Dispatching Incremental TDD

Core principle: Write ONE test at a time. Dispatch the agent instead of writing multiple tests or full implementation upfront.

When to Dispatch

Always dispatch incremental-tdd-agent when:

• Implementing new features
• Adding functionality to existing code
• Building new components or modules
• Creating APIs or interfaces
• Extending classes or services

Especially when:

• Feature seems "simple" or "obvious"
• You can "see the pattern" already
• Writing similar or related tests
• Implementing integration tests
• Under time pressure

How to Dispatch

Use Task tool with

subagent_type='incremental-tdd-agent'

:

Task(
  subagent_type='incremental-tdd-agent',
  prompt=`
    Feature: [What to implement]
    Context: [Existing code, patterns to follow]
    Goal: [Expected behavior when complete]
  `
)

What the Agent Does

The incremental-tdd-agent maintains ONE TEST AT A TIME discipline throughout feature implementation:

For each test:

1. RED Phase (Two-Stage)
   • Stage 1: Make it compile (write stub)
   • Stage 2: Make it fail correctly (verify test works)
2. GREEN Phase
   • Write minimal code to pass THIS test only
3. REFACTOR Phase
   • Evaluate design after each test
   • Make improvements before next test

Agent enforces:

• NO writing multiple tests at once (even if similar)
• NO skipping to implementation without failing test
• NO batching implementation for multiple tests
• NO skipping REFACTOR phase
• ONE test → GREEN → REFACTOR → repeat

Red Flags - Dispatch Agent Instead

If you catch yourself thinking:

• "Let me write tests for all scenarios first"
• "I can see the pattern, let me implement it"
• "These tests are similar, I'll batch them"
• "The feature is simple, I'll just write it"
• "Let me write the implementation first, then tests"
• "I'll test the full integration flow at once"
• "Refactoring can wait until feature is complete"
• "Making it compile is obvious, skip to implementation"

ALL of these mean: STOP. Dispatch incremental-tdd-agent.

Why Agent Pattern

Problem: ONE TEST discipline fades from context after 3-4 test cycles. By test 5-6, agents batch remaining tests and implementation.

Solution: Agent identity IS incremental TDD. Process maintained through 10+ test cycles.

Benefits:

• No context dilution during multi-test features
• Automatic discipline enforcement per test
• Design insights emerge test by test
• Process maintained through entire feature
• Clear boundaries (dispatch → feature complete)

After Agent Returns

Agent returns when:

• ✅ Feature fully implemented
• ✅ All tests passing
• ✅ Design evaluated through multiple REFACTOR cycles
• ✅ Incremental approach maintained throughout

You can continue with completed feature.

Common Mistakes

DON'T:

• Write multiple tests before dispatching
• Skip agent for "simple" features
• Batch similar tests "to save time"
• Skip REFACTOR after early tests
• Ignore agent's ONE TEST discipline

DO:

• Dispatch before writing first test
• Trust agent's incremental process
• Let agent complete full RED-GREEN-REFACTOR per test
• Learn design insights from each test cycle
• Maintain discipline through entire feature