subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Jun 21, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill processes implementation plans from external files, which creates a surface for indirect prompt injection. 1. Ingestion points: Plan files are read by the controller and task text is passed to subagents (see SKILL.md and implementer-prompt.md). 2. Boundary markers: Prompts use markdown section headers to delimit task descriptions. 3. Capability inventory: Subagents can modify files, run tests, and commit code via referenced tools. 4. Sanitization: The risk is mitigated by a mandatory two-stage review process (spec compliance and code quality) that verifies all changes before completion.
- [COMMAND_EXECUTION]: The workflow involves subagents performing development tasks such as code implementation and test execution. These actions are within the intended scope of the skill and are supervised by the review gates.
- [PROMPT_INJECTION]: The skill contains instructions for 'Continuous execution' without pausing for human feedback. While this increases autonomy, the skill includes explicit 'Red Flags' and escalation logic (BLOCKED/NEEDS_CONTEXT) that require human intervention when necessary.
Audit Metadata