gauntlet-loop

Pass

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is primarily instructional, teaching the agent how to implement a specific iterative prompting architecture (Split-Build-Judge-Repeat). It provides references to community discussions and original articles on the technique.
  • [EXTERNAL_DOWNLOADS]: The skill mentions external tools and scripts (e.g., scripts/precheck.py and google/gemini-3.7-flash), but these are described as architectural components for a developer workflow rather than immediate malicious payloads. The referenced URLs point to legitimate GitHub repositories (mshumer/Claude-of-Duty) and an official methodology site (somethingbig.ai).
  • [COMMAND_EXECUTION]: While the skill instructions assume the agent has access to tools like terminal or execute_code, it does not provide any hardcoded malicious commands. It warns about the risks of autonomous agents (e.g., agents rewriting files without permission) and recommends using checkpoints and kill switches as best practices.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-defined 'goals' and 'reference bars'. It manages the risk of instructions within that data by recommending fresh sub-agent contexts for critics, ensuring the critic evaluates the artifact's quality rather than obeying instructions embedded within the artifact itself.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 3, 2026, 10:26 AM