DocsGuides

Coding agents

A project made with nx new carries the rules a coding agent needs to write Nexy code your way: the order of work, the code and writing rules, and one skill per topic. This page says what you get, where it goes, and how to keep it current.

What `nx new` adds

File	Read by	Holds
`AGENTS.md`	Codex, OpenCode	the stack of the project, the order of work, the rules, the list of skills
`CLAUDE.md`	Claude Code	one line, `@AGENTS.md`, so both read the same text
`.agents/skills/`	Codex, OpenCode	the skills
`.claude/skills/`	Claude Code, OpenCode	the same skills

AGENTS.md is short and tailored. A client-routed Vue project reads "client-routed, Vue, styled with Tailwind"; an API project is told there are no views. Only the skills that fit are copied: a React project gets nexy-react and no Vue skill, a server project gets the .nexy and Markdown skills, an API project gets no skill about views.

To skip all of it, create the project with nx new --no-agents.

The skills

Skill	Covers
`nexy-principles`	order of work, code rules, writing rules
`nexy-ux-audit`	the experience and information-architecture audit to run before any view
`nexy-project`	modes, folders, file conventions, configuration, commands
`nexy-file`	`.nexy` pages, props, hooks, islands
`nexy-markdown`	`.md` and `.mdx` content
`nexy-react`, `nexy-preact`, `nexy-solid`, `nexy-vue`, `nexy-svelte`	idiomatic code for the client framework
`nexy-styling`	Tailwind, UnoCSS or plain CSS, and the design tokens
`nexy-backend`	API handlers, actions, WebSocket, ISR, middleware
`nexy-modular`	controllers, modules, injection and guards of the modular router
`nexy-security`	dependencies, input, sessions, authorization, secrets
`nexy-testing`	the test-first loop and the gates

The rules they enforce: audit the experience before building a view, write the failing test first, one function does one treatment, one class carries one intention, one folder holds one set of related intentions, no emoji, real SVG icons, plain text written for people, and no hand-written DOM scripts in React, Vue, Svelte, Solid or Preact.

Add them to an existing project

nx skills

Run it at the root of the project. It reads nexyconfig.py and package.json to find the stack, copies the matching skills, and writes AGENTS.md and CLAUDE.md when they are missing.

It never overwrites a file you wrote: an existing AGENTS.md or CLAUDE.md is kept, and the command tells you so. Add the skill list to your own file by hand if you want it there. A skill of your own in .claude/skills/ or .agents/skills/ is left alone; only the nexy-* skills are refreshed.

Run nx skills again after you upgrade Nexy to refresh the copies, and commit the result so the whole team shares it.

How an agent finds them

AGENTS.md (or CLAUDE.md, which imports it) is read at the start of every session. It names the skills.
Each skill starts with a name and a description that say what it covers and when to load it. The agent reads those first and opens a skill only when the task matches.

The skills follow the open Agent Skills format, so other agents that read SKILL.md folders can use them too.

What they do not do

Skills tell an agent how to work; they do not make its output correct. Keep your own checks:

Read what the agent changed.
Run nx check, nx test and nx build. The skills ask for this and for the output to be reported, not summarised.
Audit dependencies the agent adds (nexy-security explains how).

The examples in the skills are checked by Nexy's own tests: the imports exist, the configuration fields exist, the commands exist, the Python parses. That catches drift, not misjudgement.

Next steps

For AI agents: invariants of a Nexy project
CLI: nx new and nx skills
Security: the version floors the security skill points to

Edit on GitHub