test-driven-execution

Test-Driven Execution Skill

When to Trigger

• At the start of any feature or task from a plan
• When a bug fix is requested (test first, then fix)
• When a subagent is about to implement a module
• When "done" is ambiguous and needs a clear definition

Philosophy

"A feature is not done when code is written. It is done when the tests pass."

Tests written before code serve three purposes:

1. They force clarity on what "correct" actually means
2. They prevent the agent from drifting into over-engineering
3. They provide instant verification that the task is complete

Step-by-Step Process

1. Extract Acceptance Criteria

Read the task description and extract every "must" and "should":

Example task: "Add a user registration endpoint"

Extracted criteria:

• Must accept email, password, name in request body
• Must return 201 on success with the created user (excluding password)
• Must return 400 if email is already registered
• Must return 422 if fields are missing or invalid
• Must hash the password before storing

2. Write the Test Specification

Before writing any implementation, write the tests (or test outlines):

For unit tests:

describe('registerUser()', () => {
  it('creates a user with hashed password', async () => { ... })
  it('throws DuplicateEmailError if email already exists', async () => { ... })
  it('throws ValidationError if email format is invalid', async () => { ... })
  it('never stores plaintext password', async () => { ... })
})