waza

Waza is a collection of eight engineering-habit skills for AI coding agents. The premise is that good engineers do more than write code: they pressure-test requirements, debug to root cause, review their own diffs, and read primary sources. Waza packages those habits as skills an agent can run, each with one job and a clear trigger. ## The eight skills - /think — used before building anything new. Challenges the problem, pressure-tests the design, and produces a decision-complete plan another agent can implement. - /ui — used when building frontend interfaces. Aims for distinctive UI with a committed direction, including screenshot-driven aesthetic iteration. - /check — used after a task and before merging or release. Reviews the diff against project constraints, verifies the result, and handles approved release and maintainer actions. - /hunt — used for any bug, regression, or unexpected behavior. Systematic debugging that confirms root cause before applying a fix. - /write — used when writing or editing prose. Rewrites prose to sound natural in Chinese and English and cuts stiff, formulaic phrasing. - /learn — used when diving into an unfamiliar domain. A six-phase research workflow: collect, digest, outline, fill in, refine, then self-review and publish. - /read — used for any URL or PDF. Returns a concise summary or clean Markdown for conversion, quotes, citations, saving, or further research. - /health — audits agent health: Codex, Claude Code, project instructions, verifier output, and AI maintainability, with a budget-aware summary pass before deep inspection. Each skill is a folder containing reference docs, helper scripts, and notes on gotchas drawn from real failures. ## Installation For Claude Code, Codex, Cursor and other agents, a single npx command installs the skills into the shared `~/.agents/skills` directory, with Claude Code symlinked in. Agents that read that directory pick the skills up as slash commands; agents with a private skills directory are specified after the `-a` flag. Updates use `npx skills update -g -y`. Host plugins are also supported for Claude Code and Codex, where skills are namespaced (for example `/waza:check`). Claude Desktop users can download a ZIP from releases and upload it under Customize > Skills. A Pi package is available as well. ## Chaining and context Users decide how skills chain. Documented workflows include planning a feature (/think then /check), shipping a fix (/hunt, fix, /check), research and writing (/read, /learn, /write), and debug-and-verify (/hunt then /check). Waza ships only generic habits. The /check skill becomes project-aware at runtime by reading the target repository's public context such as READMEs, package manifests, Makefiles and CI workflows, plus task constraints — not private paths, credentials or tokens. ## Extras A minimal statusline for Claude Code shows context window, 5-hour quota and 7-day quota, color-coded by usage without progress bars. Codex has native statusline items configured in `~/.codex/config.toml`. Three optional rule toggles can be installed: English coaching that appends a short correction when a prompt has an English mistake, always-on anti-pattern guardrails, and a routing hint telling non-Claude hosts to prefer Waza skills when a request matches their triggers. ## Background Waza (技) is a Japanese martial arts term for technique. The project positions itself as deliberately small compared with heavier skill collections, and says the skills were built from real projects and refined across 300+ sessions on 7 projects. It is described as part of a trilogy alongside Kaku (code) and Kami (documents). Licensed under MIT.