MeowKit

Appearance

meow:cook ​

End-to-end feature implementation pipeline with TDD enforcement, hard gates, and MeowKit's 7-phase workflow.

What This Skill Does ​

meow:cook is the primary entry point for building features in MeowKit. Given a task description, a plan file path, or a set of flags, it automatically detects your intent, selects the right workflow mode, and orchestrates the full 7-phase pipeline — orient, plan, test RED, build GREEN, review, ship, and reflect — without requiring you to invoke each step manually.

The skill enforces strict TDD: failing tests are written BEFORE implementation code. Gate 1 (plan approval) and Gate 2 (review approval) require human approval in ALL modes.

Core Capabilities ​

  • Smart intent detection — Analyzes your input to determine the right workflow mode
  • Six workflow modes — Interactive (default), fast, parallel, auto, no-test, code (from plan path)
  • Full TDD enforcement — Writes failing tests (Phase 2) before implementation (Phase 3)
  • Hard Gate 1 and Gate 2 — Plan approval and review approval enforced. Gate 2 is NEVER auto-approved
  • Model tier routing — Declares TRIVIAL/STANDARD/COMPLEX before work begins
  • Memory integration — Reads prior learnings at start, writes back at end
  • Gate validation scripts — Deterministic checks for plan/review completeness
  • Subagent delegation — Routes each phase to specialist agents via Task() tool

When to Use This ​

Use meow:cook when...

  • You're building a new feature from scratch
  • You have a plan file and want to execute it
  • You need the full pipeline: plan → test → build → review → ship
  • You want MeowKit to figure out the right workflow automatically

Don't use meow:cook when...

Usage ​

bash
# Natural language — cook detects intent automatically
/meow:cook add user authentication with JWT

# From an existing plan file
/meow:cook tasks/plans/260327-auth-flow/plan.md

# Fast mode — skip research, still requires plan + TDD-flavored tests
/meow:cook add login form --fast

# Parallel mode — spawn multiple agents for independent components
/meow:cook implement checkout system --parallel

# Auto mode — auto-fix issues, but Gate 2 still requires human approval
/meow:cook refactor payment module --auto

# No-test mode — skip TDD (use sparingly)
/meow:cook update readme --no-test

7-Phase Workflow ​

Phase 0: Orient → Phase 1: Plan [GATE 1] → Phase 2: Test RED
→ Phase 3: Build GREEN → Phase 4: Review [GATE 2] + Ship → Phase 5: Reflect
  1. Orient — Detect intent, declare model tier, read memory
  2. Plan — Research + create plan → Gate 1 (human approval)
  3. Test RED — Write failing tests from acceptance criteria
  4. Build GREEN — Implement until tests pass (TDD)
  5. Review + Ship — Code review → Gate 2 (human approval) → commit + PR
  6. Reflect — Sync plan, update docs, write memory

Gotchas ​

  • Skipping Gate 1 on "simple" features: Features that seem simple grow during implementation. Always create a plan file
  • Auto-approve sneaking bugs past Gate 2: Auto mode can auto-fix but NEVER auto-approve. gate-rules.md says NO exceptions
  • Context loss between phases: Long workflows exceed context. Update plan.md Agent State after each phase
  • Parallel mode deadlocks: Phase dependencies cause deadlocks. Map dependency graph before spawning
  • Code mode on stale plans: Running old plan against changed codebase. Warn if plan >14 days old
  • Fast mode shallow coverage: Skipping research = plan-level tests only, not edge cases
  • Missing model tier declaration: Always declare TRIVIAL/STANDARD/COMPLEX in Phase 0
  • Forgetting memory read/write: Phase 0 reads memory/lessons.md; Phase 5 writes back
  • Using Agent() instead of Task(): Task() enables tracking and blocking. Always use Task() for phases 2-5

Released under the MIT License.

Built with MeowKit