Spec-driven Spock Skill

Spec-driven Spock

Spock has a small but underused set of metadata annotations that turn an ordinary Specification into living design documentation:

| Annotation | What it pins | | ----------------- | ------------------------------------------------------------------------ | | @Title | One-line statement of what the spec is about (class) | | @Narrative | Multi-line "given / so that" backstory (class) | | @Subject | The class under test (class or field) | | @See | URL(s) to external references — ADR, RFC, ticket, docs (class or method) | | @Issue | URL(s) to issues / tickets specifically (class or method) | | @PendingFeature | Marker for a not-yet-implemented behaviour — see below | | @Unroll | Per-row report names for data-driven specs | | @Snapshot | Inject a snapshot for snapshot-driven tests (Spock 2.4+) |

Spock renders these into HTML reports and surfaces them in IDE plugins, which is why they pay rent. Stuffing the same information into Groovydoc comments is invisible to the runtime.

For the per-annotation cheat sheet, read references/annotations.md.

When to reach for this skill

A spec exists but a reader can't tell why it's there. Add @Title + @Narrative + @See.
A design decision (ADR, RFC, issue) lands and you want a test that fails if the decision is silently abandoned. Use the executable contract pattern.
A class spec has grown past ~250 lines and now mixes audiences. Split by concern, give each new file its own @Title/@Narrative.
A flag matrix or exit-code table is duplicated across N specs. Collapse with @Unroll.
A behaviour is documented but not yet implemented. Pin it as @PendingFeature — Spock fails CI if the implementation lands without the marker being removed.

Pattern 1 — Class-level scaffold

Every non-trivial spec should open with this header:

@Title("One sentence — what is this spec asserting?")
@Narrative('''
Given/so-that prose. Two or three short paragraphs at most. Explain
the *audience* for this spec and what kind of change should make this
spec evolve.
''')
@See([
    "https://github.com/<org>/<repo>/blob/master/adr/YYYYMMDD-<title>.md",
    "https://github.com/<org>/<repo>/pull/<n>"
])
@Subject(ClassUnderTest)
class MyFeatureSpec extends Specification { ... }

The @See URLs render as links in Spock HTML reports and most IDE plugins, so prefer stable URLs (a merged file path, a PR number, an issue URL) over local paths.

Pattern 2 — Per-method `@See`

Each feature method links to its design source:

@See("https://github.com/<org>/<repo>/blob/master/adr/2026-foo.md#d3--checksum-only")
def 'file outputs are compared by recorded checksum only'() {
    expect: ...
}

Use @Issue instead of @See when the link is specifically a bug or ticket. Both accept a single String or a list of Strings, and both can be repeated on the same target (Spock allows multiple @See annotations on one feature).

Don't add a @See if the only reasonable link is the file the test is already in — that's noise.

Pattern 3 — The executable contract

When an ADR / RFC lands, ship an executable contract alongside it: every numbered decision becomes a @PendingFeature stub that runs but stays skipped until the implementation fills it in.

@PendingFeature
@See("https://github.com/.../adr/foo.md#d7--auto-detect-ci")
def 'CI=true switches default output mode to --json'() {
    expect: false
}

Why expect: false?

Spock's @PendingFeature skips a test that fails and fails the build if it passes.
A body of expect: false always fails, so the stub stays skipped — green CI.
When a dev implements the behaviour, they replace the body with a real assertion. If they forget to remove @PendingFeature, the now-passing test triggers Spock's unexpected-pass and CI goes red — that's the signal.
A bare empty body won't compile (Spock requires at least one expect/then/when block).

This is the killer pattern for ADR-driven development. See references/adr-contract-pattern.md for a full worked example (a 4-file split of an 18-decision ADR), and assets/contract-spec-template.groovy for a copy-pasteable skeleton.

