plan-harder
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by consuming and processing external data from a user's codebase and a subagent's review feedback.
- Ingestion points: The agent reads the local codebase in Phase 0 (Research) and subagent feedback in Phase 5 (Review).
- Boundary markers: Absent. There are no instructions to use delimiters or to treat external data as untrusted instructions.
- Capability inventory: The skill can write files to the system (Phase 3), trigger subagent reviews (Phase 5), and prompt the user for input.
- Sanitization: Absent. The instructions do not include steps for validating or escaping content retrieved from the codebase or the subagent before it is incorporated into the saved implementation plan.
Audit Metadata