The 10 Best Free Claude Code Skill Repos (Honestly Ranked, 2026)
Best free Claude Code skill and agent repos ranked: anthropics/skills, wshobson/agents, davila7 templates — honest strengths, tradeoffs, and token costs.
The best free Claude Code skill repos are anthropics/skills (official reference, start here), wshobson/agents (~192 free agents, largest collection), and davila7/claude-code-templates (biggest community hub). Several are genuinely excellent — use them before paying anyone, including us. The one gap every free repo shares: zero publish per-skill context-token cost. That silence is the single clearest signal for when a paid, curated kit starts making economic sense.
We sell a paid product. We list ourselves last, with that disclosure stated now so the rest of this can be read straight.
How did we rank these?
We weighted four criteria: breadth (surface area covered), quality (how well the skills/agents are actually built), maintenance (is the repo alive in 2026), and usability (can you adopt it without a weekend of plumbing). We explicitly did not weight token transparency in the ranking score, because almost nobody in the free ecosystem has it. We flag it for every entry anyway, because token cost is the operational variable nobody talks about when they tell you to "just install more skills."
We measured token footprints across published SKILL.md files using a tiktoken-compatible counter, falling back to the ~4-chars-per-token heuristic where we had to read rather than clone. Every number we cite is labeled so you know which it is.
The Agent Skills open standard landed December 18, 2025, and the ecosystem grew 18.5x in 20 days. There are now roughly 90,000 skills on skills.sh. The quality range is enormous. That growth makes an honest ranking more useful, not less.
What does the full ranked list look like?
| Rank | Repo | Best For | Token Data? | Maintenance |
|---|---|---|---|---|
| 1 | anthropics/skills | Correctness, learning the format | No | Active |
| 2 | wshobson/agents | Sheer agent breadth (~192 agents) | No | Active |
| 3 | davila7/claude-code-templates | Discovery, community hub | No | Active |
| 4 | claude-seo | SEO production workflows | No | Active |
| 5 | claude-blog | Blogging and long-form content | No | Moderate |
| 6 | marketingskills | Broad marketing coverage | No | Moderate |
| 7 | superpowers | Experimental capability extensions | No | Moderate |
| 8 | jeremylongshore | Practical, coherent task skills | No | Moderate |
| 9 | K-Dense scientific | Research and science workflows | No | Moderate |
| 10 | alirezarezvani | Personas and C-suite frameworks | No | Moderate |
Token data: none. That column is entirely blank because no major free repo publishes measured per-skill context cost. We are not saying that to sell you something. We are saying it because it is the first number you need when deciding what to keep installed.
Should you start with anthropics/skills?
Yes, always. The canonical first-party repo. If you want to understand what a well-formed SKILL.md looks like — the structure the whole ecosystem imitates — this is the source of truth.
Best at: correctness and learning the format. The patterns here are what the Claude Code team itself ships. Start here before installing anything from a third party so you can tell the difference between a well-formed skill and a prompt dressed up in skill clothing.
Trade-off: it is a reference, not a full vertical toolkit. You will outgrow its breadth quickly. No token cost published.
Is wshobson/agents worth installing?
It is the largest free collection: roughly 192 agents and ~84 plugins. Nothing else free comes close on breadth. We reviewed it in detail — the honest version with the numbers — in our full wshobson/agents breakdown. Short answer: it is worth spending an hour in, and not worth bulk-installing.
Best at: sheer breadth of subagents for engineering and general tasks. If a common agentic workflow exists, there is a reasonable chance something here covers it.
Trade-off: 192 agents means quality is uneven and curation is entirely on you. There is no per-agent context-cost figure to help you decide what to drop. The agents you do not prune sit in your context window every session, running up token cost for capabilities you never trigger. That is the invisible tax the repo does not mention.
What is davila7/claude-code-templates good for?
Discovery. It is the largest community hub — a broad, actively maintained collection of templates, agents, and configurations contributed by many people. If a pattern exists in the Claude Code community, it is likely indexed here.
Best at: answering "is there already a template for X" before you write one from scratch. Great starting point.
Trade-off: hub-of-everything means wildly inconsistent depth and house style across contributions. You are assembling a system, not adopting one. No token data anywhere in the repo.
