Claude Code Skills: The 10 Best, Ranked by Verdict
Claude Code skills are SKILL.md folders that Claude Code loads when a task matches, or when you type /skill-name. Designer Skills Collection (78) leads our ranking of 10 skills verified in Claude Code. Each row below carries a verdict, a safety grade and the commit we reviewed. Our Claude Code setup guide shows how to add them.
- Ranked
- 10 verdicts, each with a verified Claude Code run
- Average score
- 76.4
- Verdicts
- 10 Recommended
- Safety grade
- A 5 · B 4 · C 1
- Confidence
- Moderate 7 · Low 3
- Verified in Claude Code
- 10 of 10
- Reviewed
- 2026-09-30
| # | Claude Code skill | Verdict | Score | Safety | Conf. | Best for | Verified in | Category | Commit |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Designer Skills Collection | Recommended | 78 | B | Moderate | Research plans, specs, handoff and critiques | Claude Code | Design | 9a6930c |
| 2 | Remotion Best Practices | Recommended | 77 | B | Moderate | Correct Remotion compositions, Studio preview first | Claude Code | React and web | a9b199e |
| 3 | Animate | Recommended | 77 | A | Low | Fixed curves and durations for one component | Claude Code (Haiku 4.5) | Animation | d16ebe6 |
| 4 | Emil Design Engineering | Recommended | 76 | A | Moderate | Reviewing animation code against concrete rules | Claude Code (Sonnet 4.6) | Animation | d16ebe6 |
| 5 | Frontend Design | Recommended | 76 | A | Moderate | New landing pages from a loose brief | Claude Code (Sonnet 4.6, Opus 5.5) | Design skills | 8a1541c |
| 6 | Brainstorming | Recommended | 76 | B | Moderate | Clarify and get sign-off before code | Claude Code, Codex | Planning | 8ca22db |
| 7 | Systematic Debugging | Recommended | 76 | B | Moderate | Root cause before any fix | Claude Code, Codex | Testing and debugging | 8ca22db |
| 8 | Using Neon | Recommended | 76 | C | Moderate | Neon connections, migrations, read-only diagnostics | Claude Code | Backend | b8250e6 |
| 9 | Semgrep Rule Creator | Recommended | 76 | A | Low | One tested custom Semgrep rule at a time | Claude Code | Security | 82fe822 |
| 10 | Terraform Skill | Recommended | 76 | A | Low | Terraform/OpenTofu modules with plan-before-apply | Claude Code | DevOps and cloud | 0a3a4a6 |
These are the top 10 of the 28 reviewed skills verified in Claude Code, sorted by score. Browse every Claude Code skill we reviewed in the full directory.
What is a skill in Claude Code, and do skills work there?
Yes, skills work in Claude Code. A skill in Claude Code is a folder with a SKILL.md file of instructions and optional scripts. Claude Code loads it when a task matches, or when you type /skill-name.
Claude Code follows the open Agent Skills standard, the same SKILL.md format other agents read. Three facts from the Claude Code skills documentation cover the basics:
- Load: "a skill's body loads only when it's used." Only the short description sits in context before that.
- Invoke: "Claude uses skills when relevant, or you can invoke one directly with
/skill-name." - Store: personal skills go in
~/.claude/skills/, project skills in.claude/skills/, plugin skills in<plugin>/skills/.
Skills differ from slash commands in how they start. Anthropic's Claude Code 101 course says: "Slash commands require you to type them, skills don't." Install steps and the /skills menu sit in our guide to add, invoke and manage skills in Claude Code. The Claude app uses the same format with different loading rules, explained in Claude skills vs Claude Code skills.
Which skills come bundled with Claude Code?
Claude Code ships bundled skills that need no install. The docs list these in their introduction:
/doctor/code-review/batch/debug/loop/claude-api
Bundled skills "are prompt-based," per the same documentation. They sit outside this ranking because nobody chooses to install them. The Claude API skill exists as a standalone Agent Skill too. Our Claude API skill verdict scores it 72, Recommended with caveats, with safety grade B.
