ysskrishna logoysskrishna
Claude Code Subagents for Cursor logo

Claude Code Subagents for Cursor

Your Cursor subagents can finally run on the model you picked in each agent file.

Get it
Cursor Directoryskills.shGitHub
Built with
AI Apps & Assistants
Chat & AI
Agent Skills
Bash
Details
FreeMIT licenseCreated October 31, 2026Updated November 12, 2026
Share this product

You tuned an agent for Haiku. Cursor still billed you like Opus.

You wrote a scout agent with a cheap model in its frontmatter. You spawned it from Cursor and watched the meter climb anyway. With a personal Anthropic API key, Cursor staff confirm the Task tool keeps subagents on the parent chat model. If the model is not enabled in Cursor, the same thing happens.

That is not a prompt problem. It is routing.

Claude Code Subagents for Cursor is a skill that sends your custom subagents through Claude Code headlessly, so each run respects the model in that agent file. Every reply tells you which model actually ran.

Pay for the model you chose, not the parent chat.

What changes when you turn it on

  • Honest model routing. Custom agents in your Cursor, Claude, or Codex agents folders run on the model in their file, not the chat you are in.
  • A clear receipt. Each subagent answer ends with a line naming the model that ran, so you can spot a mismatch before the bill lands.
  • One toggle for the session. Invoke the skill once, stay in Claude Code subagent mode until you ask to switch back to Cursor subagents.
  • Install the way you already work. Cursor plugin from GitHub, skills.sh, GitHub CLI, or copy the skill folder by hand. Same skill either way.
  • Docs for the sharp edges. Permissions when Claude Code leaves Cursor's sandbox, troubleshooting, and a sample project with two ready-made agents.

Picture the next expensive agent call

You ask for a quick scout pass on a big repo. The agent file says Haiku. The skill routes the work through Claude Code, the job finishes, and the footer says the model you expected. You stop second-guessing every subagent spawn.

Ready to try it?

You need Cursor desktop locally, Claude Code installed and signed in, and a POSIX shell on macOS or Linux. Cloud agents cannot see the Claude CLI.

Keep exploring

Frequently asked questions

Why not just trust Cursor's Task tool?

You can, when the parent model is the one you want and your setup does not force BYOK routing. The README documents the two cases where Task ignores the agent file model: a personal Anthropic API key, or a model that is not enabled in Cursor. This skill exists for when you already wrote per-agent models and want those honored.

How do I run Cursor subagents on the model in each agent file?

Install the skill from the GitHub repo, invoke it in chat, then ask for a custom subagent as usual. For the rest of the session, the skill routes those agents through Claude Code and reports the model each one used.

What do I need before it works?

Cursor desktop on your machine, Claude Code installed and signed in, and macOS or Linux. Runs bill your Claude Code account. Cloud Agents are not supported because they do not have the Claude CLI.

Will Claude Code ask for approval on every command?

Claude Code runs outside Cursor's sandbox. The project docs walk through Cursor approval modes that balance safety and fewer prompts, including Auto-Review with Sandbox and Run Everything for trusted repos.

How is this different from AI Agent Skills?

AI Agent Skills adds thinking workflows to your assistant. This skill fixes model routing when you spawn custom subagents from Cursor. They complement each other; many people use both.

Can I switch back to Cursor's built-in subagents?

Yes. Tell the assistant to use Cursor subagents and it routes through the Task tool again.

Is it free?

Yes. It is open source under the MIT license.

Continue exploring

Explore All Products
ProductThinking Methods

Thinking Methods

17 well-known thinking methods your AI coding assistant can actually run.
Agent Skills
AI Apps & Assistants
Bash
Gemini
BlogCover image for Your AI Agent Shouldn't Start From Zero Every Session

Your AI Agent Shouldn't Start From Zero Every Session

AI Apps & Assistants
Agent Skills
3 min read
ToolCover image for AI Token Counter

AI Token Counter

AI token counter with model pricing and context-window comparison in your browser.

Chat & AI
AI Apps & Assistants
+1
BlogCover image for Teach AI Agents to Think From Multiple Angles with Six Thinking Hats Skill

Teach AI Agents to Think From Multiple Angles with Six Thinking Hats Skill

AI Apps & Assistants
Agent Skills
6 min read
ProductAI PPT Slide Generator

AI PPT Slide Generator

A topic goes in. A downloadable PowerPoint file comes out. Host the API yourself.
AI Apps & Assistants
Gemini
FastAPI
PostgreSQL
BlogCover image for Thinking Skills for AI Agents: From Brainstorming to Critical Analysis

Thinking Skills for AI Agents: From Brainstorming to Critical Analysis

AI Apps & Assistants
Agent Skills
3 min read