Skip to content
Back to skills

Dispatching Incremental Tdd

ASecurity

Use when implementing features or adding functionality - dispatches incremental-tdd-agent that enforces ONE test at a time, preventing batch test writing that defeats incremental design discovery. Maintains discipline through entire feature implementation (10+ test cycles)

  • 666 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added February 8, 2026
developmenttypescriptrustgorefactoringapi

Works with

  • api

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add majiayu000/claude-skill-registry --skill dispatching-incremental-tdd --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Dispatching Incremental Tdd?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Dispatching Incremental Tdd
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-dispatching-incremental-tdd/badge)](https://www.skillsdirectory.com/skills/majiayu000-dispatching-incremental-tdd)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: dispatching-incremental-tdd
description: Use when implementing features or adding functionality - dispatches incremental-tdd-agent that enforces ONE test at a time, preventing batch test writing that defeats incremental design discovery. Maintains discipline through entire feature implementation (10+ test cycles)
---

# 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'`:

```typescript
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

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.