If you're using Claude Kit to accelerate your software development workflow, you've probably asked yourself: "Why should I run /plan first and then /code, when /cook can do everything in a single command?"
The short answer: /cook is indeed a powerful weapon, but like any powerful tool, it requires a smart user with a clear strategy.
1. The "/plan + /code" Combo: For Those Who Prefer Control
The standard Claude Kit workflow is typically recommended to separate two phases:
Planning (/plan): The Planner Agent researches, analyzes the codebase, and creates a detailed markdown guide file.
Execution (/code): The Developer Agent reads that plan file and starts coding, testing, and reviewing.
Why Choose This Approach?
Absolute Control: You have the opportunity to read and edit the plan file before any code is written. If the AI misunderstands your intent, you fix it on paper - much cheaper than fixing code.
Token Savings: This is a critical factor. The /code command is designed to execute an existing plan. It doesn't need to spend resources thinking from scratch.
Rigorous Process: /code has mandatory quality control gates: Tests must pass 100%, Code Review must be clean of Critical errors, and User must Approve before merging.
2. The "/cook" Command: The Power of Full Automation
The /cook command embodies convenience. When you type /cook [feature description], you're triggering a massive automated pipeline:
The /cook Flow:
Planning (if none exists) → Scout (find files) → Implementation → Testing → Code Review → Documentation → Report
It does everything. You can go make coffee and come back to a completed feature. But precisely because of this power, /cook is easily misused.
3. Why "/cook" Is Only for Smart Users
Many people mistakenly think /cook is a button for lazy people. Wrong! /cook is a tool for those who know how to give precise commands.
Here's why you need to be "smart" when using /cook:
A. You Must Know Exactly What You Want (Prompt Engineering)
Because /cook automatically transitions from planning to code immediately, if your input is vague, the output will be a mess.
# ❌ Poor Input
/cook add stuff
# → AI will be confused and produce wrong results
# ✅ Smart Input
/cook implement password reset flow with email verification
# → AI clearly understands the scope of work
If you can't describe in detail, use /plan to have AI help clarify your idea first. Using /cook for unclear ideas is burning tokens (money) uselessly.
B. You Must Know Resource Management (Token Management)
A "fatal" mistake of beginners: Already ran /plan, but then ran /cook again with the same description.
Consequence: /cook will create a new plan, wasting double the tokens for the same work.
Golden Rule:
- No existing plan → Use
/cook - Plan file exists (
plans/xyz.md) → Use/code @plans/xyz.md
C. You Must Have Discipline (Trust the Process)
When /cook is running, it triggers a series of Agents: from Planner, Scout, to Tester and Code Reviewer. Don't interrupt it midway.
Interrupting will lead to incomplete code, untested features, or outdated documentation.
A smart user knows to be patient to receive complete results: Code + Tests + Docs.
Conclusion
Use /plan + /code when:
- You want to save costs
- You need certainty
- You're working with extremely complex features requiring step-by-step review
Use /cook when:
- You want speed
- You're confident in your prompt-writing ability
- You want to maximize the orchestration power of 14 Agents in Claude Kit
Use /cook like a good manager delegating to specialists - not like a gamble of chance!