tw93/Waza

Waza – AI‑driven engineering‑habit skills

What it is
Waza is a small collection of skills that can be installed into LLM‑powered coding assistants such as Claude Code, Codex, Cursor, Gemini‑CLI, Copilot, etc. Each skill encodes a concrete engineering habit (planning, UI design, debugging, code review, research, summarising, writing, health‑checking) and can be invoked with a slash command like /think or /check. The skill runs, produces a structured output, and then hands control back to the user or to the next skill in a chain.

Why it matters
LLM agents are great at generating code, but without disciplined workflows they tend to produce noisy, generic results. Waza supplies the process layer: a clear trigger, a well‑defined outcome, and guardrails that keep the model from drifting. By turning familiar engineering habits into reusable, model‑agnostic plugins, Waza lets you get the same disciplined workflow whether you are using Claude, Codex, or any other LLM‑based IDE.

Key features

Skill	When you use it	What it does
/think	Before starting a new feature	Challenges the problem, pressure‑tests the design and returns a complete implementation plan.
/ui	Building front‑ends	Generates UI mock‑ups and iterates on screenshots, giving a concrete visual direction.
/check	After a change, before merge	Reviews the diff against project constraints, verifies the result and can trigger a release.
/hunt	When a bug appears	Systematic debugging workflow that confirms the root cause before a fix is applied.
/write	Writing prose (Chinese/English)	Rewrites text to sound natural, removing stiff phrasing.
/learn	Exploring an unfamiliar domain	Six‑phase research pipeline: collect → digest → outline → fill → refine → self‑review.
/read	Any URL or PDF	Returns a concise summary or clean Markdown for quoting, citation, etc.
/health	Auditing the agent itself	Checks the health of the LLM environment, budget usage and maintainability.

Installation & usage

• Install with a single npx command; the skills are placed in ~/.agents/skills and become available to any compatible agent.
• For GUI‑based tools (Claude Desktop, Codex plugin marketplace, etc.) you can upload a ZIP release.
• Skills can be chained manually (/think → /check) or combined with optional rules that add persistent instruction guards (e.g., English‑error correction, anti‑pattern checks, routing hints).

Status line & budget awareness
Waza ships a tiny status‑line script for Claude Code that shows the model’s context window usage and the 5‑hour / 7‑day quota, helping you stay within limits. Codex has native support for similar indicators.

Scope & limits
The skills are generic – they do not contain any project‑specific code. At runtime, /check reads the public parts of the target repository (README, package.json, Makefile, CI config) to become project‑aware, but it never accesses private credentials.

Relation to other projects
Waza is part of a trilogy by the author: Kaku (code generation), Waza (habit‑drilling skills), and Kami (document shipping). It focuses purely on the workflow layer rather than raw code synthesis.

Who might use it

• Developers who already use Claude, Codex, Cursor, Gemini‑CLI, etc. and want reproducible, disciplined AI assistance.
• Teams that want to standardise AI‑augmented engineering habits without building a custom plugin framework.
• Anyone looking for a lightweight, eight‑skill set that can be extended with custom rules.

License
MIT – free to use, modify and contribute.

---

All information above is taken directly from the repository’s README.