consulting-experts
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided goals and chat history to dynamically generate instructions for subagents.
- Ingestion points: The user's goal is extracted from the request and chat context in
references/executing-consultation.md(Step 1). - Boundary markers: There are no explicit boundary markers or 'ignore embedded instructions' warnings used when interpolating the user's goal into subagent briefs.
- Capability inventory: Subagents are instructed to load any installed skills and may have read/write access to files depending on their assigned roles (Step 3).
- Sanitization: No sanitization or validation of the user-provided goal is described before it is used to define the 'Role' and 'Job' in subagent prompts.
Audit Metadata