What are some examples of Claude Code skills?
Six of the 10 ranked skills show the range of formats. They run from one 71-line file to a bundle of 111 skills.
| Job | Claude Code skill | What Claude Code does differently | Format |
|---|---|---|---|
| UI design | Frontend Design | Plans design tokens and checks them against the brief before code | 1 SKILL.md, 71 lines |
| Design process | Designer Skills Collection | Writes research plans, specs and handoff packages with named UX laws | Bundle: 111 skills, 9 plugins |
| Motion | Animate | Decides whether to animate, then uses fixed curves and durations | 1 skill, 2 files |
| Video | Remotion Best Practices | Writes compositions, opens Studio, renders only on request | Router plus 11 sub-guides |
| Planning | Brainstorming | Asks questions and waits for approval before code | 1 skill plus visual companion |
| Infrastructure | Terraform Skill | Shows a reviewed plan before any apply or destroy | 1 skill plus 8 references |
What is the best skill for Claude Code?
The best skill for Claude Code in our ranking is Designer Skills Collection (78, Recommended). Remotion Best Practices and Emil Kowalski's Animate follow at 77. All 10 ranked skills have a verified Claude Code run.
Designer Skills Collection ranks first on breadth and structure. It packs 111 design skills and 34 commands into nine plugins of plain Markdown, with no runtime code. A task-to-skill index and a /start-here router point Claude Code to the right skill. Its weak side is taste. Outside reviewers rate it stronger on process than on visual judgment, and output quality has no measured comparison.
How were these Claude Code skills ranked?
Only skills with a verified Claude Code run qualify. The overall score is weighted toward the worst failure. Read how we score and rank skills, and see why rankings cannot be bought.
What changed since our review?
Three ranked skills changed upstream after our 2026-09-30 review. The verdicts below stay as reviewed.
- 2026-10-10: obra/superpowers released v7.0.0, which "rebuilt brainstorming" (commit bb92a77). The commit touches both Brainstorming and Systematic Debugging. We reviewed v6.4.2 at 8ca22db.
- 2026-10-15: the
skills/neon-postgrespath changed in commit f333dd0, a merge of PR #141. We reviewed Using Neon at b8250e6. - No path change: the other seven skills show no change to their skill paths after 2026-09-30.
Updates to any verdict appear in our log of verdict corrections and updates.
What are the top 10 Claude Code skills, one by one?
The top 10 Claude Code skills span design, motion, video, planning, debugging, databases, security and infrastructure. Each card states the verdict, safety grade and confidence. It names the main limitation at the reviewed commit.
1. Designer Skills Collection (78, Recommended, B, Moderate)
Designer Skills Collection is a Markdown-only marketplace from Owl-Listener. Its 111 skills cover research plans, specs, handoff packages and critiques. The maintainer merges outside fixes and adds CI guards afterwards. Load and routing defects (#22, #27, #48) reached users in 2026 before fixes landed. Installing all nine plugins adds about 6,500 tokens of descriptions to every session, by our estimate. Compare it with other UI/UX and frontend design skills, ranked.
2. Remotion Best Practices (77, Recommended, B, Moderate)
Remotion Best Practices comes from the framework's own maintainers, in the remotion-dev/skills repository. A small router sends Claude Code to eleven sub-guides for creating, previewing, captioning and rendering React video. Studio opens first, and rendering waits for your request. Broken imports (remotion#11669) and dead links (remotion#9964) were fixed within about a day. A side-by-side test found compositions correct but visually bare. A stale global install silently overrides the project copy (remotion#11809, open). The full Remotion Best Practices review has the evidence.
3. Animate (77, Recommended, A, Low)
Animate is Emil Kowalski's recipe skill for web UI motion. It first asks whether an element needs motion at all. Fixed easing, duration and spring tables follow, plus a dozen ready recipes. Issue #28 measured 90% recall on Claude Haiku 4.5, with no firings on 80 critique or audit requests. Every miss produced finished-looking code with invented curves. Confidence stays Low because the one independent test measured routing, not output. Open Emil Kowalski's Animate, reviewed.
