How to Share Skills Between Claude and Cursor (and Copilot)
Some Claude skill instructions translate directly to Cursor and GitHub Copilot. Others need adaptation. Here's what transfers, what needs adjustment, and how to keep the approved reusable guidance in one place.
Updated 6 min read
The instructions inside a Claude skill — the text that tells the assistant what to do — can be adapted for use in Cursor and GitHub Copilot. Neither tool has a "Skills" interface like Claude's, so you adapt the instructions into the format each tool reads: a file in the repository, a project rule, or a custom instruction. The prompt text transfers; the Claude skill packaging does not.
This means reusing skills is a deliberate adaptation step, not a one-click sync. fenbs can hold the approved version of the instructions and a link to where each tool reads them, so the next person to adapt or update the skill starts from the same source.
Where each tool reads reusable instructions
Claude Code reads skill files from .claude/skills/ in a repository (project skills) and from ~/.claude/skills/ on your machine (personal skills). It also reads AGENTS.md natively for project-wide context and rules.
Cursor reads .cursorrules in a project root, .cursor/rules/ for project rules defined in its interface, and AGENTS.md files. If you have standard review or documentation instructions in a Claude skill, the text of those instructions can be placed in a .cursorrules file or a Cursor rule for the same project.
GitHub Copilot reads .github/copilot-instructions.md for repository-level instructions. Organization owners can also set custom instructions at the organization level. A skill's purpose and instructions can be adapted into this file, following Copilot's documentation for the format it expects. Check GitHub's current Copilot documentation for the full scope of what custom instructions support.
What translates directly
Plain-language instructions translate well. A review checklist that says "check every changed function for unhandled error paths" is useful in Claude Code, Cursor, and Copilot. A style guide that says "use British English" or "write functions under 30 lines" applies regardless of which assistant reads it.
Scope restrictions translate well too. "Review only the files supplied" and "do not run, build, or deploy" are instructions any assistant can follow. The specific phrasing may need adjustment for a different model's tendencies, but the intent is clear.
Step-by-step procedures translate if the steps do not depend on Claude-specific tool calls. A three-step research process — "read the supplied document, identify claims that need a source, produce a numbered list of those claims" — works in any assistant that can read files.
What needs adaptation
File references need updating. A Claude Code skill that references .claude/skills/resources/template.md must be rewritten for Cursor to point to a path Cursor can actually read. The file may need to move, or the reference may need to become a relative path from the project root.
Tool-specific calls do not transfer. If a Claude Code skill uses Claude's built-in tools by name — for example, a specific MCP tool or a Claude Code subagent command — those calls are Claude Code-specific. Cursor and Copilot have their own agent and tool models; adapt the intent into the equivalent approach for each.
Output format instructions may need adjustment. Claude Code, Cursor, and Copilot each have preferences in how they present code suggestions, comments, and longer text. Test the same instruction in each tool and compare whether the output format matches your expectation before relying on it.
| Instruction type | Transfers? | Notes |
|---|---|---|
| Plain-language task description | Yes, with minor wording adjustments | Test in each tool; models interpret instructions differently |
| Review or quality checklists | Yes | File references must point to paths the tool can access |
| Style and format rules | Yes | Add to each tool's instruction file; no special conversion needed |
| Claude Code-specific tool calls | No | Rewrite using the equivalent in each tool's own docs |
| MCP tool references | Partial — depends on whether the other tool supports MCP | Check whether Cursor or Copilot supports the MCP server you use |
| Skill file structure (.claude/skills/) | No | Each tool reads its own file paths and formats |
A practical approach to reusing skill instructions
Start with the skill you want to reuse. Strip it down to the plain-language instructions — the checklist items, the constraints, the output format. Remove file references, tool calls, and Claude-specific syntax.
Adapt that core for each tool. For Cursor, create or update .cursorrules or a project rule. For Copilot, add the instructions to .github/copilot-instructions.md. Test with a realistic task in each tool before committing the file. Compare outputs and adjust the wording until the result meets your standard.
Decide which version is canonical. When you improve the skill based on a test in Cursor, update the Claude skill too, and vice versa. Without a single source, the versions drift, and different tools apply different versions of what is supposed to be the same standard.
Keep the canonical instructions and their source links in one place
Store the approved version of the skill instructions in fenbs Skills and prompts, with links to where each tool reads its copy. When you update the skill after a test in one tool, update the fenbs record so every connected assistant and the next person to adapt the skill starts from the current version. Store the source links in Links and listings so you can find each tool's file quickly.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
How to do this with fenbs
- Create a free account, free for one person with no credit card, and add a project for the shared skill instructions.
- Save the canonical (platform-neutral) version of the instructions in Skills and prompts in fenbs with the purpose and required inputs.
- Add the file paths for each tool's copy to Links and listings, with the tool name and the date last updated.
- Connect Claude, and if supported, Cursor or Copilot, to fenbs through Integrations. Each connected assistant reads the canonical instructions before starting.
- When you revise the skill, update the fenbs record first, then update each tool's file to match.
fenbs does not write to .cursorrules or .github/copilot-instructions.md for you. Placing the instructions in each tool's file is a manual step. What fenbs provides is the canonical source that prevents the versions from drifting independently.
For how Claude Code and Cursor compare as coding tools, see Claude Code vs. Cursor. For the full comparison including Copilot, see Claude Code vs. Cursor vs. Copilot.