think Skill (tw93) for Claude Code: Plan Before Coding

Makes Claude turn a rough idea into an approved plan with a recommended approach, risks and tests, and write no code until you sign off.

Installs: 15.1K; on skills.sh
Stars: 7.5K; on GitHub
Files: 3; besides SKILL.md
Updated: Nov 2026; last repo commit

think makes Claude plan first. You describe an idea, and it reads your project guide (AGENTS.md or CLAUDE.md), checks official solutions in current docs, recommends one approach with a minimal option, names the most fragile assumption and waits for your approval. It writes no code, scaffolding or pseudo-code until you approve.

It has a lightweight mode for a defined problem, an evaluation mode for keep, kill or pivot calls, and a triage mode for a bundle of asks. It is not for bug fixes or small edits: a bug report gets routed to the hunt skill. It starts its first line with a 🥷 emoji, as the SKILL.md instructs.

Good for

Developers who want a plan agreed before the agent writes code
Anyone deciding whether a feature is worth building or keeping
Teams who need a handoff another engineer can run without re-deciding

Install think

Claude Code

Copy this prompt.

Install the think skill in Claude Code. Run this command:

npx skills add tw93/Waza -a claude-code codex cursor -g -y

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

The README route installs all eight Waza skills into ~/.agents/skills (one shared copy). think works on its own, so the single-skill routes are enough. Pick one route, not several. The plugin routes namespace the skill, for example /waza:think. The marketplace file says the single-skill plugin needs Claude Code v2.1.143 or newer.

From the project README. Source.

What's in the skill

When it loads

Turns rough ideas into approved, decision-complete plans before coding. Use when planning architecture, judging whether to build, or writing a handoff. Not for bug fixes or small edits.

Sections in SKILL.md

Outcome Contract
Durable Context Preflight
Lightweight Mode
Evaluation Mode
Triage Mode
Before Reading Any Code
Check for Official Solutions First
Propose Approaches
Validate Before Handing Off
Simplicity Gate
Implementation Handoff
Hard Rules
Gotchas
Output

Key facts

Author: tw93 (repo owner)
Repo: tw93/waza
Installs on skills.sh: 15,123
Repo stars: 7,488
License: MIT
Last repo commit: 2026-11-01
Files in the skill folder: 4 files including SKILL.md
Part of: Waza, a set of 8 skills by tw93 (/check, /think, ...)

What it includes

One recommended fix in 2 to 3 sentences, then wait for approval
For a defined problem where only the how is open. Names the brute-force version first, lists files, flags more than 8 and states one risk.
Lightweight Mode
A Kill, Keep or Pivot verdict with three reasons
For value, viability or commercialization calls on one target. Loads references/mode-evaluation.md and ends with an entity delta count.
Evaluation Mode
A per-item table for a bundle of asks or screenshots
Loads references/mode-triage.md so each ask is accepted or rejected on its own.
Triage Mode
Check prior decisions and live config values first
Skims ADRs and design docs, and opens the real config file instead of quoting a default from memory.
Before Reading Any Code
Prefer built-in or ecosystem solutions
Checks framework built-ins against live docs, and for hard problems studies 2 to 3 mature open-source projects.
Check for Official Solutions First
One recommendation, one minimal option, the most fragile assumption
Includes attack angles for dependency failure, scale and rollback cost when the plan needs them.
Propose Approaches
Checks on file count, components, tests, rollback and credentials
Uses read-only checks for tool availability and does not install tools or change state to validate a plan.
Validate Before Handing Off
Shrink plans that add files, layers or config nobody asked for
Compares against the brute-force path and counts new commands, env vars, flags and services.
Simplicity Gate
A plan another engineer or agent can execute
Scope, approach, interface changes, verification commands, follow-through steps and rollback.
Implementation Handoff
No placeholders, phases that merge on their own, bugs routed out
Forbids TBD and TODO in approved plans and sends error reports to /hunt.
Hard Rules
Approved design summary
Building, not building, approach, 3 to 5 key decisions and deferred unknowns.
Output

From the official docs

Permissions

Low risk

The skill plans and reads, and the SKILL.md says it writes no code until you approve. It reads your project guide files and can look up live docs, so keep secrets out of those files.

What it can do

Read your data
Create and change things
Delete things
Read-only modeHow to turn it on

Watch out for

It reads your project guides and memory notes
It reads AGENTS.md or CLAUDE.md, the matching rules, and a local agent-memory summary if you point to one. The durable-context reference also says it may read an obvious local MEMORY.md and open one or two summaries. Do not keep notes there that you would not show the assistant. github.com

Technical details

Turning on read-only	SKILL.md says no code, scaffolding or pseudo-code until you approve, and plan validation uses read-only checks.
Talks to	Live documentation lookups, when your environment has doc-lookup tools

Questions

What does the think skill do?

It makes Claude design before it builds. It produces an approved plan with the goal, chosen approach, rejected alternatives, tests and handoff steps, and waits for your approval before any implementation.

Why does Claude start its reply with a ninja emoji?

The SKILL.md tells it to prefix the first line with 🥷 inline. This is part of the Waza skill set.

When should I use think and when hunt?

Use think for a new feature, an architecture choice or a value judgment. Use hunt for an error, crash or regression. The think skill says a bug report belongs to /hunt and routes you there.

Will it start coding after I approve?

Only if you ask. If you say to implement the plan or otherwise authorize it, it carries out the approved direction without another round, but a settled design alone does not authorize implementation.

How do I install it?

Use the install box above. The Waza README documents the skills CLI for several agents and a plugin route, in which skills are namespaced like /waza:think.

Is it safe to install?

It bundles no scripts, only markdown. It plans and reads, and the SKILL.md says no code is written before you approve. It does read your AGENTS.md or CLAUDE.md and can use doc-lookup tools.

Information

updated Nov 2, 2026

Repository: tw93/waza
Folder: skills/think

tw93/waza skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 7

Waza README · github.com
Waza marketplace.json · github.com
think SKILL.md · github.com
think references/mode-evaluation.md · github.com
think references/mode-triage.md · github.com
think references/durable-context.md · github.com
Waza README · github.com