testing-skills-with-subagents

Test-driven development for documentation ensures skills are robust by running baseline scenarios without the skill to observe failures, then validating the skill prevents those failures. Document agent rationalizations verbatim to identify and plug compliance loopholes before deployment.

Create pressure scenarios combining time, sunk cost, and authority constraints, then run RED-GREEN-REFACTOR cycles: document exact failures without the skill, verify the skill fixes them, and iteratively tighten rules to ensure compliance across multiple pressures.

Capture agent rationalizations verbatim during pressure testing to expose where skills fail or get circumvented. Use these exact rationalization tables to identify gaps and strengthen skill rules before production deployment.

Yes. TDD for documentation maps RED-GREEN-REFACTOR to skill creation: start with baseline tests showing agents fail without the skill, validate the skill prevents failures, then refactor rules to close loopholes and ensure compliance.

Pressure-based evaluation tests agent compliance under realistic constraints—time pressure, sunk cost, and authority—rather than isolated conditions. This reveals rationalization patterns and ensures skills hold when agents face competing priorities.

Yes. Baseline failure documentation is the first step in the RED-GREEN-REFACTOR cycle. Observe and record how agents fail without the skill, then design the skill to prevent those specific failures and achieve compliance.