tool-install

Pass

Audited by Gen Agent Trust Hub on Jul 31, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing untrusted data from external URLs via fetch_content and web_search to determine installation steps.
  • Ingestion points: The workflow fetches user-provided links to research tool descriptions and installation methods (Step 1 in SKILL.md).
  • Boundary markers: The skill includes an explicit approval gate requiring the user to reply with "approve" or "go" before any mutation occurs. It also utilizes plannotator --gate if available.
  • Capability inventory: The agent can execute shell commands, modify shell configuration files (e.g., .zshrc), and install background services (LaunchAgents/LaunchDaemons).
  • Sanitization: No explicit sanitization or filtering of external content is described; the agent relies on its own "research" to interpret the data.
  • [COMMAND_EXECUTION]: The skill is designed to execute shell commands to inspect the system environment, perform installations, and verify success.
  • [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch metadata from user-specified URLs and download software packages via package managers like Homebrew, npm, pip, and cargo.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 31, 2026, 04:01 PM
Security Audit — agent-trust-hub — tool-install