Guides

    Awesome Claude Skills: Curated Lists Worth Your Time

    We checked five awesome Claude skills lists on GitHub for size, activity and dead links on 13-11-2026. Which list to open first, and what to install.

    November 13, 202614 min read
    Share:

    Published 13-11-2026. Last verified on 13-11-2026, reviewed monthly because star counts and list contents change every week.

    Quick answer: which awesome Claude skills list should you open first?

    This guide is for developers and small teams who want to pick five to ten Claude skills worth installing this week, without spending a day inside the awesome Claude skills lists on GitHub and without installing something that was abandoned months ago. Our short answer: open travisvn/awesome-claude-skills first if you are new to skills, because it is short, starts with the official Anthropic skills and has a security section. Open ComposioHQ/awesome-claude-skills when you want breadth, because it is the largest skills list we counted. Go to VoltAgent/awesome-claude-code-subagents when what you want is a subagent rather than a skill. And treat every list as a starting point for your own check, not as a guarantee: a link in a list says nothing about whether the skill is maintained or safe.

    How we checked the lists

    Our method in three lines: we open every source we cite, we label what GitHub shows separately from what we counted ourselves, and we name what we could not check. The page-specific gate: a list only gets a recommendation if it is not archived, shows ongoing commit activity on its repository page and none of the first five entries we could open returned an error. Everything below was read on 13-11-2026.

    Two parts of the gate we planned did not hold up in this run, and we would rather say so than imply coverage. GitHub's commit history pages were not readable by our fetcher, so we did not verify the date of each list's last commit; we used the commit count shown on the repository page instead. And we did not run a security scan on list entries ourselves, so every safety statement in this article is a method for you to apply, marked "not tested by us" where it matters.

    The gate did filter. hesreallyhim/awesome-claude-code has the second highest star count in our set, but its README body did not render for us, so we could not count entries or open its links. It stays in the table and loses its place as a first stop. jimmysong.io, which ranks for the same search, fell out completely: the page describes an awesome list but lists no skills and no dates.

    The five lists side by side

    These are the numbers we collected on 13-11-2026. Stars, forks, commits and licence are as shown on GitHub that day; the entry counts are ours, made by counting linked list items in each README.

    ListStars (shown)Entries (our count)Commits (shown)LicenceFirst five links we could open
    ComposioHQ/awesome-claude-skills78.5kabout 164 (86 skills plus 78 app skills)77Apache-2.01 of 5 opened, loaded; 4 point to anthropics/skills folders we could not open
    travisvn/awesome-claude-skills15.5kabout 2942not shown0 of 5 opened; all 5 point to anthropics/skills folders
    VoltAgent/awesome-claude-code-subagents26.3k162 subagents535MIT4 of 5 loaded, 1 returned an error
    hesreallyhim/awesome-claude-code56.8knot readable2,266not shownnot readable
    anthropics/skills187.5knot listed in README61mixed, see belownot applicable

    The order of our recommendation is not the order of the stars. travisvn sits first for most readers because 29 entries is a list you can read in ten minutes, and it begins with the official skills you should try before anything else; what would move it down is a long gap between updates, which we could not date. ComposioHQ sits second because its 86 skills cover the widest range of jobs, from document processing to security; it loses first place because its opening section sells Composio's own app connector, which needs a separate account. VoltAgent's subagents list sits third because it is the only one built for a different file type, and the most active by commit count apart from hesreallyhim. hesreallyhim/awesome-claude-code sits fourth: 2,266 commits suggests a well-kept list, but we could not read it, and we do not rank what we cannot check. anthropics/skills is not a curated list at all; it is the source the other lists point to, so we treat it as the reference rather than a competitor.

    Two numbers to read with care. The star counts on the README pages differed from the repository pages for three lists (Composio showed 74.3k on its README page), so treat stars as a rough signal of attention, not of quality. And the licence matters more than people think: the anthropics/skills README says many of its skills are Apache 2.0, while the document skills (docx, pdf, pptx, xlsx) are source-available, not open source. If you plan to fork one into a product, read that line first. For a ranking of individual skills rather than lists, our guide to the best Claude Code skills of 2026 does that job.

