brainstorming

Warn

Audited by Gen Agent Trust Hub on Aug 5, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill launches a custom Node.js server (scripts/server.cjs) to support its browser-based visual companion. This server binds to local network ports and can execute system commands via child_process to open browser sessions. Control scripts (scripts/start-server.sh) also interact with the system process table to manage the server lifecycle.\n- [DATA_EXFILTRATION]: A local file server is established to serve project mockups via the /files/* endpoint. Although it implements security measures—including path traversal checks and a mandatory 32-byte authentication token—the server creates a path for exposing local project data if sensitive files are placed within its content directory.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Ingestion points: Design specs via spec-document-reviewer-prompt.md and browser events via $STATE_DIR/events. Boundary markers: No explicit delimiters or 'ignore' warnings are present in the subagent prompt. Capability inventory: Shell command execution, local server management, and file system writes. Sanitization: No evidence of sanitization for captured event text or spec content before prompt interpolation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 5, 2026, 04:18 PM
Security Audit — agent-trust-hub — brainstorming