Built for AI (@cruzjs/skills)

CruzJS ships an AI toolkit — @cruzjs/skills — that teaches AI coding assistants how the framework works. Instead of re-explaining your architecture every session, you install a canonical knowledge base, a suite of slash commands, and a set of agent personas once, and your assistant follows CruzJS patterns out of the box.

It is multi-harness: the same content is distributed into the native config format of five different AI coding tools.

Install

npx @cruzjs/skills init

Interactive — pick which harnesses to set up. Non-interactive:

npx @cruzjs/skills init --harness=cursor
npx @cruzjs/skills update          # refresh to the latest content

The canonical knowledge base

Everything starts from a single source of truth installed at .cruzjs/knowledgebase/ — harness-agnostic markdown. From there, init distributes the right subset into each tool’s expected location and format, so all your assistants share one consistent picture of the project.

What gets installed:

34 knowledge base docs — architecture, DI, Drizzle/D1, tRPC routers, auth & org scoping, data ownership, UI patterns, events, jobs, testing, deployment, and more.
13 slash commands — repeatable workflows (see below).
8 agent personas — architect, developer, developer-fix, code-reviewer, qa-engineer, auto-monitor, ui-designer, ux-designer.
Project instructions — the entry file each harness reads automatically (CLAUDE.md / AGENTS.md / GEMINI.md).

Supported harnesses

Tool	Entry file	Where content lands
Claude Code	`CLAUDE.md`	`.claude/` — `commands/`, `kb/`, `agents/personas/`, `agents/shared/`, `agents/workflows/`
Cursor	—	`.cursor/rules/*.mdc` (MDC frontmatter; domain rules auto-attach via globs)
Codex (OpenAI)	`AGENTS.md`	`.agents/skills/`, `.agents/kb/`
OpenCode	`AGENTS.md`	`.opencode/skills/`, `.opencode/kb/`, `opencode.json`
Antigravity (Google)	`GEMINI.md`	`.agent/rules/`, `.agent/skills/`

For Cursor, domain-specific rules attach automatically by file glob — e.g. the events KB attaches when you edit files under **/events/** — so the assistant pulls the right context without being told.

Commands

Command	What it does
`/dev`	Full autonomous dev pipeline (spec → build → review → QA → PR)
`/new-feature`	Scaffold a complete feature module (service, router, UI, tests)
`/add`	Add a field, event, test, or job to an existing feature
`/debug`	Diagnose and fix an issue
`/fix-lint`	Fix TypeScript and lint errors
`/code-review`	Automated review (security, patterns, data ownership)
`/qa`	Automated QA with Playwright
`/create-ui-component`	Build a production UI component with Storybook
`/build-application`	Interactive wizard to build a complete app
`/pm`	Turn a feature request into a product spec
`/architect`	Turn a spec into a detailed implementation plan
`/new-ui`	Create UI (route / component / modal) for an existing backend
`/roadmap`	Execute tasks from a `MASTER_PLAN.md`

Why it matters

Because the assistant already knows the framework’s conventions, it produces code that fits the rest of your app — correct DI wiring, the right tRPC procedure type, tenant-safe queries that filter by userId/orgId, CUID keys, D1-compatible schemas — without you having to spell them out each time. The same guarantees hold whether your team is on Claude Code, Cursor, Codex, OpenCode, or Antigravity.