Of the hundreds of repos the two big awesome-claude-skills lists link to, start with Anthropic’s anthropics/skills, add obra/superpowers if you write code, then pick by job: Playwright for browser testing, Trail of Bits for security, Expo for React Native, scientific-agent-skills for research. Skip anything archived (get-shit-done is), unlicensed or untouched for a year. The lists look the same whether a link is solid or dead, which is why this page exists.
So I went through them with the GitHub API on 2 November 2026 and checked every linked repo I could for stars, licence and last push date. What follows is the short list I would install, the ones I would skip, and the checks to run before you trust any skill on your machine.
One honesty note up front: this is a documentation-based curation. I read READMEs, SKILL.md files and licence files and pulled repo metadata. I did not run benchmark tests on these skills for this article, and nothing below claims test results.
If you are new to the format, read what Claude Code skills are first.
The short answer: which awesome Claude skills to install first
Start with Anthropic’s own anthropics/skills repo (skill-creator, frontend-design, webapp-testing), then add obra/superpowers if you write code. After that, pick by job: lackeyjb/playwright-skill for browser automation, trailofbits/skills for security review, expo/skills for Expo apps, and K-Dense-AI/scientific-agent-skills for research work. Skip anything archived, unlicensed or untouched for a year.
Here is the shortlist. All numbers come from the GitHub API on 2 November 2026.
| Repo | What it does | Stars | Licence (per GitHub) | Last push | Verdict |
|---|---|---|---|---|---|
| anthropics/skills | Official skills: skill-creator, frontend-design, webapp-testing, docx, pdf, pptx, xlsx and more | 186,355 | Per skill (see below) | 2026-10-28 | Install first |
| obra/superpowers | TDD, debugging, planning and review workflow skills | 310,623 | MIT | 2026-10-29 | Install if you code |
| lackeyjb/playwright-skill | General browser automation with Playwright | 3,261 | MIT | 2026-11-01 | Install for testing and scraping |
| trailofbits/skills | CodeQL, Semgrep, variant analysis, code audit | 7,937 | CC-BY-SA-4.0 | 2026-10-28 | Install for security work |
| expo/skills | Official Expo app development skills | 4,008 | MIT | 2026-11-01 | Install if you ship Expo |
| K-Dense-AI/scientific-agent-skills | Scientific libraries and databases | 55,663 | MIT | 2026-10-26 | Install for research |
| zarazhangrui/frontend-slides | Animated HTML slide decks, PPTX conversion | 31,702 | MIT | 2026-06-23 | Install for decks |
| yusufkaraaslan/Skill_Seekers | Turns documentation sites into skills | 15,263 | MIT | 2026-09-30 | Install as a tool |
| conorluddy/ios-simulator-skill | iOS simulator build, navigation, testing | 1,307 | MIT | 2026-10-19 | Install for iOS |
| mhattingpete/claude-skills-marketplace | Plugin marketplace of mixed skills | 689 | Apache-2.0 | 2026-07-25 | Browse, cherry-pick |
Takeaway: ten repos cover most real needs. Everything else in the lists is either a niche tool or a risk you do not need to take.
What the awesome-claude-skills lists are (and what they are not)
An awesome list is a curated README of links. It does not host, test or audit the skills it points to. The two lists people mean when they search “awesome claude skills” are ComposioHQ/awesome-claude-skills and travisvn/awesome-claude-skills.
ComposioHQ/awesome-claude-skills
This is the bigger one by stars: 78,518 on 2 November 2026, with a last push on 18 September 2026. GitHub detects no licence for the repo itself. Unlike a pure link list, it also ships skill folders directly in the repo (32 top-level directories when I checked), plus a connect-apps plugin that uses Composio’s own service to let Claude “send emails, create issues, post to Slack”, in the README’s words.
Worth knowing: Composio is a commercial integrations company, and the README carries banners and links to its MCP Gateway and dashboard. That does not make the list bad. It does mean the list is also marketing, and the integration plugin routes actions through a third-party service.
travisvn/awesome-claude-skills
The smaller and tidier list: 15,666 stars, last push 28 April 2026, no licence detected. Its README badge reads “updated Feb 2026”. It splits official skills from community ones, has a security warning before the community section (“Skills can execute arbitrary code in Claude’s environment”), and keeps a short table of individual skills.
