launching-agent-teams

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill includes a Python script (scripts/update-docs.py) that downloads documentation from code.claude.com, which is an official and trusted domain associated with Anthropic. This is a legitimate documentation update mechanism.
  • [COMMAND_EXECUTION]: The SKILL.md file suggests running shell commands for updating docs and managing tmux sessions. These are standard developer operations relevant to the skill's purpose (parallel agent coordination).
  • [REMOTE_CODE_EXECUTION]: While the skill references fetching documentation, it does not execute remote code or download executable scripts from untrusted sources. The fetched content is stored as reference markdown files.
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by ingesting documentation from the web and processing user-provided spawn prompts for agent teams. However, it incorporates clear boundaries and coordination principles (e.g., plan approval requirements, explicit task decomposition) that mitigate risks of accidental instruction obedience.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 05:49 PM
Security Audit — agent-trust-hub — launching-agent-teams