Compare / Claude Code skills libraries compared (2026)
Claude Code skills libraries compared: Skillry, AgenKit, Agensi, skills.sh, HookCrate
Published 2026-10-11 · Last verified 2026-10-11 · 891 words · figures link to their sources; numbers change daily.
A "skill" for Claude Code is a folder with a SKILL.md — instructions, examples, sometimes scripts — that the agent loads when a task matches. Because the format is a plain file, a whole market of libraries has appeared to sell, host or index them. They differ far more in delivery and trust than in price. This page compares the ones named most often, plus one that search kept returning, with each figure linked to the page it was read from on the "last verified" date.
A disclosure: HookCrate is the author's product and the smallest catalogue on this page. Several competitors are free, several are much larger, and the table says so.
The comparison
| Library | Price | Catalogue size (as stated) | Delivery | Preview before paying | Updates | Source |
|---|---|---|---|---|---|---|
| Skillry (skillry.dev) | Free skills with a free account; Premium "$9.99 / month", "$79 / year", "$169 one-time" (founding price); "Payments are non-refundable" | "400+ Skills so far" | Hosted; install "from the web or CLI", or download a ZIP; browser-authorised CLI | Shows "a real result it made" per skill; file contents before paying not stated | "Ongoing Skill updates" by the same team; per-skill version shown | skillry.dev |
| Skillry (skillry.ai — a different site) | No pricing shown | No total shown; 11 skills on the home page | Download of .md files | Not stated | Listings all dated "Mar 1, 2026" | skillry.ai |
| AgenKit | — | — | — | — | — | agenkit.dev is an Apache-2.0 agent framework (pip install agenkit), not a skills library, and does not mention Claude Code; the closest product by name is "AgentKit", listed at $99 per kit / $149 bundle on aisotools.com (one third-party source — unverified) |
| Agensi | "Some skills are free, but most are paid"; paid skills are one-time purchases; "Flat 30% platform fee", creators "Keep 70%" | "8,500+ Skills" (hero) / "9,011 live skills … across 32 use cases" (same page) | Download ZIP → "Unzip into ~/.claude/skills/"; or connect over MCP and let the agent buy and install | Not stated | "every week"; "8-point automated security scan" per version | agensi.io |
| skills.sh | No paid tier shown; the CLI is open source (MIT, vercel-labs/skills, 33,719 stars per GitHub API) | Leaderboard "All Time (1,273,765)"; top skill shows 3.8M installs | npx skills add <owner/repo> from any GitHub repo; "Made with love by Vercel" | Yes — skills are public repos | Continuous (community repos); public "Audits" section | skills.sh, Vercel blog, 2026-01-20 |
| ClaudSkills | Free "$0"; Pro "$9/mo" or "$79 per year"; Lifetime "$149 one-time" | "200,723+ skills" (headline) vs "3,268 Skills indexed" (stats box, same page) | Free: manual download; Pro: one-click install via desktop app (macOS, Windows) | Free tier can browse and download, so yes in practice | Not stated | claudskills.com |
| HookCrate | "$9 / month" or "$99 lifetime"; two hooks free (MIT) | 8 packs | Hosted CLI (hookcrate install) into .claude/; free hooks as a plugin marketplace | Yes — full sample-session transcript on every pack page | Versioned packs, hookcrate update | this site |
What the differences mean
Index versus catalogue. skills.sh and ClaudSkills are indexes of skills that live in other people's repositories; their headline numbers count the world, not what they wrote. Skillry (.dev), Agensi's paid listings and HookCrate are catalogues: someone is accountable for each item. Indexes win on breadth and price; catalogues win when you want a maintainer to call.
Price is nearly identical where there is one. Monthly tiers cluster at $9–$9.99 and lifetime at $99–$169, so price is not the deciding factor. What varies is what the money buys: Skillry and ClaudSkills sell access to a library, Agensi sells individual skills with the author taking 70%, HookCrate sells a small set with previews.
Delivery decides your workflow. npx skills add and hookcrate install put files in place from a terminal. Agensi's ZIP-and-unzip and Skillry.ai's downloads are manual. ClaudSkills' one-click requires a desktop app. If you install into many projects, a CLI matters more than catalogue size.
Previews are rarer than you would expect. Only skills.sh (public repos), ClaudSkills' free downloads and HookCrate's transcripts let you read what you are getting before you commit. Skillry shows the output a skill produced, which is a different kind of evidence — useful for creative skills, less so for workflow ones.
Trust signals differ. Agensi scans every version with an "8-point automated security scan"; skills.sh publishes audits; HookCrate's packs have tests you can run; most others say nothing. For anything that executes scripts, this is the column to read first.
Picking one
- Free, broad, community: skills.sh — install from any repo, read the source first.
- Many niche paid skills, pay per item: Agensi.
- One subscription, creative-output skills (slides, images, video): Skillry (.dev).
- Index with a desktop installer: ClaudSkills.
- A few engineering workflow packs with transcripts and tests, hooks included: HookCrate — see research-fleet and bitter-pill-audit, or read the research fleet guide to judge the depth before paying.
FAQ
Is "AgenKit" a Claude Code skills library?
Not as far as the public web shows on the verification date. agenkit.dev is an Apache-2.0 distributed-agent framework for Python and Go that does not mention Claude Code. The similarly named "AgentKit" skill kits appear only on one third-party listing with $99/$149 prices; treat that as unverified.
Why are there two Skillrys?
skillry.dev ("Agent Skills with taste", paid tiers) and skillry.ai ("Discover, Rate & Download AI Skills", free downloads) are different sites with different models. Check the domain before comparing prices.
Do any of these work with agents other than Claude Code?
Yes. skills.sh lists use with 17 agents in Vercel's launch post; Skillry.dev lists Claude Code, Codex, Cursor and others; Agensi says the open SKILL.md format works across agents. HookCrate's packs target Claude Code specifically because its hooks depend on Claude Code's hook contract.
Related guides
- Auditing your CLAUDE.md: cut, keep, sharpen
- Running a research fleet with Claude Code subagents and verifying every URL
Related packs
- Bitter Pill Audit — Audits a CLAUDE.md or instructions file for redundant, vague and contradictory rules with keep / cut / sharpen verdicts and token estimates.
- Research Fleet — Decomposes a research question into sub-questions, runs parallel subagents, cross-checks claims and verifies every URL before reporting.