subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill instructions and subagent prompt templates were analyzed and contain no evidence of malicious intent, credential harvesting, or unauthorized network activity.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as task specifications and implementer reports. This surface is managed through the use of specialized subagents with isolated context windows and prompt templates that employ structural delimiters to separate data from instructions. The verification process requires subagents to inspect code directly, which helps mitigate the risk of instructions embedded in reports.\n- [COMMAND_EXECUTION]: The workflow incorporates standard development activities like git commits and automated testing performed by subagents. These operations are restricted to the context of the specific development tasks and are performed within isolated workspaces as recommended in the integration guidelines.
Audit Metadata