    What are Claude skills, and why do lists exist at all?

    A Claude skill is a folder with a SKILL.md file: instructions, and sometimes scripts and reference files, that Claude loads when a task matches the skill's description. That is the difference with a long prompt. The skill sits in the background until it is needed, so you can install twenty without filling every session with them.

    Lists exist because skills are easy to publish and hard to find. Anyone can put a SKILL.md in a public repository, and there is no central index on GitHub. Awesome lists fill that gap with a human filter: someone decided an entry was worth linking. That filter is the whole value, and it decays. A list that was well kept in spring links to repositories that have since been renamed, archived or taken over, and nothing in the list tells you which.

    All three pages that rank for this search start in the same place, and so should you: the official skills in anthropics/skills. They cover document work (Word, PDF, PowerPoint, Excel), creative tasks and a skill for writing skills. We send almost everyone there first, because those skills show what a good SKILL.md looks like, and they give you a baseline to judge community entries against.

    Which list fits which job?

    The right list depends on the job you want to hand off, not on the star count. Here is where we would start for the four jobs we see most often.

    Documents and office work. Start with the official document skills, reached through travisvn/awesome-claude-skills, which lists them by type. For a team that produces proposals, decks and spreadsheets every week, those four skills cover most of the need. When you want more, our productivity skills collection has skills for briefings, checklists and meeting notes that the lists barely touch.

    Code and reviews. ComposioHQ/awesome-claude-skills has 32 entries under Development and Code Tools in our count, the largest single category in any list we checked. That is where we would look for test, Git and review helpers. If you would rather filter by job than scroll, the code quality and review collection groups skills that review diffs, catch regressions and enforce conventions.

    Subagents for parallel work. That is a different file type with its own list, covered further down.

    Business apps such as CRM, mail and calendars. The Composio list has 78 app skills, but they run through Composio's own connector. That is a fair choice if you already use Composio, and an extra account plus API key if you do not. We would only go that way for a team that needs many SaaS apps at once.

    Our verdict across all four: no single list wins every job. The pattern that works for teams we talk to is one list for discovery and one check you always run before installing, which is the next section.

    How do you get from a list entry to an installed skill?

    You copy the skill folder into Claude's skills directory, restart the session and check that Claude picks it up. The list gives you a link; the work is in the four steps around it.

    1. Read the SKILL.md before anything else. The description field decides when the skill fires. A vague description means it will rarely trigger; an overly broad one means it fires on tasks you did not intend.
    2. Copy the folder, not just the file. Many skills include scripts or reference files next to the SKILL.md. Copy the whole folder into ~/.claude/skills for personal use or .claude/skills in the repository for the team. Our step-by-step guide on installing Claude skills from GitHub covers paths and the common mistakes.
    3. Test it on a real task. Ask for the job the skill is meant to do and see whether Claude names the skill. If it does not, the description is the first suspect.
    4. Write down where it came from. A one-line note with the repository and the date you copied it saves an afternoon when something changes upstream.

    Two of our skills shorten this. Skill Finder: Find and Install the Right Skill is free and, by its listing, searches local skills, directories and GitHub for a task, ranks the candidates and installs the one you pick. Agent config auditor, also free, checks a Claude Code or Cursor setup for shallow skills, missing hooks and vague tool descriptions once you have installed a handful. Neither was run by us for this article.

    Link rot is the practical problem at step two. A README with 164 links will have dead ones, and checking them by hand is an afternoon. Your agent can fetch the pages for you through the scraping tools in Agensi Tools: on 13-11-2026 that page listed a single page fetch from $0.0013 per call as a starting price, with no account at the scraping provider needed, and failed uses are not charged. And when no list has the skill you need, writing your own is often faster; our skill creator collection is where we would start.

    How do you spot an abandoned or unsafe entry?

    Check three things before you install: when the repository last changed, who maintains it and what the skill can run. A skill is code with instructions attached, so the same caution applies as for any dependency.

