What problem does it solve?
This Skill addresses the critical problem of creating ineffective or ambiguous documentation for AI agents. It ensures that new skills are rigorously tested and bulletproofed against agent rationalization, leading to higher quality, more reliable agent behavior, and reduced debugging time for skill authors.
Core Features & Use Cases
- Test-Driven Documentation (TDD): Applies the RED-GREEN-REFACTOR cycle to skill authoring, requiring authors to run failing scenarios with subagents before writing skill content.
- Rationalization Countering: Guides authors to identify and explicitly counter agent rationalizations observed during testing, closing loopholes and enforcing desired behaviors.
- Claude Search Optimization (CSO): Provides guidelines for crafting effective skill names, descriptions, and keywords to maximize discoverability and efficient loading by future Claude instances.
- Use Case: When developing a new skill, an author first runs a subagent on a task without the skill. If the subagent fails or rationalizes, the author documents this, then writes the minimal skill to address the failure, and re-tests until the subagent complies.
Quick Start
Before writing any new skill or editing an existing one, first create a pressure scenario with a subagent to observe its baseline behavior without the skill, documenting any failures or rationalizations.