4. Emil Design Engineering (76, Recommended, A, Moderate)
Emil Design Engineering is one Markdown file of motion and polish rules. It gives named cubic-bezier curves, duration tables and a review checklist. The checklist catches ease-in, scale(0), transition: all and centered popover origins. Tessl reports a 1.20x lift on Claude Sonnet 4.6 over two scenarios. The Motion useSpring example never tracks the pointer (issue #26, PR #7). Its checklist flags durations over 300ms, while its own table allows 500ms for drawers. Read the Emil Design Engineering verdict.
5. Frontend Design (76, Recommended, A, Moderate)
Frontend Design is Anthropic's 71-line SKILL.md for new interfaces. It names stock AI looks with hex values. It forces a token plan, reviewed against the brief, before any code. Tessl reports a 1.27x lift on Claude Sonnet 4.6 across 3 scenarios. Element-level defaults still slip through: issue #1808 found pill buttons on 10 of 21 pages. Briefs bound to an existing design system get no guidance. See Frontend Design vs UI/UX Pro Max, or open our Frontend Design verdict.
6. Brainstorming (76, Recommended, B, Moderate)
Brainstorming is obra's design gate from the Superpowers repo. Claude Code clarifies intent, compares approaches and waits for sign-off before writing code. A three-path router (spike, bounded, architectural) scales the ceremony to the task. Tessl reports a 1.81x lift on its single scenario. Questions sometimes advance before the user answers (#1989). The default spec path commits files into docs/ (#1690, #1246). Grade B reflects an optional local Node server. Open the Brainstorming verdict in full.
7. Systematic Debugging (76, Recommended, B, Moderate)
Systematic Debugging is obra's root-cause-first discipline for coding agents. Four phases run in order: investigate, compare patterns, test one hypothesis, then fix. The rules forbid any fix before Phase 1 and require a failing test first. The agent stops after three failed fixes. At v6.4.2 the multi-layer example prints the IDENTITY secret (#2375, fixed only on dev). The waitFor snippet does not compile (PR #1839 open). Unreferenced fixtures ship with every install (#2277). Read Systematic Debugging, reviewed.
8. Using Neon (76, Recommended, C, Moderate)
Using Neon (neon-postgres) is Neon's Postgres skill for agents, published in neondatabase/agent-skills. It covers pooled versus direct connections and Prisma prepared-statement errors. A symptom-to-check table keeps Claude Code on read-only commands during diagnosis. Grade C reflects what it assumes: Neon account access, a DATABASE_URL and a tool that runs SQL. Routing overlaps with the parent neon skill, which absorbed a neon-postgres prompt in a small Haiku pilot (#76). Learn what a C safety grade means, then open the Using Neon verdict.
9. Semgrep Rule Creator (76, Recommended, A, Low)
Semgrep Rule Creator from Trail of Bits drafts one custom Semgrep rule at a time, test first. Rules carry ruleid/ok annotations and must pass 100% of semgrep --test checks. A contributor's benchmark, with 5 runs per arm, shows validating and passing rules. Every run fetches seven external documentation pages. PR #322 reports about 944,000 billable tokens and roughly $0.48 per rule. Local references omit advanced taint features such as pattern-propagators. See the Semgrep Rule Creator review.
10. Terraform Skill (76, Recommended, A, Low)
Terraform Skill by Anton Babenko covers Terraform and OpenTofu modules, tests, CI and state operations. Its response contract requires a reviewed plan before any apply. A targeted destroy with auto-approve once wiped Route53 records (issue #42). A plan-destroy-first rule shipped within three days (PR #43). The eight references total about 44,000 tokens of context, by our estimate. The maintainer declined Terragrunt support (issue #52). Read the Terraform Skill verdict.
Which Claude Code skills should you avoid?
Avoid Landing Page Design (50, Not Recommended) for Claude Code work: it asks for invented statistics. Use Skill Creator (57) with care. Its scripts start Claude sessions that run with your permissions.
