architect
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill manages a surface for indirect prompt injection as it analyzes existing codebases and user-provided design tasks to synthesize software architectures.\n
- Ingestion points: Data is ingested from external subsystems during the grounding phase and design specifications during the sketch phase.\n
- Boundary markers: The process enforces isolation between model runners using dedicated git worktrees or independent subdirectories.\n
- Capability inventory: The skill coordinates several sub-skills (arena, how, why, interrogate) and performs local file system operations.\n
- Sanitization: Includes a screening step against specific design red flags and provides an explicit human checkpoint (Phase C) to review the synthesized design before implementation.\n- [COMMAND_EXECUTION]: The workflow involves local execution of version control operations, specifically using
git worktreeto maintain isolated environments for parallel architectural explorations.
Audit Metadata