executing-plans-with-subagents
Pass
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill establishes an orchestration workflow centered on bounded tasks and isolation. It mandates that subagents operate with limited scope ('Touch only what this task requires') and requires the orchestrator to integrate and verify results independently, reducing the risk of unauthorized or harmful changes.
- [PROMPT_INJECTION]: The skill involves a surface for indirect prompt injection as it processes code and status reports from subagents. However, it implements robust mitigations: (1) Ingestion points: Subagent reports and file diffs; (2) Boundary markers: Prompts use explicit sections for 'Task', 'Context', and 'Boundaries'; (3) Capability inventory: Subagents are permitted to inspect repo state and edit files; (4) Sanitization: The orchestrator uses a 'Spec Compliance Review' to compare code to requirements line-by-line and runs 'integrated verification' commands to ensure the final state matches expectations. These checks effectively prevent the orchestrator from blindly following instructions embedded in subagent output.
Audit Metadata