Cursor Skills vs Claude Code Skills: What Actually Transfers Between Them (2026)

By Rakshit Yadav (@yadavrakshit60)•Oct 2026•9 min read

Short answer: the SKILL.md format itself is identical in both tools — YAML frontmatter with name and description, then a markdown body. What breaks when you move a skill between them is the folder path (.cursor/skills/<name>/SKILL.md vs .claude/skills/<name>/SKILL.md) and a couple of path-relative assumptions inside the body, not the content. Below: why solo founders end up running both tools, the exact directory gotchas that make a working skill silently fail to load in the other editor, what actually ports with zero edits, and a migration checklist.

Why founders end up with both tools installed

Most solo SaaS founders don't pick one editor and stay there. A common pattern: Cursor first because it's the familiar VS Code fork with inline diffs, then Claude Code gets added for anything that benefits from a long, autonomous run — a migration, a refactor across a dozen files, a CI-triggered fix — because its agent loop is built for exactly that. Some founders run Claude Code inside Cursor's terminal pane and get both UIs at once. Either way, you end up with a repo that has skills written for one tool and a growing urge to reuse them in the other instead of writing everything twice.

That's the actual problem this post solves: you have a SKILL.md that works in Cursor, and you want it to also fire correctly in Claude Code (or the reverse), without maintaining two divergent copies.

Directory structure differences that break skill loading

Both tools implement the same open agentskills.io standard for the file itself, but they read it from different places:

CursorClaude Code
Skill path.cursor/skills/<name>/SKILL.md.claude/skills/<name>/SKILL.md
Agent definitions.cursor/agents/.claude/agents/
Commands.cursor/commands/*.md.claude/commands/*.md
Project instructionsCURSOR.md (or AGENTS.md)CLAUDE.md (or AGENTS.md)
Trigger modelAgent matches your task against description, or you invoke with /Agent matches your task against description up front, loads full file only on match

The gotcha is almost never the frontmatter — it's forgetting the file has to physically exist under the other tool's folder too. Copy a skill from .cursor/skills/stripe-payments/SKILL.md into .claude/skills/stripe-payments/SKILL.md and it works immediately, because both tools parse the same two-field frontmatter and treat the body as a plain instruction set. Leave it only in .cursor/skills/ and Claude Code will never see it — there's no auto-discovery across tool boundaries, and no error telling you it's missing. It just doesn't load, and your "it worked yesterday" instinct will send you looking for a bug in the skill body that isn't there.

A second, subtler gotcha: skill bodies that reference other repo paths relative to the tool's own config folder. A skill that tells the agent to "check the rule in ../rules/stripe.mdc" is a Cursor-only reference — Claude Code has no .cursor/rules/ to resolve that against, and there's no equivalent always-on rules folder in Claude Code (the closest parallel is CLAUDE.md itself, which has no glob-scoping). If your skill cross-references another file by relative path, that line needs rewriting per tool, even though the rest of the skill is portable as-is.

What content actually ports

Everything inside the markdown body ports with zero changes:

  • The numbered procedure or checklist — this is the actual instruction set, and it's tool-agnostic markdown.
  • Code snippets, bash commands, SQL — none of it is Cursor- or Claude-specific.
  • Links to other skills, agents, or docs, as long as you keep both tools' copies in sync at the same relative skill name.
  • The description field's trigger phrasing ("Use when the user asks for X, or before Y") — both tools use it the same way, to decide whether to load the skill into context for the current task.

What does not port automatically:

  • The file's location (you need a copy in both .cursor/skills/ and .claude/skills/, not a symlink across tool boundaries unless you've tested that your version control and either tool's file watcher tolerate it).
  • Any reference to the other tool's rules folder, since Claude Code has no glob-scoped always-on rules file the way Cursor's .cursor/rules/*.mdc does.
  • Invocation habits: Cursor lets you type /skill-name to force-load a skill manually in addition to auto-matching; Claude Code's model is purely description-match-triggered. If your skill's instructions assume the user manually invoked it, double-check that assumption still holds when Claude Code decides to load it on its own.

A migration checklist: Cursor skill → Claude Code skill

  1. Create the mirrored folder: .claude/skills/<name>/SKILL.md.
  2. Copy the frontmatter and body verbatim.
  3. Grep the body for .cursor/ or any relative path outside the skill's own folder — rewrite or drop those lines.
  4. Reread the description field as if you were the Claude Code agent deciding whether to load it for an unrelated task. If it only makes sense in Cursor's vocabulary ("when the user runs /review"), rephrase it in terms of the task, not the invocation method.
  5. Test it: give Claude Code a task that should match the description and confirm the skill actually loads (check the agent's reasoning or ask it directly which skill it used).
  6. Keep both copies under version control in the same commit going forward — a skill edited in one folder and forgotten in the other is the single most common way these drift.

