Testing Skills With Subagents

TDD-style testing for subagents involves documenting baseline failures, writing minimal rules to fix them, and re-running scenarios under pressure to prevent regressions. You capture verbatim rationalizations and iterate through red-green-refactor cycles until behavior remains compliant.

Rationalization tracking captures verbatim excuses or bypass attempts generated during subagent testing scenarios. It maintains a running table of these rationalizations to refine rules iteratively, preventing the tested skill from regressing or being easily bypassed under maximum pressure.

Baseline validation starts by designing realistic pressure scenarios and running them without the skill enabled. This reveals genuine failures and bypass attempts, which you then address by writing a focused minimal skill and re-running the tests to confirm the fixes.

Yes, testing skills under pressure requires no external tooling. The process relies entirely on clear scenario design, verbatim failure capture, and iterative refinement of rules to validate subagent behavior and prevent regressions without additional frameworks.

Red-green-refactor cycles are needed during skill creation and validation to ensure robustness. You run RED to observe failures, implement GREEN to fix them, REFACTOR until no new rationalizations emerge, then perform VERIFICATION GREEN and meta-test for compliance.

Verbatim failure capture in TDD skill testing depends on the quality of scenario design and explicit rationalization tracking. Its limitation is that it requires no external tooling, meaning the effectiveness relies heavily on manual iterative refinement and scenario pressure realism.