Why that matters
| Question | ComposioHQ list | travisvn list |
|---|---|---|
| Stars (2 Nov 2026) | 78,518 | 15,666 |
| Last push | 2026-09-18 | 2026-04-28 |
| Repo licence detected | None | None |
| Hosts skill code itself | Yes | No, links only |
| Vendor tie-in | Composio integrations | None obvious |
| Security warning in README | Not prominent | Yes, above community skills |
Neither list removes dead links on a schedule. The travisvn list has not been pushed in six months as of this writing, and two repos it links (get-shit-done and, on Composio’s side, notebooklm-skill) are now archived. Treat both lists as a starting map, then check each repo yourself.
How I picked: the four checks every repo had to pass
A repo made the shortlist only if it passed four checks: active within roughly six months, a licence GitHub can detect, a clear SKILL.md, and a scope that matches what it claims. Stars were a tiebreaker, not a pass mark.
- Last push. I pulled
pushed_atfrom the API. Claude Code’s skill and plugin format has changed through 2025 and 2026, so a repo untouched since October 2025 may predate parts of it. - Licence. No licence means no granted right to copy or modify, under default copyright. For a personal experiment you may not care. For a team repo you should.
- Archived flag. If
archivedis true, the owner has stopped accepting changes. Bugs stay bugs. - Scope versus permissions. I read what each skill tells Claude to do. A d3 chart skill that only writes JavaScript is low risk. A skill that drives a logged-in browser session or deploys code carries far more.
What I did not do: run each skill against a fixed task and score the output. If you want that, the sources and methodology section at the end lists exactly what a fair test would need.
Stars are a weak signal on their own
obra/superpowers has more stars (310,623) than anthropics/skills (186,355). That says it is popular, not that it suits you. Superpowers is a full development methodology. If you do not write code, its planning and TDD skills will fire on prompts where you never wanted them. Pick by job first and use stars only to break ties between similar repos.
The official pick: anthropics/skills
Anthropic’s repo is the safest first install because it is maintained by the company that builds the skill format, and it was pushed on 28 October 2026. It currently has 19 skill folders, including skill-creator, frontend-design, webapp-testing, mcp-builder, doc-coauthoring, theme-factory, claude-api and the four document skills.
Which ones to actually turn on
- skill-creator. Interviews you and writes a SKILL.md. This is the fastest way to stop collecting other people’s skills and start writing your own. Our guide to building a Claude Code skill walks the same steps by hand.
- frontend-design. Pushes Claude away from generic layouts. The travisvn list says it “works very well for React & Tailwind”. That is the list’s view, not a test I ran.
- webapp-testing. Playwright-based testing of a local app. Overlaps with lackeyjb/playwright-skill, so pick one.
- mcp-builder. Only if you build MCP servers.
- docx, pdf, pptx, xlsx. Useful if you produce office files.
The licence catch most posts skip
GitHub detects no single licence for the repo, because licences sit per skill. I opened two. frontend-design/LICENSE.txt is Apache 2.0. docx/LICENSE.txt opens with ”© 2025 Anthropic, PBC. All rights reserved” and says use is “governed by your agreement with Anthropic regarding use of Anthropic’s services.”
So the document skills are source-available, not open source. Using them with Claude is fine under your Anthropic terms. Forking them into a product or redistributing them is a different question, and you should read the file before you do.
| Skill | Licence file says | Fine to modify and redistribute? |
|---|---|---|
| frontend-design | Apache 2.0 | Yes, under Apache terms |
| docx (and likely pdf, pptx, xlsx) | Anthropic proprietary, tied to your service agreement | Check the file first |
Takeaway: install from this repo first, but read LICENSE.txt in each folder before you copy one into a shared repo.
The best community picks for developers
For developers, obra/superpowers, lackeyjb/playwright-skill, trailofbits/skills and expo/skills are the four community repos that pass every check. All four were pushed within the last week of October 2026 and all carry a detectable licence.
obra/superpowers (310,623 stars, MIT)
Jesse Vincent’s library. The travisvn list describes “20+ battle-tested skills including TDD, debugging, and collaboration patterns” with /brainstorm, /write-plan and /execute-plan commands. Install it through its marketplace: /plugin marketplace add obra/superpowers-marketplace.
Who it suits: people who want Claude to plan before it edits, write tests first, and debug in a set order. Who it does not suit: anyone who wants a quick answer and finds a planning step annoying. It is opinionated by design.
