testing-skills-with-subagents

Apply RED-GREEN-REFACTOR cycles to diagnose and prevent skill bypass.

Updated Nov 3, 2025
One-click install
npx skills add https://github.com/ayourtch/kimichat --skill testing-skills-with-subagents-ayourtch
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: testing-skills-with-subagents
Source: https://github.com/ayourtch/kimichat/tree/main/skills/testing-skills-with-subagents
Command: npx skills add https://github.com/ayourtch/kimichat --skill testing-skills-with-subagents-ayourtch

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill ensures that new or edited skills are robust and effective by applying Test-Driven Development (TDD) principles to process documentation, verifying compliance under pressure.

Core Features & Use Cases

  • TDD for Documentation: Guides through RED (baseline test without skill), GREEN (write skill), and REFACTOR (close loopholes) cycles for skill content.
  • Pressure Scenarios: Creates realistic scenarios with combined pressures (time, sunk cost, authority) to test agent compliance and expose weaknesses.
  • Rationalization Prevention: Identifies and explicitly counters agent rationalizations, making skills "bulletproof" against shortcuts.
  • Use Case: When developing a new test-driven-development skill, use this skill to run a scenario where an agent is tempted to skip TDD, then refine the skill until the agent complies.

Quick Start

I'm using the testing-skills-with-subagents skill to verify my new code-review-process skill.

Frequently Asked Questions about testing-skills-with-subagents

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I test-drive documentation for AI agent skills to prevent them from bypassing intended behavior?

Test-driven documentation applies RED-GREEN-REFACTOR cycles to skill processes: run baselines without the skill to capture failures, write minimal skill fixes, then stress-test with combined pressures (time, sunk cost, authority) to expose rationalizations and close loopholes before deployment.

What's the best way to verify a new skill won't fail under pressure or rationalization?

Create multi-pressure scenarios that combine time constraints, authority influence, and sunk-cost bias to reveal where agents compromise skill compliance. Iterate through RED-GREEN-REFACTOR cycles, explicitly capturing and countering each rationalization until the skill remains bulletproof.

How do I identify when an AI agent rationalizes away from a skill's intended process?

Run baseline tests without the skill to establish verbatim failure modes, then apply pressure scenarios that combine competing incentives. Compare agent behavior against documented expectations to isolate rationalizations, then refine skill language or process clarity to eliminate escape routes.

Can I use test-driven documentation principles to audit existing skills for compliance gaps?

Yes. Run RED-phase baselines with the existing skill under normal and pressure conditions, document where behavior deviates from intent, then iterate GREEN-REFACTOR cycles to close gaps. This ensures deployed skills remain compliant across realistic scenarios.

What happens if a skill passes normal tests but fails under combined pressure?

That signals incomplete skill design. Pressure scenarios (time scarcity, authority claims, sunk costs) reveal rationalizations that normal tests miss. Return to GREEN-REFACTOR to tighten process wording, add explicit compliance checkpoints, or redefine boundaries until the skill withstands all pressures.