team-brainstorm

Pass

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill manages a multi-agent pipeline by dynamically spawning background worker agents using the Agent tool. It also includes functionality to generate and load new role-specification files at runtime within the session directory to handle capability gaps identified during the brainstorming process.\n- [COMMAND_EXECUTION]: Uses the Bash tool for local environment checks and workspace discovery. It also interacts with the maestro suite for project-level searches and loading architectural constraints, focusing on local project data.\n- [PROMPT_INJECTION]: The orchestration pipeline is susceptible to indirect prompt injection due to the way it handles user input and inter-agent communication.\n
  • Ingestion points: The system ingests untrusted user topics via command-line arguments and processes markdown artifacts generated by various worker agents (ideas, critiques, and synthesis reports).\n
  • Boundary markers: Task templates use structural headers but lack explicit delimiters or instructions to treat interpolated external data as strictly non-executable text.\n
  • Capability inventory: The skill possesses significant capabilities including the ability to write files, execute shell commands, and create new agent instances.\n
  • Sanitization: Content from the initial topic and subsequent agent-generated files is interpolated directly into instruction templates for downstream agents without specific sanitization or escaping mechanisms.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 2, 2026, 06:13 AM
Security Audit — agent-trust-hub — team-brainstorm