arena
Pass
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes outputs from multiple subagents during evaluation and grafting phases, creating a surface for indirect prompt injection.
- Ingestion points: Phase D (Pick a base) and Phase E (Graft) require the agent to read candidate artifacts and rationales produced by subagents.
- Boundary markers: The instructions do not mandate the use of delimiters or specific directives to ignore instructions that may be embedded within the candidate outputs.
- Capability inventory: The skill utilizes the
spawn_agenttool to create sub-processes and performs file operations within local paths and git worktrees. - Sanitization: No explicit steps are provided to sanitize or validate the content generated by subagents before it is synthesized into the final artifact.
Audit Metadata