    Abandoned. Open the repository and look at the date of the last commit, open issues without replies and whether the README still matches the folder. Our cut-off for a skill we would put into a team setup is activity in the last 90 days, unless the skill is so small it has nothing to maintain. Doing that for twenty entries is tedious. Your agent can pull public repository data such as releases, issues, contributors and forks through the GitHub tools in Agensi Tools without a GitHub token of your own. On 13-11-2026 that page showed starting prices from $0.0002 per call for repository issues and files, and the agent tells you the cost of each use.

    Unsafe. Read every script the skill ships. Look for network calls to unknown hosts, commands that read your environment variables or SSH keys and instructions that tell Claude to skip confirmations. Our article on whether AI agent skills are safe lists the patterns we see most, and how the Agensi security scan works explains what we check before a skill goes live on our marketplace.

    Two skills automate the first pass. Pre-Install Skill Security Gate is free and gives a SAFE, CAUTION or DO NOT INSTALL verdict on a third-party SKILL.md or skill folder before you install it, based on a static checklist. Skill Provenance Audit, at $5, is built for the other direction: it checks a skill folder against its upstream source, licences and aliases before you ship or sell it. A static check catches the obvious; it does not replace reading the code. For more tools in that category, our security skills collection is the place to browse.

    We sell skills that Claude Code, Cursor, Codex CLI and Copilot load through one connection to Agensi (see the MCP setup guide), and pay-per-call tools your agent uses through one MCP connection, and we name ours in each section of this article. No list owner or vendor paid to be included.

    Are there awesome lists for Claude Code subagents?

    Yes. VoltAgent/awesome-claude-code-subagents is the one we would use: 162 subagents in our count, in ten categories from core development to research, MIT licensed and 535 commits shown on 13-11-2026. Subagents are not skills. A subagent is a separate Claude instance with its own instructions and context window that the main session hands work to; a skill is knowledge the current session loads.

    Of the five links we opened from this list, four loaded and one returned an error, which is the best result in our sample and the reason it passes our gate. The catch is the size: 162 agents is more than any team needs. Our advice is to pick two or three for the jobs you repeat weekly, such as a reviewer and a test writer, rather than installing a category.

    Running several subagents well is mostly about splitting the work. Agent Orchestrator: Parallel Coding Agents, free, breaks a large task into a task graph and runs Claude Code, Codex or Gemini CLI in parallel in git worktrees, with review and ordered merges. Sub-Agent Task Briefs, $5, writes a self-contained brief per subagent with inputs, outputs and who owns which files. The wider agents and orchestration collection has more of this kind.

    When is a marketplace better than a GitHub list?

    A GitHub list is better for discovery and free experiments; a marketplace is better when someone in your team has to answer for what was installed. The difference is accountability, not price.

    A list gives you links. Nobody checks whether an entry still works, nobody scans it and when it breaks there is no one to ask. That is fine for a developer trying things on a side project. It is not fine for a team of five whose agent touches client repositories, where one bad script is a long day.

    A marketplace adds a review step, a security scan and a page that tells you price and supported agents before you install. On Agensi, every skill page shows which agents it supports and whether it is free or paid, and many are free; our list of the best free Claude Code skills shows how far you get without paying. Where we would still send people to GitHub: the official Anthropic skills, and very new skills that no marketplace has picked up yet.

    If your bottleneck is writing better prompts and instructions rather than finding skills, the prompt and skill engineering collection is the more useful place to spend an hour.

    Your next step

    Pick one list from the table, take three skills for jobs you repeat every week and run them through the install and safety checks above before anything goes into a shared repository. If you want your agent to do the checking, connect it once through Agensi MCP and use Skill Finder plus the GitHub tools named above; the agent shows the price of each tool use and your balance pays only for what runs. Star counts and list contents move weekly, so check the numbers on GitHub against ours before you rely on them.

    Written by Samuel Rose, Founder of Agensi. Next scheduled review: 13-12-2026.