What are the best free vertical repos by domain?
This is where the free ecosystem earns its keep. Several domain-specific repos are genuinely strong for their target use case.
claude-seo (#4): Solid free SEO production coverage — audits, schema, clusters, programmatic pages. It optimizes for output completeness rather than falsifiable, revenue-tied recommendations, but for pure production volume it covers a lot. We built our own SEOKit specifically to add the falsifiability layer on top — not because the free SEO repos are bad, but because a "green checkmark" audit is not the same as a revenue-ranked recommendation with a confidence interval.
claude-blog (#5): A focused free repo for blogging workflows. Tight scope, useful if writing articles is your main job. Not a content department, just a coherent blogging workflow without the bloat of a general kit.
marketingskills (#6): Broad marketing coverage across campaigns, copy, and channels. Better for teams that want a free starting set across many tasks rather than deep expertise in one channel.
superpowers (#7): Experimental capability-extending skills that you will not find in the conventional repos. Good for tinkerers; expect less polish and more unpredictability. Install selectively.
jeremylongshore (#8): A practical, get-it-done collection from a single maintainer. Single-maintainer repos often have more coherent voice than big community hubs. Breadth is limited by one person's priorities, but what is there is consistent.
K-Dense scientific (#9): The only free option we know of aimed at research and science workflows. Genuinely useful if this is your domain. Almost entirely irrelevant outside it.
alirezarezvani (#10): Known for persona-style agents, including C-suite and advisory personas. One improvising persona is not the same as a gated multi-agent workflow — it gives you a voice, not a quality bar or an artifact pipeline — but for quick framework-driven advisory prompts it works.
How should you actually combine free repos without wrecking your context?
A ranking is useless if it just ends with "install all ten." It should not. Here is the operational approach that does not blow up your token budget:
- Anchor on one broad base. Pick one foundation — anthropics/skills for correctness, wshobson/agents if you mostly want subagents — rather than pulling from everywhere at once.
- Add exactly one vertical. Layer in the single domain repo that matches your actual daily work (claude-seo, claude-blog, K-Dense for research) instead of installing all the verticals because they were free.
- Cherry-pick from the hubs, do not bulk-install. From davila7 and wshobson, take the specific agents you will use this week. With ~192 agents available, installing all of them bloats your context window with descriptions for capabilities you will never trigger — and you pay for that in tokens every session.
- Measure what you keep. None of these repos publish token cost, so count the ones you adopt yourself. A tiktoken-compatible counter over the
SKILL.mdfile, or the ~4-chars-per-token estimate for a quick ballpark. Our token-budget playbook walks the full audit-and-prune routine; it applies just as much to free skills as paid ones. - Audit quarterly. Skills you installed six months ago and never invoke are pure overhead. The claude folder context audit pattern is worth running every few months.
The discipline matters more with free repos than paid ones because there is no cost signal stopping you from over-installing. A lean, curated free stack of 8 well-chosen skills beats a 60-skill kitchen sink every time.
Where do free repos fall short compared to paid kits?
Three gaps show up consistently across every free repo on this list.
No token transparency. The entire free ecosystem produces zero per-skill cost data. You cannot make an informed install decision without knowing what you are adding to your standing context. We measured ClaudeKit's 82,197 total tokens across all five kits and publish the ledger on every install. That is not a marketing claim — it is just the number, available to inspect. The real cost of free skills post runs the comparison with actual measurements.
No coherence across verticals. Mixing anthropics/skills with wshobson/agents and claude-seo gives you three different naming conventions, three different output formats, and no shared quality standard. A curated kit is opinionated by design: same command patterns, same evidence-first finish (every command ends with a diff, a report, or a verified file — not a handwavy suggestion).
No read-only specialist agents. Free repos ship agents as autonomous executors. The v2 architecture in ClaudeKit ships agents as read-only specialists: reviewers, auditors, researchers that inform your decisions without taking destructive actions. That is not a safety marketing line — it is a practical workflow difference that matters when you are running 20 commands a day.
If none of that matters for your use case, the free repos above are genuinely good and you should use them.
And then there's us — the paid option, disclosed
We make ClaudeKit — five vertical kits for engineering, marketing, SEO, video, and ecommerce. We are listing ourselves after ten free repos on purpose.
