test-driven-development— agent skill

Constraints

• Apply @rules/php/core-standards.mdc
• Apply @rules/code-testing/general.mdc
• If the current project uses Laravel, also apply @rules/laravel/laravel.mdc, @rules/laravel/architecture.mdc, @rules/laravel/filament.mdc, and @rules/laravel/livewire.mdc
• Follow test conventions from @skills/create-test/SKILL.md

Core principle

If you did not watch the test fail, you do not know whether it tests the right thing.

Iron law

NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST

Use when

• Implementing a new feature
• Fixing a bug
• Changing behavior
• Refactoring code that should remain behaviorally stable

Read, Map & Verify before the first RED (mandatory pre-flight)

Reading, mapping, and verifying come first; implementing comes last. This pre-flight is blocking — do not add or modify a single line of production code until all three steps pass, and never act on an assumption you have not confirmed by reading the code.

1. Read — open and read the actual target files and the code they depend on (callers, called methods, related tests, configuration). Confirm what the code does by reading it, not by guessing from names or the assignment description.
2. Map — map the change's blast radius: every call site, caller, data-flow path, and existing test that the behavior touches, plus the conventions and helpers already in the codebase to reuse instead of reinventing.
3. Verify — check your assumptions against the real code and its observed behavior (reproduce the current behavior so the first RED test asserts the real gap). If what you read contradicts the assignment framing, stop and surface the discrepancy instead of writing a test on a wrong premise.

Only after Read, Map, and Verify are complete may the first RED test be written.

Pre-flight (mandatory before the first RED)

Before writing the first failing test, run @skills/prepare-issue-context/SKILL.md with MODE=tdd and the assignment reference, scoped to the scenario(s) the upcoming RED step will cover. The skill seeds the development database with the records the failing test will depend on and captures a reproduction record (entry point + inputs + observed output) that becomes the arrange block of the first test. If the skill returns blocked: <count> open gap(s), stop and surface the gaps — writing a RED test against missing or guessed fixtures is the most common cause of stub-grade tests that drift from real behavior.

Required cycle

1. RED

Write one minimal test for the next behavior.

• Keep the test focused and readable
• Prefer real code paths; mock only where appropriate by project testing rules
• Do not generate covers()

2. VERIFY RED

Run the test and confirm:

• it fails
• it fails for the expected reason
• it is not failing because of syntax, setup, or typo issues

If the test passes immediately, it does not prove the new behavior.

3. GREEN

Write the smallest production change needed to make the test pass.

• Do not add extra features
• Do not broaden scope
• Do not refactor unrelated code yet

4. VERIFY GREEN

Run the relevant tests and confirm:

• the new test passes
• affected existing behavior still passes

5. REFACTOR

Only after green:

• remove duplication
• improve naming
• simplify code
• keep behavior unchanged

6. REPEAT

Move to the next behavior and repeat the cycle.

Commit boundary — the RED step is never its own commit

RED is a state of the working tree, never a state of the published history. The cycle above runs entirely before anything is committed: the failing test from step 1, the production change from step 3 that makes it pass, and the refactor from step 5 land together as one commit, made only once step 4 is green.

• Never commit a failing test. A committed RED step is a commit that cannot be deployed or cherry-picked, and git bisect will blame it for a regression it does not carry (@rules/git/general.mdc Git Rules — Every commit is green).
• Never simulate a failure in a committed test to represent work still to come — no ->skip() / ->todo() / markTestIncomplete(), no assertion inverted "for now", no commented-out assertion, no fixture pinned to the buggy output. If the behavior is not fixed yet, the test does not get committed yet.
• Several cycles, one commit each. When an item needs more than one RED → GREEN cycle, either commit each completed cycle (green at every step) or commit the whole item at the end — never a commit that stops between RED and GREEN.
• Repair in place. If a commit turns out red after a rebase or reshape, amend that commit; never append a repair commit at the tip (@skills/git-workflow/SKILL.md).

Bug-fix rule

Never fix a bug without first writing or updating a test that reproduces it.

Scope control

• Fix obvious blocking issues only when necessary for safe implementation
• Keep unrelated cleanup out of scope unless it is trivial and low risk

Post-cycle validation

1. Verify 100% code coverage for all changed or added code paths — if coverage tooling exists, run it.
2. Discover available fixers and checkers (prefer Phing targets from build.xml/phing.xml; fall back to Composer scripts in composer.json).
3. Run available fixers on changed files and fix any violations.
4. Run available checkers/analyzers on changed files and resolve all reported errors by rewriting the flagged code, never by adding a suppression annotation (@rules/php/core-standards.mdc PHP Practices).
5. Run a quick code review of all tests written during the TDD cycle against @rules/code-testing/general.mdc and fix any findings.

Done when

• Every implemented behavior is backed by a test
• Each new test was observed failing before implementation, and no failing or simulated-failing test was committed
• Production code was added only to satisfy failing tests
• Changed behavior, edge cases, and failure paths are covered
• Relevant tests pass
• 100% code coverage is verified for all changes
• Code style and quality checks pass (fixers and checkers ran clean)
• Test review passed with no findings
• Refactoring did not introduce new behavior