Its sibling, obra/superpowers-lab (435 stars, last push 1 June 2026), is labelled experimental. Leave it alone unless you already use Superpowers daily.
lackeyjb/playwright-skill (3,261 stars, MIT)
General browser automation. Pushed on 1 November 2026, so it is current. It overlaps with Anthropic’s webapp-testing. My rule: use Anthropic’s if you only test your own local app, and lackeyjb’s if you need general automation across sites.
trailofbits/skills (7,937 stars, CC-BY-SA-4.0)
From the security firm Trail of Bits. Covers static analysis with CodeQL and Semgrep, variant analysis and code auditing. The licence is CC-BY-SA-4.0, which is a share-alike content licence. If you modify and redistribute these skills, the share-alike terms apply to your version too.
expo/skills (4,008 stars, MIT)
Official skills from the Expo team. If you build React Native apps with Expo, install this before any community alternative, because the maintainers of the framework wrote it.
conorluddy/ios-simulator-skill (1,307 stars, MIT)
Builds, navigates and tests iOS apps through the simulator. Pushed 19 October 2026. Mac only, by nature.
The best picks for non-developers and research
If you do not write code, the skills worth installing are Anthropic’s document and design skills, frontend-slides for decks, scientific-agent-skills for research, and Skill_Seekers to turn docs you rely on into a skill. Our Claude Code skills for non-developers guide goes deeper on setup.
zarazhangrui/frontend-slides (31,702 stars, MIT)
Creates animated HTML slide decks from scratch or converts PowerPoint files. Last push 23 June 2026, about four months before this check. Still within my window, but watch it.
K-Dense-AI/scientific-agent-skills (55,663 stars, MIT)
Both lists still link it as claude-scientific-skills. The API now returns the name scientific-agent-skills, so the repo has been renamed and GitHub redirects the old URL. It covers specialised scientific libraries and databases. If you do bioinformatics, chemistry or data-heavy research, this is the biggest single collection in either list.
yusufkaraaslan/Skill_Seekers (15,263 stars, MIT)
A tool, not a skill. It converts documentation websites into Claude skills. Handy when you rely on a library whose docs Claude gets wrong. Read what it generates before you install the output, since it is only as good as the source docs.
mhattingpete/claude-skills-marketplace (689 stars, Apache-2.0)
A marketplace bundle from the Composio list. Last push 25 July 2026. Fine to browse. Install single skills from it, not the whole set.
A note on marketing and writing skills
Neither awesome list is strong on marketing, SEO or writing. For those, see Claude Code skills for marketers, Claude Code skills for SEO and skills for humanizing AI writing.
Skip these: the awesome-list entries I would not install
Skip a skill if it is archived, has no licence, has not been pushed in over a year, duplicates something better, or asks for more access than its job needs. Here is how that played out. Every row is from the GitHub API on 2 November 2026.
| Repo | Stars | Licence | Last push | Why I would skip it |
|---|---|---|---|---|
| gsd-build/get-shit-done | 64,184 | MIT | 2026-05-31 | Archived. No more fixes. |
| PleasePrompto/notebooklm-skill | 7,791 | MIT | 2026-09-10 | Archived, and it works through your Google account in a browser. Two reasons in one. |
| asklokesh/loki-mode | 1,114 | NOASSERTION | 2026-11-01 | Active, but the licence is unclear, and the README pitches 37 agents that “build, deploy, and operate” a startup. That is a lot of autonomy to grant. |
| jthack/ffuf_claude_skill | 215 | None | 2025-10-16 | No licence, a year stale, and it is web fuzzing. Only for authorised pentests, and Trail of Bits covers security better. |
| chrisvoncsefalvay/claude-d3js-skill | 251 | None | 2025-10-18 | No licence, a year stale. Claude writes d3 well without it. |
| coffeefuelbump/csv-data-summarizer-claude-skill | 481 | None | 2025-10-16 | No licence, stale, and the document skills overlap. |
| omkamal/pypict-claude-skill | 102 | NOASSERTION | 2026-03-22 | Unclear licence, low activity. |
| smerchek/claude-epub-skill | 170 | MIT | 2025-10-18 | Licensed, but untouched for a year. Fine to read and copy ideas from. |
| obra/superpowers-lab | 435 | MIT | 2026-06-01 | Self-described as experimental. |
| Composio connect-apps plugin | n/a | Repo licence not detected | 2026-09-18 | Lets Claude send email and post to Slack through a third-party service. Only install it if you want exactly that, and scope the connected accounts. |
Why “no licence” is a real problem
Under default copyright, code with no licence grants you no right to copy, modify or redistribute it. Most people install these skills for personal use and nobody sues anyone. But if you commit an unlicensed SKILL.md into your company repo, you have copied someone’s work without permission. Open an issue asking the author to add a licence, or write your own version.