- Landing Page Design (50, Not Recommended, safety grade A) claims Claude Code support. Its content rules ask for organic-looking figures like 47.2% and invented brands. PR #1 adds a fix but sits unmerged. Read the Landing Page Design verdict.
- Skill Creator (57, Conditional, safety grade C) ships ten Python scripts that spawn
claude -psessions with your permissions. Its trigger-eval harness returns false negatives (#1721, #1352, #1412) and counts crashes as passes (#1478). See the Skill Creator review. - Loki Mode runs
claude --dangerously-skip-permissionsby default. Its rules tell the agent never to ask or stop. - Skill Seekers launches Claude with permissions skipped on prompts built from scraped pages.
Our checklist shows how to vet a skill before installing.
Which Claude Code skills are best for developers and coding?
For developers, the best Claude Code skills are obra's Brainstorming and Systematic Debugging (both 76). Terraform Skill and Semgrep Rule Creator (76) cover infrastructure and security rules. Using Neon (76) covers Postgres on Neon.
Six ranked skills fit coding work, in four groups:
- Process: Brainstorming (76) scopes features before code. Systematic Debugging (76) finds root causes before fixes.
- Infrastructure and data: Terraform Skill (76) guards apply and destroy. Using Neon (76) picks connection strings and migration paths.
- Security: Semgrep Rule Creator (76) writes tested custom rules.
- Framework: Remotion Best Practices (77) writes correct React video compositions.
The cross-agent ranking of best Agent Skills across agents adds skills verified in other tools.
Which Claude Code skills should you install first?
Install three Claude Code skills first: Brainstorming before new features, Systematic Debugging for bugs and Frontend Design for interfaces. Add Terraform Skill for infrastructure. All four earn Recommended with a verified Claude Code run.
| You are | Install first | Why |
|---|---|---|
| A developer on any stack | Brainstorming · Systematic Debugging | Scope before code; root cause before a fix |
| Building interfaces | Frontend Design · Animate | A deliberate look; motion with fixed values |
| Running infrastructure or data | Terraform Skill · Using Neon | Plan-first changes; safe Neon migrations |
| Doing AppSec | Semgrep Rule Creator | Tested custom rules, one at a time |
| Making product videos | Remotion Best Practices | Correct compositions with Studio preview |
Each skill adds context when it loads. The measured file size per load runs from 1,830 tokens (Semgrep Rule Creator) to 5,105 tokens (Terraform Skill). For the Claude app instead of the terminal, see the best Claude skills for the Claude app.
Which Claude Code skills are verified, and which only claim support?
All 10 ranked Claude Code skills are verified in Claude Code. Of all the skills we have reviewed, 28 are verified in Claude Code. Two of the 10 are verified in Codex.
| Agent | Verified (of 10) | Claimed (of 10) |
|---|---|---|
| Claude Code | 10 | 9 |
| Codex | 2 (Brainstorming, Systematic Debugging) | 5 |
| Cursor | 0 | 5 |
| OpenCode | 0 | 4 |
| Gemini CLI | 0 | 3 |
| GitHub Copilot | 0 | 2 |
Animate claims no agent at all, yet it has a verified Claude Code run on Haiku 4.5. Remotion Best Practices claims Codex. Its only non-Claude run used GPT Sol through a Codex-style client (remotion#11258). We do not count that run as a Codex verification. A claim of Cursor support is not a verified Cursor run. Claude on the web and desktop has its own ranking of skills for Claude on web and desktop.
Where are the best Claude Code skills on GitHub?
