What is the Red-Green-Refactor cycle for skills?

It involves running a scenario without the skill to watch the agent fail (Red), writing a minimal skill to make the agent comply (Green), and iterating to close loopholes (Refactor).

How do pressure scenarios work in this context?

Pressure scenarios combine factors like tight deadlines, authority overrides, and sunk costs to see if the AI will prioritize convenience over following its skill instructions.

Why shouldn't I test simple reference skills?

Pure reference skills like API guides don't have rules to violate; stress testing is most effective for skills that enforce discipline or have high compliance costs.

What are 'rationalization tables' in skill testing?

They are lists of specific excuses an AI might use to bypass a rule, paired with explicit negations that prevent the AI from using those excuses in the future.

What is the goal of meta-testing?

Meta-testing involves asking the AI why it failed a test to determine if the skill was poorly written, poorly organized, or simply ignored.

Skill TDD & Stress Testing

Name: Skill TDD & Stress Testing
Author: seanGSISG

byseanGSISG

보안 및 테스팅

Validates and bulletproofs Claude Code skills by applying a Red-Green-Refactor cycle to process documentation using high-pressure subagent scenarios.

소개

Skill TDD & Stress Testing provides a rigorous framework for developing and verifying Claude Code skills through Test-Driven Development (TDD) principles. It enables developers to verify that their AI instructions hold up under pressure by running baseline tests where agents are expected to fail, writing skills to address those specific failures, and then iterating to close rationalization loopholes. This systematic approach ensures that AI agents remain compliant with complex workflows, such as strict TDD or security protocols, even when tempted by time constraints, sunk costs, or authority pressure.

주요 기능

Foundational principle enforcement for AI compliance
Verbatim rationalization capture and plugging
Red-Green-Refactor framework for skill documentation
High-pressure scenario generation (Time, Sunk Cost, Authority)
Meta-testing for instruction clarity and prominence
0 GitHub stars

사용 사례

Hardening process documentation against AI 'cutting corners' or excuses
Ensuring AI agents follow strict coding standards under tight deadlines
Verifying internal skills before deployment to production environments

소개

주요 기능

Foundational principle enforcement for AI compliance
Verbatim rationalization capture and plugging
Red-Green-Refactor framework for skill documentation
High-pressure scenario generation (Time, Sunk Cost, Authority)
Meta-testing for instruction clarity and prominence
0 GitHub stars

사용 사례

Hardening process documentation against AI 'cutting corners' or excuses
Ensuring AI agents follow strict coding standards under tight deadlines
Verifying internal skills before deployment to production environments