Here is what each kit actually contains, measured:
| Kit | Commands | Skills | Agents | Measured Tokens |
|---|---|---|---|---|
| EngineerKit | 25 | 4 | 4 | 20,413 |
| MarketingKit | 20 | 3 | 2 | 16,714 |
| SEOKit | 19 | 4 | 2 | 16,004 |
| EcomKit | 20 | 3 | 2 | 16,464 |
| VideoKit | 17 | 5 | 3 | 12,602 |
| Total | 101 | 19 | 13 | 82,197 |
Agents are read-only specialists: reviewer, auditor, researcher. No orchestrator agents, no blocking quality gates, no runnable Python tools. Commands end with EVIDENCE — a diff, a report, a verified file — not a reviewer loop.
Install is ck auth <key> then ck install <kit> (global to ~/.claude by default, --local for project scope). Token ledger prints on every install. You can also install via /plugin marketplace add Madni-Aghadi/claudekit-<kit>. Diagnose with ck doctor, recount tokens with ck tokens <kit>.
Pricing: $14.99/mo single kit, $29.99/mo Pro (any 3 kits, swap 1 per cycle), $49.99/mo All-Access. Annual plans: $119/$239/$399. Lifetime $99 per kit as shipped, no future updates included. 14-day refund window. 3 devices per license. See full pricing.
The honest pitch: pay for token transparency, vertical depth, and a coherent command system — or stay free and assemble from the excellent repos above. Both are valid choices depending on how much your time costs relative to the subscription.
FAQ
What are the best free Claude Code skill repos in 2026?
Start with anthropics/skills for the official reference, then wshobson/agents (~192 free agents, largest collection) and davila7/claude-code-templates (biggest community hub). For verticals, claude-seo, claude-blog, marketingskills, superpowers, jeremylongshore, K-Dense (science), and alirezarezvani (personas) all cover specific ground well. All are free; none publish per-skill token cost.
How many agents does wshobson/agents have?
Roughly 192 agents plus about 84 plugins, making it the largest free Claude Code agent collection by count. Quality is uneven across that breadth and curation is left entirely to you — there is no per-agent context-cost figure to help you decide what to keep. See our full review for the breakdown of which categories hold up.
Do any free Claude Code skill repos publish token costs?
No. As of June 2026, no major free repo publishes per-skill context-token cost. It is the single gap the entire free ecosystem shares. You can estimate a skill's cost yourself by running the SKILL.md through a tiktoken-compatible counter, or use the ~4-characters-per-token heuristic for a quick ballpark. Our token cost measurement post walks the methodology.
Should I use free repos or pay for a kit?
Use free repos if they cover your needs — several on this list are genuinely excellent. Pay for a kit if you specifically want published token costs, vertical depth with a consistent command system, and read-only specialist agents that end workflows with evidence rather than suggestions. We make a paid product and still recommend trying the free options first, because a tool you chose intentionally beats one you paid for and partially understand.
What is the Agent Skills open standard and does it affect these repos?
The Agent Skills open standard launched December 18, 2025, and has been adopted by 32+ tools. The ecosystem grew 18.5x in 20 days after launch, with roughly 90,000 skills now on skills.sh. Most repos on this list predate or are adjacent to that standard. It means the quality floor is rising and discovery is getting easier — but it also means there is more noise to filter through than there was six months ago.
How do I avoid overloading my context with too many installed skills?
Pick one base repo, add one vertical, and cherry-pick from hubs rather than bulk-installing. The skills you keep installed sit in your context window every session — you pay for them in tokens whether you invoke them or not. Run an audit every quarter using the claude folder context audit pattern: list everything installed, count its token weight, and drop anything you have not used in 30 days.
If your primary use case is engineering — code review, debugging, shipping features — EngineerKit is the most complete vertical we have built: 25 commands, 4 read-only agents, 20,413 measured tokens, and a /eng debug command that starts from root cause rather than surface symptoms. The getting started guide walks from install to first workflow in under 10 minutes.
Give Claude Code a real team
Five kits, 101 commands, every token measured. Pick the team that matches your work and install it in five minutes.
See the kits
