Which agents support agent skills

The Agent Skills standard is supported by:

• OpenAI Codex — reads from ~/.codex/skills/
• VS Code with compatible extensions — reads from .agents/skills/ in your project
• Any other agent that supports the .agents/skills/ project-local directory convention

If you are using Claude Code, Cursor, Cline, Roo Code, or Windsurf, use the MCP server instead. It is a better fit for those tools and does the same job.

Installing the skill

The skill is a folder called lightningchart-js. Get it from GitHub and copy it to the right location for your agent.

For project-local use (any agent skills compatible tool):

git clone https://github.com/Lightning-Chart/lightningchart-js-agent-skill.git
cp -r lightningchart-js-agent-skill/lightningchart-js .agents/skills/

For OpenAI Codex (global install, available across all projects):

# macOS / Linux
cp -r lightningchart-js-agent-skill/lightningchart-js ~/.codex/skills/

# Windows PowerShell
Copy-Item -Recurse .\lightningchart-js $env:USERPROFILE\.codex\skills\

That is the entire installation. No npm install, no config file edits, no restart required. The skill directory is just a folder the agent reads.

Using it in prompts

Many agents pick up locally installed skills automatically when the task is relevant. Try this in a JavaScript or TypeScript project after installing the skill:

If the agent picks up the skill, you will see it use correct LightningChart JS API patterns without being told to. If it does not pick it up automatically, reference the skill explicitly:

The $lightningchart-js prefix tells the agent explicitly which skill to load for this task. Using it explicitly is the most reliable approach if your agent does not auto-discover skills.

What the skill teaches the agent

Without any special guidance, AI coding agents writing LightningChart JS code tend to make several consistent mistakes:

• Using @arction/lcjs (the old package name) instead of @lightningchart/lcjs
• Mixing in Chart.js or Recharts method signatures that do not exist in LCJS
• Using API patterns from old LCJS versions that have been updated
• Guessing method names rather than checking documentation
• Missing the license key setup step

The skill gives the agent a reliable starting point that prevents all of these. It loads official LCJS agent instructions maintained by the LightningChart JS maintainers and updated when the API changes, so the guidance stays current.

The skill vs the MCP server: which one to use

They do the same job through different mechanisms. The right choice depends on your tool:

• Use the MCP server for Claude Code, Cursor, Cline, Roo Code, Windsurf, and any MCP-compatible agent. It integrates more deeply and the agent can call it dynamically during a session.
• Use the agent skill for OpenAI Codex and VS Code-compatible tools that support the skills directory convention.

If your tool supports both, the MCP server is generally the better option because it can be called on demand rather than being loaded once at the start of a session.