Pattern 4 — `@Unroll` for table-driven contracts

When the ADR / design pins a matrix (exit codes, format flags, status columns), use one @Unroll spec instead of N copy-pasted methods:

@Unroll
@PendingFeature
@See("https://.../adr/foo.md#d6--exit-codes")
def 'exit code is #code when #scenario'() {
    expect: false
    where:
    code | scenario
    0    | 'runs are semantically equivalent'
    1    | 'runs differ in any failing category'
    2    | 'load failure or schema mismatch'
}

Spock renders each row as a separately-named test (exit code is 0 when runs are semantically equivalent), so the HTML report reads as if you wrote three methods, but the source stays DRY.

Pattern 5 — Splitting large contracts

If a contract spec passes ~250 lines or mixes audiences, split it. Naming heuristic: one file per concern, one concern per audience.

For the lineage-validate ADR (18 decisions / 46 stubs), the split was:

| File | Decisions | Audience | | -------------------------------- | ---------------------- | ------------------- | | LineageValidateCliFlagsSpec | flags, env, exit codes | pipeline authors | | LineageValidateEquivalenceSpec | equivalence rules | reviewers | | LineageValidateBaselineSpec | resolver SPI, schema | plugin authors | | LineageValidateReportingSpec | diff shape, categories | tooling integrators |

Each file gets its own @Title / @Narrative / class-level @See. Each method keeps its own @See to the specific decision.

The split is organisational, not structural — the implementation side stays a single shared core. Resist proliferating implementation classes to match the spec split.

Pattern 6 — `@Snapshot` for snapshot-driven specs

Spock 2.4 ships first-class snapshot support. Inject a Snapshotter into the spec, point it at a directory, assert against it:

@Snapshot
Snapshotter snapshotter

def 'renders the report as documented'() {
    expect:
    snapshotter.assertThat(actualReport).matchesSnapshot()
}

Set spock.snapshots.update=true (or run with -Dspock.snapshots.update=true) to refresh snapshots when the contract intentionally changes. Keep the snapshot files in version control next to the spec so a PR's diff shows the contract change explicitly.

Pitfalls

Don't reference a static final String URL inside @See or @Issue. Annotation values must be compile-time constants, and Java/Groovy reject STATIC + "#section" concatenation in annotation arguments. Inline the full literal URL.
Don't put @Subject on a field that isn't actually the SUT. It's a documentation hint, not a wiring directive — claiming the wrong subject misleads readers.
Don't use @PendingFeature as a way to silence a flaky test. The marker is for "designed but not yet implemented", not "we know it's broken." Use @Ignore (with a reason) for flake.
Don't lose the safeguard. If you remove @PendingFeature from a stub without replacing the body, you get a normal red build — that's fine. But never leave a @PendingFeature stub with a body that always passes (e.g., expect: true); Spock will fail the build immediately, which defeats the contract.
GitHub anchors are fragile. Section headers change, anchor slugs drift. When the ADR is in the same repo, prefer linking to the file (no anchor) plus a // D7 — ... Groovy comment. Or use a tiny test that parses the ADR and verifies every @See anchor resolves.

Workflow

When the user asks for help on a Spock spec:

Read the spec. Note what the test asserts and what design decision motivates it.
Find the source of truth. ADR, RFC, ticket, or PR. If there isn't one, ask whether one should exist — the spec may be the de facto source.
Apply the class-level scaffold. @Title + @Narrative + @Subject + class-level @See pointing at the source.
Add per-method @See only where the link is more specific than the class-level one.
For ADR-driven work, draft a *ContractSpec companion with @PendingFeature stubs for every decision; split by audience if it grows past ~250 lines.
Run the suite to confirm @PendingFeature stubs are skipped (not failing, not unexpectedly passing).

When inheriting a legacy *Test class that doesn't follow this style, don't blanket-rewrite. Add annotations to the methods the current task touches; leave the rest. Consistency across a codebase matters less than making each spec individually self-explanatory.

Agent Skills: Spec-driven Spock

Install this agent skill to your local

Skill Files