tdd

# TDD Skill — Red-Green-Refactor

You are guiding the developer through strict Test-Driven Development. You write code directly to the real files — the user can always undo with git. Pause only when the user's input is needed, not at every step.

## Phase 1: Setup

1. Determine the input type:
   - **If the input is a file path** (ends in `.md`, `.txt`, or exists on disk): read the plan file.
   - **If the input is an inline description** (e.g., `"add JWT authentication with refresh tokens"`): use it directly as the feature specification. Ask clarifying questions with `AskUserQuestion` only if the description is too vague to decompose into increments (e.g., "build an app"). A single sentence like "add user login with email and password" is enough to start.
2. Detect the language and test framework from the project. If ambiguous, ask:
   - Python → pytest
   - TypeScript → vitest
   - Go → testing (stdlib)
   See `references/language-configs.md` for runner commands and conventions.
3. Break the feature into small, testable increments. See `references/increments.md` for decomposition patterns.
4. Create a task list using `TaskCreate` — one task per increment.
5. Present the increment list to the user with `AskUserQuestion`:
   - "Here are the increments I've identified. Want to reorder, add, remove, or modify any before we start?"
   - Options: "Looks good, let's start" / "I want to modify the list"

## Phase 2: TDD Loop

For each increment, follow Red-Green-Refactor strictly. Mark the current increment as `in_progress` via `TaskUpdate`.

### RED — Write a Failing Test

1. Write the failing test **directly to the real test file** (e.g., `tests/test_feature.py`, following existing project conventions). Show the test in a fenced code block and briefly explain what behavior it verifies.
2. Run the test using the appropriate runner (see language-configs.md). Use `Bash`.
3. Confirm the test **fails**. If it passes unexpectedly, stop and flag this:
   - "The test passed already — this means either the behavior is already implemented or the test isn't asserting the right thing. Let's investigate before moving on."
4. Pause with `AskUserQuestion`:
   - "RED: test fails as expected. Review the test above — ready to move to GREEN?"
   - Options: "Looks good, write the code" / "I want to change the test first"

### GREEN — Write Minimal Code to Pass

5. Write the **minimal** production code **directly to the real source file** to make the failing test pass. Show it in a fenced code block. Write only enough to pass, nothing more.
6. Run **all** tests (not just the new one). Use `Bash`.
7. If all tests **pass**, move directly to REFACTOR — no pause needed.
8. If any test **fails**, show the failure output and pause:
   - "A test failed unexpectedly. Here's the output. Want me to fix it, or do you want to handle it?"
   - Options: "Fix it" / "I'll handle it"

### REFACTOR — Improve the Code

9. Review the current code. If refactoring opportunities exist (duplication, naming, extraction, simplification), apply them directly and run all tests to confirm green. Show what you changed in a fenced code block.
10. If no refactoring is needed, say so and move on — no pause.
11. If refactoring causes a test failure, revert and pause to discuss.

### NEXT

12. Mark the increment as `completed` via `TaskUpdate`.
13. Briefly summarize: what test was added, what code was written, what was refactored (if anything).
14. Move directly to the next increment — go back to RED. No pause between increments.

## Phase 3: Wrap-up

After all increments are complete:

1. Summarize what was built:
   - List each increment and its test
   - Note the final test count and all-passing status
2. Suggest any remaining work (edge cases, integration tests, documentation).

## Rules

- **Write directly to real files.** No temp files, no asking the user to move code. The user has git for undo.
- **Pause only when needed:** after RED confirms failure (so the user can review the test), and when something goes wrong (unexpected pass, test failure at GREEN, refactoring breakage). Do NOT pause at GREEN success or REFACTOR success.
- **Run the full test suite** at Green and Refactor steps, not just the new test.
- **Keep tests behavior-focused.** Test what the code does, not how it does it.
- **Simplest case first.** Start with the degenerate/edge case, build toward the general case.
- **One assertion per test** when possible. Each test should verify one behavior.
- **Track progress visually** using TaskCreate/TaskUpdate so the user sees where they are.
- **Follow existing project conventions.** Match the test file locations, naming patterns, and style already in the codebase.