Guide · Agent skill
How to use the Think agent skill: plan before you build
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Think skill from tw93/Waza makes an AI agent stop and plan before it writes code. You describe a feature, an architecture question or a problem, and the agent returns a plan for your approval instead of starting to edit files. This guide explains what the skill does, how to install it, how to trigger it, and where it is not the right tool.
Think: Design Before You Buildtw93/Waza · MIT Download the skill Quick start
- 1Download the ZIP. Get the Think skill ZIP from this page.
- 2Unzip the skill. Unzip the archive so the skill folder contains SKILL.md and its references folder.
- 3Place the skill folder. Move it to ~/.claude/skills/ for all projects, or to a project's .claude/skills/ folder for one project.
- 4Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 5Describe the problem. Ask for a plan in plain language, for example 'plan how to add this feature, no code yet'.
- 6Review and approve. Read the recommended approach, the stated assumptions and the handoff steps, then approve or ask for changes.
What the Think skill makes the agent do
Think turns a rough idea into an approved, decision-complete plan. Under this skill the agent writes no code, scaffolding or pseudo-code until you approve the plan. The plan covers the goal, success criteria, constraints, the chosen approach, rejected alternatives, tests and handoff steps.
Before recommending a design, the agent checks existing project docs, reads the actual config values in your project, and looks at official or framework solutions. It is instructed to give one clear opinion rather than a list of vague options, and to say what evidence would change that opinion.
- Evaluation mode: for keep-or-remove or value judgments about a single feature or target.
- Triage mode: for judging a batch of requests, classifying each as a bug, already working, accepted improvement, cosmetic or out of scope.
- Lightweight mode: for a defined problem where only the fix is open; it gives one recommended fix and waits for approval.
Install the Think skill
Download the ZIP for the skill from this page and unzip it. To make it available across all your projects, place the unzipped skill folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.
If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
Trigger the skill with natural requests
You do not need a special command. Describe the problem in plain language and ask for a plan, a design or a judgment. The skill's trigger phrases include requests such as 'plan this', 'how should I', 'what's the best approach', 'is it worth it' and 'should we keep this'. Its description says to use it when planning architecture, judging whether to build, or writing a handoff.
Ask for the outcome you want. For example, say that you want a plan you can hand to a teammate, or that you want to know whether an idea is worth building before anyone writes code.
- 'Plan how we should add offline sync to the notes app. I don't want code yet.'
- 'Is it worth building an export-to-PDF feature, or should we skip it?'
- 'Give me a handoff plan for moving sessions from cookies to tokens.'
- 'Should we keep this legacy import script? Evaluate it.'
What you get back
The output is one recommended direction or a handoff plan, with assumptions and verification steps. The agent states the most fragile assumption behind the plan and says what happens if that assumption fails. When a plan conflicts with a hard rule in your project's guide files (AGENTS.md or CLAUDE.md), the agent is meant to point out the conflict rather than silently override it.
The agent also checks the plan against a few validation points: whether it touches more than 8 files or adds a new service, whether data flows between more than 3 components, which test paths are covered, and whether it can be rolled back without touching data. Every API key or third-party account the plan needs should be listed up front.
- The goal and success criteria
- The chosen approach, with effort, risk and existing code it builds on
- Rejected alternatives and why
- Test paths: happy path, errors and edge cases
- Handoff steps that someone else can execute without re-deciding
Requirements and safety notes
The skill lists no requirements. According to its listing, it does not run scripts and does not use the network. It reads project files and documents as part of planning, and the skill tells the agent not to launch unverified applications, install tools or change state to validate a plan.
Check the MIT licence in the source repository before reuse. Because the skill is read-only planning, the main risk is a plan that looks finished but rests on an unchecked assumption, so review the stated assumptions before approving.
- Requires: nothing listed
- Runs scripts: no
- Uses network: no
- Files: SKILL.md, references/durable-context.md, references/mode-evaluation.md, references/mode-triage.md
Tips for getting good plans
Give the agent real context: the repository, the constraints you already know, and any prior decisions. The skill is designed to check current repo state and project docs before it plans, and to avoid asking you to restate things your project already establishes.
Approve or reject the plan explicitly. The skill is written to wait for approval before any implementation starts, so a clear yes or no keeps the workflow moving.
- Name the constraint that matters most, such as a deadline or a compatibility rule.
- Ask for the minimal option alongside the recommended one if you want to compare.
- Use triage mode when you have several requests from one issue or screenshot set.
What people use it for
- Planning the architecture for a new feature before any code is written.
- Deciding whether a proposed feature is worth building at all.
- Writing a handoff plan that an agent or teammate can execute without re-deciding the design.
- Evaluating whether an existing feature should be kept or removed.
- Triaging a batch of user requests into bugs, existing features, accepted improvements, cosmetic items and out-of-scope items.
Frequently asked questions
How do I install the Think agent skill?
Download the ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ to use it everywhere, or in your project's .claude/skills/ folder to use it in one project. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
Is the Think skill free?
The source repository is licensed under MIT. Check the licence file in the repository for the full terms before reuse. This page does not state any price.
What does the Think skill require?
Its listing says it requires nothing, runs no scripts and uses no network access.
Will Think write code for me?
No. Under this skill the agent writes no code, scaffolding or pseudo-code until you approve the plan. It is for planning and judgment, not for producing finished code or documents.
When should I not use the Think skill?
Do not use it for simple bug fixes or small edits. The skill itself says it is not for those tasks, because the extra planning step adds delay. It is also not the right tool if you only want a finished document or design.
How is Think different from just asking the agent to plan?
Think sets specific rules: wait for approval before coding, check project docs and live config values, give one opinion with evidence that would change it, and list rejected alternatives and tests. Those rules apply each time the skill is triggered.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Think suits developers and teams who want a checked, written plan before code is written, or who need to decide whether an idea is worth building. Skip it for simple bug fixes, small edits, or when you only need a finished deliverable.
Download the skill