orchestrating-swarms
Pass
Audited by Gen Agent Trust Hub on Apr 2, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill documents an architecture for multi-agent swarms where a 'Leader' agent processes messages and findings from 'Teammate' agents. This introduces an indirect prompt injection surface if those teammates are tasked with analyzing untrusted data from external sources.
- Ingestion points: Teammate inbox files (~/.opencode/teams/{name}/inboxes/*.json) containing messages, pull request content for review agents, and external web data retrieved via google_search or webfetch.
- Boundary markers: The orchestration message format does not define explicit delimiters or 'ignore' instructions for teammate-generated content.
- Capability inventory: Teammate agents have access to the Bash tool for command execution, file system operations, and the ability to spawn further agents.
- Sanitization: No documentation exists for sanitizing or validating the structured JSON messages passed between agents.
- [EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to install the it2 CLI tool using standard package managers (pip, uv) to enable iTerm2 backend support. This is a reference to a well-known utility for a common development environment.
Audit Metadata