Why “archived” beats “stale” as a red flag
A stale repo might still work and might get updated. An archived one is read-only by the owner’s decision. If Claude Code changes how plugins load, an archived skill stays broken. get-shit-done has 64,184 stars, which shows that popularity and maintenance are separate questions.
The overlap trap
Every installed skill puts its description into Claude’s context so Claude can decide when to use it. Two skills that both claim “browser testing” compete for the same prompts. Pick one per job.
Security checklist before you install any skill
Read the SKILL.md and every script before you install, check what the skill can reach, and prefer repos from known orgs. The travisvn list warns that “Skills can execute arbitrary code in Claude’s environment”. That is the reason for this list.
- Read SKILL.md end to end. It is plain markdown. Look for instructions to fetch remote URLs, run shell commands, or read files outside the project.
- Open every file in
scripts/. Anthropic’s own docx skill ships ascriptsfolder. Scripts are where a skill stops being advice and starts being code. - Search for network calls.
curl,wget,fetch,requests,http. Ask why a writing skill needs the network. - Search for credential access.
.env,~/.ssh,~/.aws,token,cookie, browser profile paths. - Check the licence file in the repo root and inside each skill folder.
- Check
archivedand last push on the repo page. - Pin a commit. Install from a specific commit hash, not a moving branch, so an upstream change cannot alter your skill silently.
- Install one skill at a time. If behaviour changes, you know which skill caused it.
- Keep Claude Code permission prompts on for shell and network tools while you trial a new skill.
- Scope connected accounts. For anything that sends email, posts messages or deploys, use a test account first.
- Prefer org-owned repos (Anthropic, Trail of Bits, Expo) when two skills do the same job.
- Re-read after updates. A skill you trusted in June is a new skill if it changed in October.
Our Claude Code skill install guide shows where skills live on disk, so you can inspect them after install too.
How to install a skill from an awesome list
Most repos install in one of two ways: as a plugin through /plugin marketplace add owner/repo, or by copying a skill folder into ~/.claude/skills/ (personal) or .claude/skills/ (project). The README of each repo tells you which.
Plugin marketplace route
The travisvn list gives these examples:
/plugin marketplace add anthropics/skills
/plugin marketplace add obra/superpowers-marketplaceAfter adding a marketplace, you choose which plugins from it to install. That is where you trim.
Folder copy route
git clone https://github.com/lackeyjb/playwright-skill
cd playwright-skill
git log -1 --format=%H # note the commit you reviewedThen copy only the skill folder you want into your skills directory. Copying one folder rather than a whole repo keeps your context lean.
Project versus personal
| Location | Scope | Use when |
|---|---|---|
~/.claude/skills/ | Every project on your machine | Skills you use everywhere |
.claude/skills/ in a repo | That project, shared via git | Team skills, reviewed in PRs |
Takeaway: project-level skills get code review for free, because teammates see them in the diff.
Common mistakes people make with awesome-list skills
The biggest mistakes are installing whole lists, trusting star counts, and never reading the files. Here are the ones I see most.
- Cloning the entire Composio repo into your skills folder. You get 32 directories of mixed quality and a pile of descriptions in context. Pick folders.
- Treating stars as a quality score. get-shit-done has 64,184 stars and is archived.
- Ignoring renamed repos.
claude-scientific-skillsis nowscientific-agent-skills. Old links redirect today, but pin the new name. - Assuming “official” means open source. Anthropic’s document skills carry a proprietary licence.
- Running two skills for one job. Anthropic’s
webapp-testingand lackeyjb’s Playwright skill overlap. - Installing integration plugins with real accounts on day one. Test with a throwaway workspace.
- Never writing your own. The best skill for your workflow is usually one you wrote in 20 lines. Start with
skill-creator. - Not re-checking. Repos go stale. Re-run the four checks every quarter.
A 30-minute setup plan
Spend ten minutes installing, ten reading, and ten writing your own first skill. That order gets you value fast without blind trust.
Minutes 0 to 10: install the base
- Add
anthropics/skillsand enableskill-creatorplus the one or two others you need. - If you code, add Superpowers through its marketplace.
