Claude Skills vs Cursor Skills vs ChatGPT Skills (2026)

Claude skills, Cursor skills, and ChatGPT / Codex skills are the same idea — folders with SKILL.md — discovered in different directories and invoked with slightly different UX. Gemini, Kimi, and GLM clients that adopt Agent Skills read the same files. For UI work, pick the agent you live in daily, keep skills portable, and attach DESIGN.md from the gallery so model choice does not reset your taste.

Short answer

Dimension	Claude Code	Cursor	ChatGPT / Codex
Format	SKILL.md	SKILL.md	SKILL.md
Typical path	.claude/skills/	.cursor/skills/ (+ Claude/Codex dirs)	.agents/skills/ / .codex/skills/
Invocation	Auto + /skill	Auto + /skill	Auto + client skill mentions
Strength	Deep repo reasoning	IDE-native edits + @ context	Ecosystem + coding agents in ChatGPT
Design tip	Strong with explicit specs	Best when @DESIGN.md + skill	Feed the same skill + template text

There is no separate "Claude-only design language." Taste lives in the skill and the template — Claude, ChatGPT, Kimi, and GLM all consume them.

What is the same across Claude, Cursor, and ChatGPT

• Open Agent Skills shape: frontmatter name + description, Markdown body
• Progressive disclosure: descriptions first, full body when relevant
• Optional scripts/, references/, assets/
• Portability: write once, run on multiple agents

That is why a design skill from /skills can move from Claude Code to Cursor without a rewrite.

What differs in practice

Claude skills

Best when: You want the agent to explore, plan, and execute multi-step UI work from the terminal.

Skills behavior: Strong at following long procedural checklists. Good at "read DESIGN.md, then implement hero, then stop."

Watch outs: No visual canvas — you preview in the browser. Vague skills still produce Claude-flavored slop (clean prose, generic layout).

Design stack: Claude Code + taste skill + gallery DESIGN.md + optional Cursor for fine-grained file edits.

Cursor skills

Best when: Daily IDE vibe coding. Composer edits real files with @ references.

Skills behavior: Skills sit beside rules and @ context. Cursor may also load .claude/skills/ and Codex-compatible locations — useful for teams that standardize on one folder.

Watch outs: Wide Composer scopes ignore skill discipline. "Build the whole marketing site" overwhelms even a great skill.

Design stack: /taste-skill + @DESIGN.md + @components/ui + one section per request. See Design UI with Cursor.

ChatGPT / Codex skills

Best when: Your team already works in OpenAI's ecosystem, or you want ChatGPT-orchestrated coding agents on a repo.

Skills behavior: Same SKILL.md standard with Codex-oriented discovery paths and UI for enabling skills.

Watch outs: Workflow varies by ChatGPT surface vs Codex CLI. Always confirm the skill folder is on the machine that runs the agent.

Design stack: Identical skill body + template brief. ChatGPT will not invent better taste than Claude without constraints — skills equalize the floor.

Where Gemini, Kimi, and GLM fit

Model / client	Skills role	Design advice
Gemini (CLI / IDE)	Agent Skills folders under Gemini paths	Use the same taste skill; keep DESIGN.md token-exact
Kimi	Follow vendor skills discovery if Agent Skills-compatible	Great for cost/speed loops — skill quality matters more
GLM	Same — portable SKILL.md when supported	Pair with templates; do not rely on model "design IQ"
Claude	First-class skills in Claude Code	Excellent checklist follower for craft skills
ChatGPT	Skills via Codex / ChatGPT coding	Excellent when the brief + skill are concrete

Open and closed models fail the same way on UI: they imitate the median. Skills + templates are the fix, not brand loyalty.

Feature comparison for design workflows

Need	Claude	Cursor	ChatGPT / Codex	Kimi / GLM
Follow anti-slop checklist	Excellent	Excellent	Excellent	Good if skill loaded
Multi-file UI edit in IDE	Via terminal/tools	Native Composer	Varies by surface	Varies by client
@ file anchoring	Tool-dependent	Excellent	Tool-dependent	Tool-dependent
Slash / explicit invoke	Yes	Yes	Client-dependent	Client-dependent
Share skill across team	Git commit .claude/skills	Git commit .cursor/skills	Git commit .agents/skills	Commit vendor path
Live skill demos	Browse /skills	Same	Same	Same

Which should you standardize on?

Choose Claude-first if planning quality and terminal agents dominate your loop.

Choose Cursor-first if engineers live in the IDE and want skills next to Composer.

Choose ChatGPT / Codex-first if the org already standardizes on OpenAI tooling.

Use Kimi or GLM for parallel cheap iterations — but run the same skill and DESIGN.md, then promote the winning diff back into Claude or Cursor for cleanup.

Most teams should standardize on one skill folder + symlinks, not three rewritten skill dialects.

Migration cheat sheet

From → To	Action
Claude → Cursor	Point Cursor at existing .claude/skills or symlink into .cursor/skills
Cursor → ChatGPT/Codex	Copy/symlink into .agents/skills
ChatGPT → Claude	Copy into .claude/skills; keep name/description intact
Any → Kimi/GLM	Install to that client's skills path; retest triggers

Do not "translate" Markdown into vendor prompt dialects unless a client truly cannot read SKILL.md.

Design QA that works on every model

After any skill-backed UI pass in Claude, Cursor, ChatGPT, Kimi, or GLM:

1. Screenshot the first viewport
2. Check type hierarchy (not one size of Inter)
3. Check layout uniqueness (not three equal cards)
4. Check color roles (accent only where CTAs need it)
5. Diff against DESIGN.md tokens

If it fails, the skill did not fire, the description is weak, or DESIGN.md is missing — not "Claude is bad at design."

Related posts

• Install Agent Skills
• Best design agent skills 2026
• Cursor vs Claude Code vs v0 for UI

Claude skills, Cursor skills, and ChatGPT skills are one standard with three doorways. Add Gemini, Kimi, and GLM as runners — keep the skill and the gallery template as the source of taste.