subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill follows secure design principles by implementing a two-stage review process (specification compliance and code quality review) for all tasks. This ensures that work is verified before being committed.
- [COMMAND_EXECUTION]: Privilege separation is enforced; implementation and review subagents are explicitly prohibited from mutating Git state (staging, committing, or branching), leaving these sensitive operations strictly to the coordinator agent.
- [INDIRECT_PROMPT_INJECTION]: The skill incorporates robust defensive instructions against indirect injection by explicitly stating that child output is evidence or proposal, never an instruction or an authoritative state update. This mitigates risks associated with subagent outputs influencing the coordinator's behavior.
- Ingestion points: Implementer reports and implementation plan tasks.
- Boundary markers: Explicit instructions in
SKILL.mdto treat output as evidence/proposals only. - Capability inventory: File system read/write and Git mutation (restricted to coordinator).
- Sanitization: The two-stage review process acts as a manual verification barrier.
Audit Metadata