subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill processes external implementation plans and task descriptions, which are then provided to subagents for execution. Because these subagents are tasked with writing code and running tests, there is a potential for instructions embedded within the plans to influence the agent's behavior.
- Reduced User Oversight: The instructions specify that the agent should execute all tasks from a plan without stopping to check in with the human partner. While this is designed to improve speed and efficiency, it reduces the frequency of manual review checkpoints where a user might otherwise intercept errors or unintended actions.
- Subagent Instruction Isolation: The skill emphasizes that subagents should have isolated context and not inherit the main session's history. This is a positive security practice that limits the potential scope of any issues to the specific task at hand.
Audit Metadata