The best Claude Code skills on GitHub live in small and large repos alike. obra/superpowers has about 293,500 stars. Designer Skills Collection, our top Claude Code skill, has about 2,800 stars at review.
| Repo | Stars at review | Ranked Claude Code skills |
|---|---|---|
| obra/superpowers | ≈293,500 | Brainstorming (6), Systematic Debugging (7) |
| anthropics/skills | ≈179,000 | Frontend Design (5) |
| emilkowalski/skills | ≈42,500 | Animate (3), Emil Design Engineering (4) |
| trailofbits/skills | ≈7,300 | Semgrep Rule Creator (9) |
| remotion-dev/skills | ≈4,800 | Remotion Best Practices (2) |
| Owl-Listener/designer-skills | ≈2,800 | Designer Skills Collection (1) |
| antonbabenko/terraform-skill | ≈2,400 | Terraform Skill (10) |
| neondatabase/agent-skills | 97 | Using Neon (8) |
Star counts are a snapshot from 2026-09-30. Stars do not predict scores here: Neon's 97-star repo scores 76.
Is there a Claude Code skills library?
Yes. Our directory works as a library: browse every reviewed skill with its verdict and safety grade. Anthropic's own collection is the anthropics/skills repo on GitHub.
How do plugins, marketplaces and MCP servers fit with Claude Code skills?
Plugins bundle Claude Code skills with commands, agents, hooks and MCP servers in one install. Marketplaces list plugins without judging them. MCP servers give Claude Code tools; skills tell it how to use them.
The Claude Code plugin docs list what a plugin adds:
Installing a plugin adds its skills, agents, hooks, and MCP servers to Claude Code.
Plugin skills appear as /<plugin>:<skill>. You install one with /plugin install <name>@<marketplace> (Claude Code plugin docs).
A marketplace is a catalog of plugins. Anthropic's official one is claude-plugins-official, hosted at anthropics/claude-plugins-official. Designer Skills Collection ships as nine plugins in its own marketplace. A listing in a marketplace carries no verdict.
MCP servers expose tools and data, while a skill carries the instructions. Using Neon names the Neon MCP server as an optional way to run SQL. Our guide explains when to use a skill or an MCP server.
What does the community report about Claude Code skills?
The community reports on Claude Code skills through Hacker News threads and GitHub issues. Four of our 10 ranked verdicts cite Hacker News. Each report sits beside our score below.
| Skill (score) | Source | Report | Tone |
|---|---|---|---|
| Frontend Design (76) | Hacker News 49804016 | A commenter calls one piece of its advice vague filler. | Negative |
| Frontend Design (76) | Hacker News 49875147 | Generic anti-slop rules swap one default for another on Opus 5.5. | Negative |
| Frontend Design (76) | GitHub #1808 | Pill buttons appeared on 10 of 21 pages. | Negative |
| Remotion Best Practices (77) | Hacker News 48016741 | The skills work as reusable API documentation for agents. | Positive |
| Remotion Best Practices (77) | Hacker News 48944096 | A user needed repeated prompting for narrative. | Neutral |
| Brainstorming (76) | Hacker News 49856911 | A user is consistently happy with results from its questions. | Positive |
| Systematic Debugging (76) | Hacker News 49871794 | A commenter finds Superpowers-style skills very useful but hard to evaluate. | Positive |
| Animate (77) | GitHub #28 | The skill missed about 10% of real motion requests. | Negative |
Common questions
Does Claude Code have a skill for creating skills?
Yes: Anthropic's Skill Creator, which the Claude Code docs use to run evals. Our verdict scores it 57, Conditional, because its eval harness misreports results.
How do you develop a Claude Code skill?
Write a SKILL.md with a clear description, then test whether it triggers. Our guide shows how to create an Agent Skill step by step.
What are Claude Code skill best practices?
Keep the description specific, the body short and references in separate files. The SKILL.md format guide covers each field.
Does Superpowers work as Claude Code skills?
Yes. Brainstorming and Systematic Debugging from obra/superpowers rank sixth and seventh, verified in Claude Code and Codex at v6.4.2.
Why does my Claude Code skill not trigger?
Claude Code matches requests against the description, and matches miss. Our guide to troubleshoot skill activation covers the fixes.
How many tokens do Claude Code skills use?
The description and when_to_use text share a 1,536-character budget. After compaction, Claude Code re-attaches the first 5,000 tokens of a skill, capped at 25,000 combined.
Are Claude Code skills free?
Yes. All 10 ranked skills sit in public GitHub repos, though Using Neon needs a Neon account.