Testing Skills With Subagents

概要

This skill treats process documentation like code, utilizing a Test-Driven Development (TDD) framework to verify that instructions are bulletproof. By running scenarios first without the skill (RED), addressing failures (GREEN), and iteratively plugging loopholes (REFACTOR), it forces agents to adhere to critical discipline even when faced with time, sunk cost, or exhaustion pressures. It is particularly useful for skills that enforce strict rules or have high compliance costs, ensuring that Claude doesn't 'rationalize away' best practices in challenging real-world scenarios.

主な機能

• Provides frameworks for explicit negation rules and rationalization tables.
• Includes meta-testing techniques to evaluate documentation clarity and prominence.
• 0 GitHub stars
• Implements the RED-GREEN-REFACTOR cycle for process documentation.
• Utilizes multi-factor pressure scenarios (Time, Sunk Cost, Authority) to test compliance.
• Identifies and counters agent rationalizations verbatim to close loopholes.

ユースケース

• Auditing and bulletproofing existing documentation to prevent agents from taking shortcuts.
• Developing new Claude Code skills that require strict adherence to specific disciplines or standards.
• Verifying that high-stakes safety or quality processes remain intact during high-pressure scenarios.