brainstorming

Warn

Audited by Gen Agent Trust Hub on May 15, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The REFERENCE.md file instructs the user or agent to install an external package globally using 'npm install -g agent-browser'. This downloads code from an external registry that is not associated with a well-known or trusted service.
  • [COMMAND_EXECUTION]: The skill documentation provides shell commands to be executed in the host environment, including 'agent-browser install', writing files to the filesystem ('cat > /tmp/mockup.html'), and interacting with the browser tool ('agent-browser open', 'agent-browser screenshot').
  • [PROMPT_INJECTION]: The skill processes untrusted project data (files, documentation, and commit history) to inform its brainstorming process, which constitutes an indirect prompt injection risk.
  • Ingestion points: Reads relevant files, project documentation, and recent commits from the local workspace.
  • Boundary markers: The instructions lack explicit boundary markers or directions for the agent to ignore embedded instructions within the processed files.
  • Capability inventory: The skill has the ability to write files to the system and execute shell commands via the 'agent-browser' tool.
  • Sanitization: There is no evidence of sanitization or filtering of the content ingested from the project environment before it is used to influence the agent's logic.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 15, 2026, 09:06 AM
Security Audit — agent-trust-hub — brainstorming