supervisor-boundary
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill architecture creates a surface where a Supervisor agent ingests and evaluates content from a Worker agent.\n
- Ingestion points: Worker results, candidate evaluations, and admission decisions mentioned in SKILL.md.\n
- Boundary markers: Specified use of SHA/fencing/validation checks and read-only mounts in SKILL.md.\n
- Capability inventory: Supervisor has read-only access to project state; Worker has workspace edit and shell access as defined in the Capability contract table.\n
- Sanitization: Referenced broker component to validate authorization and leases before side effects.\n- [DYNAMIC_EXECUTION]: The skill mentions a script used for materializing security guards and adapters.\n
- Evidence: The document refers to materialize.mts for installing the Supervisor-boundary adapter in SKILL.md.
Audit Metadata