orchestrate
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill framework involves a coordinator agent ingesting and integrating data from independent sub-agents, which introduces a standard indirect prompt injection surface.
- Ingestion points: The 'Harvest, Integrate, Verify' section of SKILL.md describes the coordinator reading summaries, deviations, and diffs from multiple worker agents.
- Boundary markers: The skill contains strong defensive instructions, such as 'a worker's summary is a claim, not evidence' and 'Inspect actual diffs and open consequential sources'.
- Capability inventory: The coordinator possesses capabilities to integrate code, commit changes, and execute verification commands.
- Sanitization: The instructions mandate human-like oversight and cross-validation of results rather than automated reliance on worker-generated text.
- [EXTERNAL_DOWNLOADS]: The skill references educational and technical documentation from well-known AI research organizations.
- Evidence: Links to Anthropic's multi-agent research case study and OpenAI's Astra guidance are provided in the 'Evidence and Limits' section of SKILL.md for technical context.
Audit Metadata