Testing Skills with Subagents

关于

Leverages the principles of Test-Driven Development (TDD) to rigorously test and refine Claude skills before deployment. This skill guides you through a RED-GREEN-REFACTOR cycle: first, observe an agent failing a task without the skill (RED), then write the skill to address those specific failures (GREEN), and finally, iterate by closing loopholes and countering rationalizations discovered through pressure testing (REFACTOR). It is essential for creating robust skills that enforce discipline and are not easily bypassed, ensuring they are effective in real-world scenarios.

主要功能

• Identifies, captures, and documents agent rationalizations for bypassing rules
• Includes a comprehensive TDD-style checklist for verifying skill robustness
• 11 GitHub stars
• Applies the RED-GREEN-REFACTOR cycle to skill and process documentation development
• Uses multi-faceted pressure scenarios (time, sunk cost, authority) to test skill compliance
• Provides a methodology for closing loopholes using explicit negations and rationalization tables

使用场景

• Developing a new skill that enforces a strict process or discipline (e.g., TDD, security protocols).
• Refining an existing skill that agents frequently bypass or find workarounds for.
• Before deploying any critical skill to ensure it performs reliably under high-pressure scenarios.