: The skill employs logical gates for implementation steps but lacks explicit delimiters for data read from the local file system.\n
  • Capability inventory: The skill can execute shell commands for server management, perform git operations, write new files to the system, and invoke downstream implementation planning skills.\n
  • Sanitization: No specific mechanisms for sanitizing or filtering data from project files or browser events are detailed in the instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 11:16 AM
Security Audit — agent-trust-hub — brainstorming