The reverse direction (Claude Code → Cursor) is the same checklist, with step 3 looking for .claude/-specific references instead.

Why we recommend Claude Code as the primary surface

Both tools read the identical SKILL.md standard, so this isn't a portability argument — it's a workflow one. Claude Code's agent loop is built for the longer, more autonomous runs a solo founder actually wants from an agent: a multi-file refactor, a CI-triggered fix, a background task that keeps going without you watching every diff. Cursor's strength is the fast, inline, human-in-the-loop edit — you're a much tighter part of the loop by design. If you're choosing where to put your primary skill catalog and your main working habit is "describe the task, let it run, review the result," that points at Claude Code. If it's "watch every change land in the editor as you type," that points at Cursor. Most founders who run both end up treating Claude Code as primary and Cursor as the fallback for quick, supervised edits — which is also why keeping skills mirrored in both .claude/skills/ and .cursor/skills/ matters: you don't want your Claude Code catalog to be the only one that's current.

Try it: AgenticKit skill catalog works in both

AgenticKit ships all 61 skills (31 engineering, 30 marketing/growth), 34 agents, and 47 commands to both .claude/skills/ and .cursor/skills/ in one install — the exact mirroring this post describes, already done:

npx agentickit-cursor init --target claude
# or
npx agentickit-cursor init --target cursor

Run it against both targets in the same repo and you get a tested, pre-synced version of the checklist above instead of maintaining the mirror by hand.

FAQ

Is SKILL.md a Claude-specific or Cursor-specific format? Neither — it's the open agentskills.io standard. Claude Code and Cursor both implement it; AgenticKit and other skill catalogs target it directly so one file set works in either tool.

Why didn't my skill load in the other tool after I copied the file? Almost always a path problem: check it actually landed under the other tool's own folder (.claude/skills/<name>/SKILL.md or .cursor/skills/<name>/SKILL.md), not just renamed in place. Also check the body for relative references to the first tool's rules folder.

Do rules port the same way skills do? No. Cursor's .cursor/rules/*.mdc glob-scoped always-on rules have no direct Claude Code equivalent — the nearest analog is CLAUDE.md, which isn't glob-scoped. If a rule is doing always-on, path-scoped enforcement, you'll need to fold that logic into a skill's description or into CLAUDE.md by hand; see Best Cursor Rules for Next.js and SaaS in 2026 for what belongs in a rule versus a skill in the first place.

Does Claude Code have a personal skills folder that works across all my repos? Yes — ~/.claude/skills/ is read by every project on your machine, in addition to the project-scoped .claude/skills/. Cursor only has the project-scoped folder; see Claude Code Skills vs Cursor Skills: Storage & Portability for what that gap means if you run several repos.

Can I just symlink one folder to the other instead of keeping two copies? You can try it, but test it against your version control and both tools' file watchers first — some setups don't follow symlinks across directories reliably, and a silently-broken symlink produces the exact same "it just doesn't load, no error" failure mode as a missing file.

Which skills are worth installing first if I'm starting from scratch? See 20 Best Claude Code Skills in 2026 or Best Cursor Skills in 2026, both ranked by setup time and real failure modes caught, not generic descriptions.

Which skills are worth mirroring first for a specific area like billing? Best Cursor Skills for Stripe Billing and Subscription SaaS walks through the 8 skill/rule combos worth porting for Stripe code specifically, since that's where generic AI coding help breaks most often.


Published 2026-10-30 by Rakshit Yadav.

Prefer the full Claude Code vs Cursor picture before committing to one as primary? Claude Code vs Cursor for Solo SaaS Founders in 2026 covers the broader tradeoffs this post assumes. Browse the full skills catalog, the free tools, or pricing to install the mirrored set directly.

Related Tools & Agents

🛠️ Free Tool: config-converter🛠️ Free Tool: rules-grader🤖 Agent: code-reviewer🤖 Agent: tech-lead⚡ Command: /auditSkill: project-conventionsSkill: security-basics

Skip the manual mirror

AgenticKit ships the same 61 skills to .claude/skills/ and .cursor/skills/ already synced.

Get the kit →