ck-cook

name

ck:cook

description

Implement a planned feature phase by phase. Use when the user says "cook this", "implement it", "let's build", "start coding", or passes a plan.md path. Spec-aware — auto-loads spec.md alongside plan for SDD+TDD. Modes (pick one): --fast (skip test/review), --hard (mandatory human approval). Composable flags (combine with any mode): --no-test (skip tester), --tdd (write failing tests before implementing).

ck:cook — Structured Implementation Pipeline

Modes — mutually exclusive, pick one (default = Standard):

• Standard — test + review, auto-approve if score ≥ 9.5 with 0 CRITICAL
• --fast — skip tester and code-reviewer; git-manager only in Step 5
• --hard — mandatory test + mandatory review, no auto-approve

Composable flags — combine with any mode:

• --no-test — skip tester; go directly to Step 3.S → Step 4
• --tdd — write failing tests first, then implement until they pass

---

Step 0 — Plan Check

When no plan path provided:

1. Search plans/ for any plan.md → ask: "Found {path}. Use this? [Y/n]"
2. If none found → ask: "No plan found. Continue anyway? [y/N]" — if No, suggest /ck:plan

After resolving plan path: check for spec.md in the same directory. If found, load it — activates spec-driven mode for Steps 1 and 2.

---

Step 1 — Load Plan / Detect Mode

Report what will be cooked:

Plan: {Feature Name}
Status: {status from plan.md}
Mode: {Standard | Fast | Hard}
Test:  {default | --no-test | --tdd}
Spec:  {plans/{slug}/spec.md — N P1 stories, N success criteria | none}
Phases remaining:
  [ ] Phase 1: ...
  [ ] Phase 2: ...

If spec loaded + --tdd not set: Spec detected. Consider --tdd: acceptance criteria in spec.md are ready-made test anchors.

If ## Session Notes exists in plan.md: output resume state and continue from where it left off.

When no plan file provided: read the feature request, ask 2–3 clarifying questions, proceed once clear.

---

Step 2 — Implement

For each phase-XX-*.md in order:

1. Read phase file — understand requirements, architecture, steps, success criteria
2. Implement following codebase conventions
3. Verify success criteria for the phase
4. If spec loaded: P1 coverage: {N}/{total} stories addressed this phase
5. Write (overwrite) ## Session Notes in plan.md, then mark phase complete - [x] Phase N: {name}
6. Report what was done

Session Notes template (overwrite, never append):

## Session Notes
<!-- Updated by cook automatically — do not edit manually -->

**Last active:** {YYYY-MM-DD HH:MM}
**Phase in progress:** {phase-XX-name}
**Status:** {one-line status}

### Decisions made this session
{bullet list of non-obvious decisions, or "(none)"}

### Next immediate action
{what cook will do next}

Review Gate — after each phase:

• Standard / --hard: pause and wait for user approval
• --fast: continue automatically

Stop if: success criterion unverifiable, unexpected blocker, or phase needs user decisions not in the plan.

---

Step 3 — Test (tester sub-agent)

--fast / --no-test: skip → Step 3.S.

[Build Gate]: verify compilation before tests. On failure: [GATE FAIL] Build gate: compilation errors — fix before testing.

Default: spawn tester → writes tests, runs full suite (100% pass required) → on failure: spawn debugger → fix → re-test.

Remediation cycles: each of cycles 1–3 must use a different approach than previous. Cycle 4: STOP.

[ESCALATION] Test remediation exhausted
File:    {path/to/failing_test}
Error:   {exact error message}
Cycles:  {approach 1} | {approach 2} | {approach 3}
Action:  Awaiting user guidance

--tdd: invert per phase:

1. tester writes failing tests (red) — from ### Tests to Write First or spec acceptance criteria
2. Confirm red before implementing
3. Implement until green, full suite passes

---

Step 3.S — Auto-Simplify

Check if SIMPLIFY_TRIGGERED in context (emitted by code-simplifier hook).

If triggered: invoke simplify skill on files edited this phase → delete simplify tracker → proceed to Step 4. If not triggered: skip silently.

Thresholds (.ck.json → simplify.threshold): totalLoc 400, fileCount 8, singleFileLoc 200.

---

Step 4 — Code Review

--fast: skip → Step 5.

[Test Gate]: all tests must pass (or --no-test set).

Spawn code-reviewer: correctness, security, regressions, quality → APPROVED / WARNING / BLOCK.

• Standard: auto-approve if score ≥ 9.5 with 0 CRITICAL
• --hard: no auto-approve — human must approve before Step 5
• Fix/re-review up to 3 cycles (different approach each), then escalate

---

Step 5 — Finalize (MANDATORY)

[Approval Gate]: code-reviewer APPROVED required (or --fast bypass).

project-manager (skip --fast): mark phases [x], update plan status.

docs-manager (skip --fast): update docs, README, API contracts.

If spec loaded: output before git-manager:

# Spec Coverage
P1 stories:        {N}/{total} covered
Success criteria:  {N}/{total} verifiable
Uncovered P1:      {list any, or "none"}

git-manager (always): conventional commits → ask to push.

---

Agents

Agent / Skill	Step	Modes
tester	3	Standard, --hard (skip for --fast, --no-test)
debugger	3	When tests fail
simplify skill	3.S	All (hook-driven)
code-reviewer	4	Standard, --hard (skip for --fast)
project-manager	5	Standard, --hard (skip for --fast)
docs-manager	5	Standard, --hard (skip for --fast)
git-manager	5	Always (mandatory)