Install Eight Engineering-Habit Skills with Waza

An agent that jumps straight to code produces plans nobody checked and fixes that treat symptoms. Waza — a 7k-star repo from tw93 — turns eight engineering habits into skills, so the pressure-testing, root-cause hunting, and self-review happen before and after the code, not never. Each skill states the outcome, the red lines, and how the result gets verified, then steps back and lets the model choose the path.

Why This Skill Matters

Waza (技, a Japanese martial arts term for a technique practiced until instinct) keeps the skill count small on purpose: eight habits, each with one job and a clear trigger. The README says the skills were built from real projects and refined through 300+ sessions across 7 projects, with every gotcha tracing to a real failure. The repo positions itself against heavier frameworks — Superpowers and gstack are named as "powerful but heavy" — and bets that restraint pays compound interest as models improve.

The eight skills, from the README's own table:

• /think — challenges the problem and produces a decision-complete plan before you build anything new.
• /ui — produces distinctive frontend UI with screenshot-driven aesthetic iteration.
• /check — reviews the diff against project constraints after a task, before merging or release.
• /hunt — systematic debugging; the root cause is confirmed before any fix is applied.
• /write — rewrites prose to sound natural in Chinese and English, cutting stiff, formulaic phrasing.
• /learn — a six-phase research workflow for unfamiliar domains.
• /read — turns any URL or PDF into a concise summary or clean Markdown.
• /health — audits your agent setup (Codex, Claude Code, project instructions, verifier output).

Installation

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

One copy lands in ~/.agents/skills, the shared skills directory. Claude Code is symlinked in; Codex, Cursor, Gemini CLI, Copilot, Amp, Kimi Code CLI, and other agents that read that directory pick the eight skills up as /check, /think, and so on. Agents with a private skills directory take their id after -a. Update with npx skills update -g -y, and remove with npx skills remove tw93/Waza -g.

Host plugin installs are also documented — in Claude Code the skills are namespaced as /waza:check:

# Claude Code (update: claude plugin update waza)
/plugin marketplace add tw93/Waza
/plugin install waza@waza

# Codex (update: codex plugin marketplace upgrade waza, then codex plugin add waza@waza)
codex plugin marketplace add tw93/Waza
codex plugin add waza@waza

Claude Desktop takes the waza.zip release asset via Customize > Skills, and Pi users can run pi install npm:@tw93/waza.

Real Workflow: Fix a Regression Without Guessing

1. Describe the bug and invoke the debugging skill:

/hunt 用户反馈登录页在 Chrome 125 之后间歇性白屏，控制台有一个 undefined 报错。
先定位根因，确认之前不要改任何代码。

2. /hunt drives systematic debugging — reproduce, narrow, and confirm the root cause before any fix lands. The README scopes this skill to "any bug, regression, or unexpected behavior", with special attention when something used to work.
3. Apply the fix, then run /check to review the diff against project constraints. The skill becomes project-aware at runtime by reading the repository's public context — READMEs, package manifests, Makefiles, CI workflows — and never private paths, credentials, or tokens.
4. Let /check handle the follow-through you approve: the README lists release, publish, push, and issue follow-through as actions it handles when needed.

Skills chain the same way for other work: /think to settle a plan, implement after approval, /check before merging; or /read to fetch sources, /learn to synthesize, /write to polish. Each skill stops at the requested outcome — an explicitly authorized workflow continues without asking you to approve every transition.

Tips

• The optional rules (English coaching, anti-patterns, waza-routing, clarity) are separate from the skills — installing Waza alone does not enable them; you opt in per rule with the setup script.
• A minimal Claude Code statusline ships alongside: context window, 5-hour quota, and 7-day quota, color-coded by usage.
• Codex has native statusline items; the README shows the [tui] config block for it.
• When /hunt confirms a root cause, record it — the gotchas in each skill folder come from real failures, and yours can feed your own notes the same way.

When Not to Use This

Eight habits is a deliberate ceiling: domain-specific playbooks (say, scientific computing or mobile release engineering) need their own skills, and teams standardized on a heavier framework will find Waza thin — that is the design, not a gap. The optional rules change your agent's persistent instructions, so read each rule file before installing it.

See the leaderboard for more skills.