Skills

Skills give AI agents durable Elements context for authoring, validating, and maintaining UI projects

Elements ships agent skills with the CLI and MCP server. Skills are not a replacement for deterministic tools. They provide workflow order, project policy, and authoring guidance while the CLI and MCP tools provide the current API data, examples, imports, validation, packages, and starter setup.

The Model Context Protocol standardizes tools, prompts, and resources. Elements maps that model directly: tools expose callable operations such as api_get, prompts provide user-invoked task flows, and skills provide reusable context that agents can keep loaded while working in a project.

Install CLI

Install the Elements CLI to your system. This creates a canonical executable path and adds the nve command to your path when possible.

Add Skills to an Existing Project

The recommended path is the project setup command. Run it from the project root:

The setup command configures Elements for common agent clients and editor tooling:

• Adds Elements MCP configuration for Claude Code, Cursor, and Codex.
• Writes the Elements skill to .agents/skills/elements/SKILL.md.
• Writes the Elements skill to .claude/skills/elements/SKILL.md.
• Adds VS Code custom data paths for nve-* tag and attribute authoring.
• Adds or updates core Elements package dependencies.

New starter projects created with nve project.create receive the same agent setup during project creation.

Install Skills Only

Elements publishes the elements skill through an Agent Skills well-known endpoint. The hosted file comes from the same registry as nve skills.list. Inspect or install it with the open skills CLI:

Target Codex and Claude Code explicitly when you don't want automatic agent detection:

This route installs skill files only. It does not install the Elements CLI, configure the MCP server, add editor data, or add package dependencies. Continue to use CLI or MCP API lookup and template validation for deterministic, current project data.

Manual Skill Setup

Use the CLI when you need to inspect or install a skill by hand:

Place the selected skill content in the directory format supported by your agent. The generated Elements skill uses this file shape:

For Codex and Cursor-compatible agents, use:

For Claude Code, use:

Available Skills

The default nve skills.list command and MCP skills_list tool expose these bundled skills.

The playground skill is available when the CLI, MCP server, or hosted build has the Elements playground service enabled.

Use Skills With MCP

Configure the MCP server once:

Then use skills as context and MCP tools as the live data plane:

• Use skills_get for workflow guidance.
• Use api_get, api_list, and api_template_validate for current component contracts.
• Use examples_list and examples_get for known UI patterns.
• Use api_imports_get to generate explicit define.js imports.
• Use project_setup and project_validate for project health.

Dynamic Context Lookup

Elements publishes context files for agents that can fetch URLs at runtime if MCP or CLI are not available:

• llms.txt is the small context index.
• llms-full.txt is the large single-file archive.

Use llms.txt when an agent can fetch links during a task. It points to the CLI/MCP context, lint context, API index, examples index, skills index, icons, and design tokens. This keeps context small and lets the agent load only the specific page or API it needs.

Use llms-full.txt when you need offline context or local retrieval-augmented generation. Download it, split it into chunks, and generate embeddings for a local index. This works well for editors, internal assistants, and environments without outbound network access during inference.

For most projects, use all four layers narrowly: install the local elements skill, configure nve mcp, let web-enabled agents start from llms.txt, and reserve llms-full.txt for local search indexes.

References

• Elements CLI
• Elements MCP
• MCP Tools
• MCP Prompts
• MCP Resources