Testing Prompts With Subagents

Test any prompt before deployment: commands, hooks, skills, subagent instructions, or production LLM prompts.

Overview

Testing prompts is TDD applied to LLM instructions.

Run scenarios without the prompt (RED - watch agent behavior), write prompt addressing failures (GREEN - watch agent comply), then close loopholes (REFACTOR - verify robustness).

Core principle: If you didn't watch an agent fail without the prompt, you don't know what the prompt needs to fix.

REQUIRED BACKGROUND:

• You MUST understand tdd:test-driven-development - defines RED-GREEN-REFACTOR cycle
• You SHOULD understand prompt-engineering skill - provides prompt optimization techniques

Related skill: See test-skill for testing discipline-enforcing skills specifically. This command covers ALL prompts.

When to Use