mirror of https://github.com/tw93/Waza.git synced 2026-09-14 19:54:25 +08:00

Files

T

Tw93 df08298346 fix: rename design skill to /ui so it stops shadowing Claude Code's /design

Claude Code ships a built-in /design (Claude Design). Waza's design skill
collided on it and silently shadowed the built-in, completely so under the
default -g install. /ui hands the command back: the built-in has no
namespaced fallback while Waza does. Routing keywords keep 设计/design so
plain-language requests still reach the skill.

Also tighten the README (fold the duplicated philosophy into one section,
lead with the skills diagram) and trim that diagram to drop the title and
install line that duplicated the surrounding README.

2026-06-27 11:38:14 +08:00

2.6 KiB

Raw Blame History

name, description

name	description
waza	Dispatcher for Waza engineering skills: think (architecture/handoff), ui (artifact-grounded interface design), check (review/release gates), hunt (runtime debugging/regression), write (prose/release copy), learn (research), read (URL/PDF fetch), health (agent config and AI maintainability audit).

Waza: Engineering Skills Dispatcher

Prefix your first line with 🥷 inline, not as its own paragraph.

Update check (non-blocking). Before routing, run bash scripts/check-update.sh once; if it prints a line, relay it to the user, then continue. It runs at most once a day, only reads a public version file, sends no data, and fails silently.

You have eight skills available. Match the user's intent to the right skill, read the matching section below, and execute it.

Routing Table

How This Works

Read the user's message and match it to a skill from the table above.
Read the matched skill section in full.
Execute that skill's instructions exactly.

If the message could match multiple skills, use these disambiguation rules:

Most specific wins: /ui is more specific than /think for UI decisions.
URL in message: start with /read. If the content is research material, chain to /learn.
Code already done vs. code broken: done/PR -> /check; error/broken -> /hunt.
Config/maintainability vs. code: Codex/Claude misbehaving, hooks/MCP, /health token usage, AI coding code rot, unclear context, missing verification, or stale verifier output -> /health; user code errors -> /hunt.
Release action vs. release prose: commit/tag/publish/push/release reactions/close issue -> /check; write release notes/changelog text -> /write.
Screenshot taste vs. screenshot regression: visual taste complaint -> /ui; broken render/state/generated output or used-to-work evidence -> /hunt.
From scratch vs. editing: new long-form output -> /learn; existing draft to polish -> /write.
"Judge this" + error -> /hunt; "judge this" + should we keep it -> /think.
Still ambiguous: read both skills' "Not for" sections; use exclusion. If still unclear, ask the user.

Path Resolution

In this distribution, sub-skill scripts live at skills/{name}/scripts/. Resolve all relative paths from this file's directory, not from a personal home-directory skill cache.

Chaining

Skills chain manually, not automatically. Each skill completes and waits for the user's next action.

Common chains: /think -> implement approved plan -> /check | /hunt -> fix -> /check -> release/push/issue follow-through | /read -> /learn -> /write

2.6 KiB Raw Blame History