Minutes 10 to 20: add one job-specific skill and read it
Pick exactly one of Playwright, Trail of Bits, Expo, iOS simulator, scientific skills or frontend-slides. Run the security checklist on it. Note the commit hash.
Minutes 20 to 30: write your own
Ask Claude to use skill-creator to turn a task you repeat weekly into a skill. Founders and solo operators get the most out of this step, see Claude Code skills for founders.
| Step | Time | Output |
|---|---|---|
| Base install | 10 min | Official skills plus Superpowers |
| One reviewed skill | 10 min | A pinned, read job skill |
| Own skill | 10 min | Your first SKILL.md |
FAQ
What is awesome-claude-skills?
It is the name of several curated GitHub lists of Claude skills. The two main ones are ComposioHQ/awesome-claude-skills (78,518 stars) and travisvn/awesome-claude-skills (15,666 stars), as of 2 November 2026.
Which awesome-claude-skills list is better?
The Composio list is bigger and hosts skill code. The travisvn list is smaller, separates official from community skills, and has a clear security warning. Use travisvn for orientation and Composio for breadth.
Are skills in awesome lists safe?
Not automatically. A list only links to repos. Skills can run code, so read SKILL.md and any scripts before installing.
What should I install first?
Anthropic’s anthropics/skills, starting with skill-creator. Add obra/superpowers if you write code.
Are Anthropic’s official skills open source?
Some are. frontend-design is Apache 2.0. The docx skill’s licence says all rights reserved, with use governed by your Anthropic agreement. Check each folder’s LICENSE.txt.
Is obra/superpowers worth it?
If you code and like planning, TDD and structured debugging, yes. It is the most-starred repo in this set at 310,623 stars. If you want quick answers with no process, it may feel heavy.
Can I use a skill with no licence?
You can read it. Copying or redistributing it, especially into a company repo, has no granted permission under default copyright. Ask the author to add one or write your own.
What does archived mean on GitHub?
The owner made the repo read-only. No new commits, issues or fixes. Treat archived skills as frozen.
How many skills should I install?
As few as cover your jobs. Each skill adds its description to Claude’s context, and overlapping skills compete for the same prompts.
Do these skills work in Claude.ai as well as Claude Code?
The travisvn list says skills can be enabled in Claude.ai under Settings, Capabilities, and are also available via the API. Skills that rely on local scripts or a local browser fit Claude Code best.
What is the difference between a skill and an MCP server?
A skill is instructions and optional scripts Claude loads when relevant. An MCP server is a running service that gives Claude tools, like database or API access. Composio’s connect-apps uses the second kind of plumbing.
Is the Composio connect-apps plugin worth installing?
Only if you want Claude to take actions in apps like email or Slack through Composio’s service. Start with a test account and minimal scopes.
Why did a skill stop working after an update?
Either the skill changed upstream or Claude Code’s loading format changed. Pinning a commit prevents the first. Choosing active repos reduces the second.
Where do installed skills live?
Personal skills go in ~/.claude/skills/, project skills in .claude/skills/ inside the repo. Plugins installed via marketplace are managed by Claude Code.
Is a high star count enough to trust a repo?
No. Stars measure attention. Check licence, archived status, last push and the actual files.
Should I build my own skill instead?
For anything specific to your workflow, usually yes. skill-creator makes a first draft in minutes and you know exactly what it does.
Sources and methodology
- GitHub REST API,
GET /repos/{owner}/{repo}for every repo named above, queried 2 November 2026:stargazers_count,license.spdx_id,pushed_at,archived. - README of travisvn/awesome-claude-skills (raw, main branch), read 2 November 2026.
- README and directory listing of ComposioHQ/awesome-claude-skills (master branch), read 2 November 2026.
skills/directory listing andLICENSE.txtfiles fordocxandfrontend-designin anthropics/skills, read 2 November 2026.- Web search for “awesome claude skills worth installing reddit”, 2 November 2026, used for reader questions only. Results were mostly blogs and aggregators, so no Reddit claims are quoted here.
No skill was run or benchmarked for this article. Descriptions of what a skill does come from its README or the awesome list entry, and are attributed where quoted. Star counts change daily.
Your next step
Install anthropics/skills, turn on skill-creator, and pick one job-specific skill from the table at the top. Run the twelve-point checklist on it before your first prompt. Then write one skill of your own. That beats any awesome list.
