subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Mar 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill defines a methodical development process using separate subagent roles. The use of isolated subagents for implementation and multi-stage review (spec compliance and code quality) is a best-practice architecture that minimizes the risk of context pollution and automated errors.
- [INDIRECT_PROMPT_INJECTION]: The skill contains a surface for indirect prompt injection as it ingests task descriptions and implementation plans from external files and interpolates them into subagent prompts (specifically in 'implementer-prompt.md' and 'spec-reviewer-prompt.md'). Evidence of this surface includes ingestion points where full plan text is pasted into prompts and the capability inventory of subagents which includes writing and committing code. The risk is mitigated by explicit instructions in the 'spec-reviewer-prompt.md' requiring independent code verification and the use of multi-stage review loops to catch deviations from the intended specification.
Audit Metadata