subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Apr 11, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions create a surface for indirect prompt injection by incorporating external data into subagent prompts.
- Ingestion points: Plan files and task requirements are ingested and interpolated into the
implementer-prompt.mdandspec-reviewer-prompt.mdtemplates. - Boundary markers: The skill does not use boundary markers like XML tags, triple backticks, or specific instructions to the subagent to ignore commands within the injected text.
- Capability inventory: The implementer subagents have high-privilege capabilities including filesystem modification, test execution, and git operations (commit/branching).
- Sanitization: There is no process for sanitizing or validating the contents of the implementation plans before they are passed to subagents.
Audit Metadata