TDD Workflow

Summon enforces Test-Driven Development with a strict separation: Tara writes the failing test, Sato makes it pass.

Why Separate Agents?

When the same person writes both the test and the implementation, there’s a natural tendency to write tests that confirm the code rather than challenge it. By splitting the roles:

• Tara writes tests from the acceptance criteria, not the implementation
• Sato writes the minimum code to make those tests pass
• Neither agent compromises the other’s perspective

The Cycle

Red Phase (Tara)

1. Read the acceptance criteria from the work item
2. Write a failing test that captures the expected behavior
3. Verify it fails for the right reason (not a syntax error)

Green Phase (Sato)

1. Read the failing test
2. Write the simplest code that makes it pass
3. No more, no less — don’t anticipate future requirements

Refactor Phase (Sato + Vik)

1. Clean up while tests stay green
2. Vik reviews for simplicity and pattern adherence
3. No new functionality during refactor

Running TDD

Use the /tdd command:

/tdd implement the user authentication feature

This invokes the full cycle: Tara writes tests, Sato implements, review follows.

Tara’s Veto

Tara has veto power on test coverage. If a work item doesn’t have adequate tests:

• The Done Gate fails
• The item doesn’t close
• No exceptions

This isn’t bureaucracy — it’s the discipline that keeps the codebase trustworthy.