skills/b-open-io/prompts/orchestrator/Gen Agent Trust Hub

orchestrator

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill is configured to transmit repository data, code snippets, and prompts to external models including Grok (xAI) and Fable (Anthropic).
  • Evidence: "A Grok dispatch can send its prompt, specification, code excerpts, and other repository content to xAI." (SKILL.md)
  • Evidence: "A Fable consult can send its consult package and repository files inspected by read tools to Anthropic." (SKILL.md)
  • Safeguards: The instructions require explicit user approval before the first transmission and strictly prohibit sending credentials, secrets, or unrelated proprietary content.
  • [PROMPT_INJECTION]: The skill's architecture involves ingesting and acting upon instructions or code generated by external agents, presenting a risk of indirect prompt injection.
  • Ingestion points: Specialist reports, implementation worker results (Grok), and advisor consults (Fable).
  • Boundary markers: Uses read-only boundaries for advisors and partitioned file ownership for implementation workers.
  • Capability inventory: The main agent possesses the authority to perform final synthesis, execute tests, commit changes, and push code to remote repositories.
  • Sanitization: Requires "adversarial review" of every worker diff and main-seat verification before finalization.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 10:49 PM
Security Audit — agent-trust-hub — orchestrator