tdd

Files (9)

• SKILL.md
• openai.yaml · agents/
• deep-modules.md · references/
• interface-design.md · references/
• mobile.md · references/
• mocking.md · references/
• refactoring.md · references/
• tests.md · references/
• web.md · references/

SKILL.md (first lines)

# Test-Driven Development

Use one behavioral contract to drive one minimal implementation increment at a time.

## Select the adapter

- Backend, library, or domain behavior: use the core workflow below.
- Web UI with Playwright: read [references/web.md](references/web.md).
- React Native UI with Maestro: read [references/mobile.md](references/mobile.md).

For deeper decisions, read only what is needed:

- Test quality: [references/tests.md](references/tests.md)
- Mocking: [references/mocking.md](references/mocking.md)
- Interface design: [references/interface-design.md](references/interface-design.md)
- Deep modules: [references/deep-modules.md](references/deep-modules.md)
- Refactoring: [references/refactoring.md](references/refactoring.md)

## Core loop

1. Confirm the observable behavior and its public interface.
2. Choose the smallest test seam that proves the behavior.
3. Write one failing test and run it to observe the expected failure.
4. Implement only enough behavior to pass.
5. Run the focused test, then relevant surrounding tests.
6. Refactor only while green.
7. Repeat with the next behavior.

```text
RED -> GREEN -> REFACTOR -> next behavior
```

## Invariants

- Test behavior, not private implementation.
- Prefer real code paths and local substitutes over internal mocks.
- Never write the entire test suite before implementation.
- Never claim RED without observing the intended failure.
- Never claim GREEN without rerunning the test.
- Keep one cycle reviewable and reversible.
- Keep production diagnostics free of secrets, tokens, passwords, and PII.

## Planning

Before code changes, align on:

- Public behavior and interface
- Highest-risk scenarios
- Test level and adapter
- Required fixtures or external dependencies
- Expected failure signal

Use the project's domain language in test names and respect documented architecture decisions.

## Completion evidence

Report:

- Test added or changed
- Observed RED failure
- Minimal implementation
- GREEN command and outcome
- Refactor performed
- Remaining untested risk

From deepaktiwari09/team-agent-skills, MIT.

More from deepaktiwari09/team-agent-skills