Zombies

ZOMBIES

Arguments

Raw arguments: $ARGUMENTS

If arguments are provided, treat them as a free-text feature description (e.g. "sign-in code login flow", "image upload validation"). Locate the relevant code and tests yourself using Grep/Glob.

If arguments are empty, run git diff main...HEAD and use the diff as the feature scope.

Goal

Identify the most valuable tests to write for the feature, using ZOMBIES as a thinking tool. The output is a list of test ideas the user can stub out themselves — do not write or stub the tests.

ZOMBIES stands for:

• Zero — no inputs / empty state
• One — a single input / the happy path
• Many — multiple inputs, ordering, pagination, concurrency
• Boundaries — limits, off-by-one, min/max lengths, timing edges, type edges
• Interface — the contract/shape of the public API (return types, status codes, redirects)
• Exceptions — invalid input, failures, expired/used/missing state, auth failures
• Simple scenarios — the common everyday usage paths a real user takes

Skip categories that don't apply. A read-only endpoint may have nothing under "Many". A pure validator may have nothing under "Interface". Only list tests that are genuinely worth writing — quality over coverage.

Instructions

1. Locate the feature

• With args: search the codebase with Grep/Glob for files matching the description. Read the implementation files (controllers, models, actions, validators) and the existing test file if one exists.
• Without args: run git diff --name-status main...HEAD and git diff main...HEAD, then read the changed implementation files. Skip auto-generated files (lockfiles, compiled assets, generated route/type definitions).

2. Generate ZOMBIES suggestions

For each ZOMBIES letter, ask: is there a test here that would catch a real bug or document real behaviour? If yes, list it. If no, skip the letter.

Cross-reference against the existing test file. This skill outputs tests to write, not a coverage map — so:

• Skip behaviours already fully covered by an existing test. A covered behaviour is not a test to write; listing it buries the gaps that matter.
• Keep a behaviour that's only partially covered — where a test exercises it but misses an important assertion (e.g. asserts a successful login redirect but never checks the code is consumed). Prefix the bullet with [partial] and name the missing assertion.
• When unsure whether a test covers a behaviour, keep the bullet rather than dropping it — a false "already covered" silently hides a real gap.

Suggestion quality bar:

• Specific, not generic. "Test maximum email length (255 chars)" beats "Test boundaries".
• Reference real values from the code when possible — column lengths from migrations, expiry windows from config, validation rules from FormRequests.
• One test per bullet. Don't combine "test A and B" into one line.
• Phrase as a behaviour to verify, not as a method name. "Expired sign-in code returns 422" beats "test_expired_code".

3. Output the report

Group by feature area first (if the diff covers multiple features), then by ZOMBIES letter within each. Letters always appear in ZOMBIES order — Zero, One, Many, Boundaries, Interface, Exceptions, Simple. Skipping a letter never reorders the rest: the letters you do show must still run top-to-bottom in that fixed sequence (e.g. show Boundaries before Interface before Exceptions, even if Zero, One, and Many were all skipped). Use this format exactly:

## 🧟 [Feature Area]

**Boundaries**
- Email field rejects values longer than 255 chars (matches migration column)
- Sign-in code expires exactly at the 15-minute mark

**Exceptions**
- Expired code returns a validation error
- Already-used code cannot be redeemed twice
- Submitting code for an unknown email fails silently (no user enumeration)

**Simple**
- Requesting a code emails the user and creates a `SignInCode` row
- [partial] valid code logs the user in — existing test asserts the redirect but never asserts the code is consumed

The 🧟 prefix and ## level are what make a feature heading stand out from the bold **Letter** sub-headings — keep both. If multiple features are in scope, repeat the block per feature, separating each with a --- horizontal rule so the boundaries between features are obvious.

End with a one-line summary: X test ideas.

If there's nothing worth testing (e.g. trivial rename, pure config change), output exactly:

✅ Nothing worth writing tests for.

Rules

• Don't stub the tests. This skill outputs ideas only — the user writes the tests.
• Skip ZOMBIES letters that don't apply. Do not write "(none)" placeholders. Quality over coverage.
• Gaps only. Skip behaviours an existing test already fully covers. Keep partially-covered behaviours, prefixed with [partial] and naming the missing assertion. When unsure, keep the bullet — never silently drop a real gap.
• Always preserve ZOMBIES order. The displayed sections must follow Zero → One → Many → Boundaries → Interface → Exceptions → Simple. Skipping letters is fine; reordering the remaining ones is not.
• One heading per letter, per feature area. Each ZOMBIES letter appears at most once within a feature area — collect all of that letter's bullets under its single heading. Never repeat a letter's heading.
• Be specific. Reference actual lengths, timings, statuses, route names from the code. Generic suggestions are worthless.
• One behaviour per bullet. No "and" joining two tests.
• No implementation hints. Don't suggest assertions, factories, or test setup — just what to verify.
• Group by feature first, then by letter. Don't dump everything under one giant ZOMBIES list when the diff spans multiple features.
• No preamble. No "Here are the tests I'd suggest…". Start with the first ## [Feature Area] heading.
• No closing advice beyond the X test ideas. summary line.