multi-agent-orch
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves establishing handoff protocols where specialized agents exchange data. This architecture creates a surface for indirect prompt injection, where a compromised or malicious upstream source could embed instructions meant to manipulate a downstream agent. The skill proactively addresses this by including specific 'Poisoned Handoff Defense' and 'Confused Deputy Prevention' instructions for the agent.
- Ingestion points: Data handoffs processed between agents, specifically via structured text/JSON and the
delegation-payload.mdfile. - Boundary markers: Instructions recommend the use of strict schema validation and explicit input/output contracts.
- Capability inventory: File system write operations for generating templates (
SWARM_ARCHITECTURE_TEMPLATE.md) and delegation payloads (delegation-payload.md). - Sanitization: The skill mandates that orchestrated agents treat upstream input as untrusted and scan specifically for prompt injections.
Audit Metadata