Install
npx skills add tw93/Waza -a claude-code codex cursor -g -yWho it is for
If you already plan features and fix defects and only want the agent to run the same habits instead of improvising, this skill fits. Officially: turn the engineering habits you already know into skills the agent can execute.
After install you get a set of slash skills. The page names /think, /check, /hunt. In Claude Code you type the slash commands directly. In Codex you call them by the installed skill name and follow the same steps. The page also mentions Cursor, Gemini CLI, Copilot, Amp, Kimi Code CLI, and other agents that read the same shared directory can pick up these eight skills. That is a directory-read relationship, not an exclusive set of names.
A better fit is someone who has a target repository they can open and wants a read-only plan or check first, instead of telling the agent to implement immediately. Official notes:
https://github.com/tw93/waza
First try in 3 minutes
Use only this official install path:
npx skills add tw93/Waza -a claude-code codex cursor -g -y
The command drops a copy into the shared skill directory ~/.agents/skills. Claude Code attaches via symlink. Codex, Cursor, and other agents that read that directory then recognize the skills as names like /check and /think. The page does not give another installer as the main path.
Prerequisites: this machine can already run the npx command above, and you have opened at least one agent that reads ~/.agents/skills. The install location is this shared directory. Do not copy it by hand into another folder.
For the first task, pick a reversible, analysis-leaning action. After opening the target repository in Claude Code, type /check, or type /think first for a plan only. In Codex, call the same check or plan by skill name. Do not say implement on the first try. Success signal: the skills appear under names like /check and /think, and /check can give a check result from the repository's public context instead of saying the skill is not installed.
Usage examples
Example 1: plan the feature first, then decide whether to implement.
Officially, feature planning is written as: /think, approve the plan, then say implement X, then /check, and only then merge. The first time, stop at /think and read the plan. The evidence does not write what an approve button looks like, and it does not write a concrete merge command. What you can confirm is this order, and that implementation should happen after the plan is accepted.
Example 2: find the root cause first, then change code.
Officially, fixing a defect is written as: /hunt, then fix, then /check, then follow-through. The page writes follow-through as release/publish/push/issue follow-through, but it does not expand those follow-through steps. The first time, run only /hunt. After the root-cause description matches the symptom, then change code. /hunt is written as confirming the root cause, not starting the change.
Both examples use the same install path. /check can follow a plan or a fix, but the first time you should still run the check alone and confirm it read public context.
Limits
Waza only supplies general engineering habits. /check reads the target repository's public context at runtime, including README, package manifests, Makefile, continuous-integration workflows, and the constraints of this task. It does not read private paths, credentials, or tokens. The page does not write that it can access private configuration or secrets.
The install command names the three targets claude-code, codex, and cursor, and the copy lands in ~/.agents/skills. Other agents that read that directory can pick up the skills, but the page does not give a trigger example for every name. Besides Claude Code slashes and Codex calls by name, the only confirmed fact for other environments is "if they read the shared directory, these eight skills appear".
The page does not write a language scope, a quota, or a required repository type. When public context is missing from the target repository, how far /check can go is not given as a separate failure sample.