do-delegation
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for a parent agent to dispatch tasks to and ingest results from multiple sub-agents. This orchestration creates a potential vector for indirect prompt injection if a sub-agent is manipulated into returning malicious instructions or deceptive data that the parent agent might act upon.
- Ingestion points: The parent agent processes completion messages and results from foreground and background sub-agents launched via the
Agenttool. - Boundary markers: The instructions recommend using strict result contracts and explicit prompts that define context and constraints for each sub-worker to limit divergence.
- Capability inventory: Sub-agents are empowered to read files, perform web searches, and edit source code (potentially using worktree isolation).
- Sanitization: The skill includes a 'Verification Contract' section that explicitly instructs the parent agent not to trust sub-agent claims. It mandates inspection of actual file diffs, evidence verification, and running tests to validate sub-agent contributions before synthesis.
Audit Metadata