ck-test

name	ck-test
description	Standalone testing skill — owns writing and running tests, verifying behavior, and two-pass TDD orchestration. Default mode analyzes changed scope and runs the relevant tests; `--unit`/`--integration`/`--e2e`/`--all` scope by test type; `--verify` reruns exactly the prior failures; `--all-phases <plan>` sweeps every completed phase for regression. `--tdd --prepare`/`--tdd --verify` bracket `ck:cook` for red-green TDD. Never edits production code, and blocks until a phase's `ck:quality` gate is APPROVED (except `--tdd --prepare`, which runs before any implementation exists).
user-invocable	true

ck:test — Standalone Verification & TDD Orchestration

ck:test owns all verification. It never touches production code and never calls ck:fix itself — a failure always hands control back to the user.

Modes (pick exactly one; --tdd further requires --prepare or --verify):

• default <plan-or-phase> — analyze the changed scope, write any missing tests, run the relevant unit + integration suites.
• --unit <target> / --integration <target> / --e2e <target> — run only that test type against <target>.
• --all <target> — run every applicable suite for <target>'s scope (not necessarily the whole repo).
• --verify <target> — rerun exactly the tests that previously failed, plus directly related regression tests; does not redesign the suite.
• --all-phases <plan-path> — regression sweep across every already-completed phase in the plan.
• --tdd --prepare <phase-file> — write failing tests before ck:cook implements anything; confirm RED (failure is due to missing implementation, not a broken test); save a RED_READY artifact.
• --tdd --verify <phase-file> — after ck:cook implements, rerun; confirm GREEN plus regression; detect tests that were weakened, deleted, or modified to dodge a failure rather than fixed by the implementation.

---

Step 1 — Quality Gate Check

Skip only for --tdd --prepare, or for --unit/--tdd --verify explicitly orchestrated by the active ck:cook phase after its Phase Checkpoint recorded tests=yes. This scoped exception lets tests and quality remain independent user choices; it does not apply to standalone test requests.

For every other mode, resolve the phase(s) in scope and verify each has a valid, current quality receipt:

python {ck-quality-skill-root}/scripts/receipt.py verify plans/{slug}/quality/{phase}-receipt.json

Resolve {ck-quality-skill-root} from the installed ck:quality skill directory for the active client; do not hardcode skills/ or .claude/skills/.

A missing or invalid receipt stops the run before any test is written or executed:

BLOCKED: phase implementation has not passed the Code Quality Gate.
Run or resume /ck:cook for the current phase.

Step 2 — Resolve Scope

• default / --unit / --integration / --e2e / --all: the phase's changed files (git diff since the phase's last session-notes timestamp), or the given <target> path.
• --verify: exactly the failures[].related_files from the report being verified, plus the tests that exercise them — refreshed against current git diff in case scope grew.
• --all-phases: every phase in plan.md/plan.json already completed with quality.status: approved.
• --tdd --prepare / --tdd --verify: the single named phase file.

Step 3 — Run Tester

If a named tester agent is available, spawn it with: scope, mode, existing test conventions (read 2-3 sibling test files first), and — for --tdd --verify — the prepared RED_READY artifact's exact test file contents. Otherwise execute the tester process inline. The inline path follows every ownership constraint and writes the same report; it exists so the skill remains standalone on Codex and Antigravity.

tester may create or edit only test code, fixtures, mocks, and test helpers. It never touches production code, never weakens an assertion to make a test pass, never deletes a failing test, and never marks a test passed without having run it.

Step 4 — Run and Verdict

Run the relevant suite(s) for the resolved scope (not necessarily the whole repo, unless --all-phases).

Mode	Verdict values
--tdd --prepare	RED_READY (every new test fails only because the implementation doesn't exist yet) or BLOCKED (a test fails for an unrelated reason, e.g. a syntax error in the test itself)
default / --unit / --integration / --e2e / --all / --verify / --all-phases / --tdd --verify	PASSED or FAILED

--tdd --verify additionally reports FAILED (category WEAKENED_TEST) if any prepared test's assertions were loosened, deleted, or its scope narrowed without a corresponding, justified change to the acceptance criteria — regardless of whether the suite otherwise passes.

Step 5 — Report and Handoff

Write the report to plans/{slug}/tests/{phase}-test-report.json (matching references/test-report-schema.json); --tdd --prepare writes plans/{slug}/tests/{phase}-tdd-ready.json (matching references/tdd-ready-schema.json) instead. --all-phases writes one report per swept phase — never one aggregate file covering multiple phases — then prints a one-line PASSED/FAILED summary per phase.

Update the phase's testing.status (JSON) or Quality and Testing State section (Markdown) to mirror the verdict: PASSED → passed, FAILED → failed, BLOCKED → blocked_on_quality. ck:test owns this field; nothing else writes it. Field shape and allowed values are defined once in ck-plan-json/rules/plan-design.md § Quality and Testing State — don't redefine them here.

On FAILED, stop — do not call ck:fix automatically. Print:

[ck:test] {mode} → FAILED
Report: plans/{slug}/tests/{phase}-test-report.json
Next: /ck:fix --from-test plans/{slug}/tests/{phase}-test-report.json, then /ck:test --verify {phase}

On RED_READY, print the artifact path and hand control back to ck:cook:

[ck:test] --tdd --prepare → RED_READY
Artifact: plans/{slug}/tests/{phase}-tdd-ready.json
Next: /ck:cook {phase-file} --tdd

On PASSED, print a short summary (passed/failed/skipped counts) and the report path; no further action is implied.

Constraints

• Never edit production code — only test code, fixtures, mocks, and test helpers.
• Never call ck:fix itself; a FAILED or BLOCKED report always hands off to the user.
• A phase without an APPROVED, current quality receipt blocks every mode except --tdd --prepare.
• Do not mark a test passed without having actually run it in this invocation.
• --verify and --tdd --verify re-run the full previously-failing set, not a subjective subset — do not declare PASSED on a partial rerun.

---

Report shape: references/test-report-schema.json. TDD-prepare artifact shape: references/tdd-ready-schema.json.