The Best Awesome Claude Skills Alternative for Verified Agent Skills in 2026
Awesome Claude Skills (ComposioHQ) is the largest community-curated collection of Claude agent skills โ 1,000+ listed skills across document processing, development, marketing, and app automation, all free on GitHub โ but it is a raw list with no security scanning, version pinning, or benchmarked proof that a skill actually improves your agent. SkillzBench is the free alternative built for exactly that gap: it will continuously crawl and dedupe skills, pin every version to a commit SHA, security-scan and license-check each one, and benchmark every skill in a sandbox to publish the measured Lift. For small businesses that want verified, safe-to-install skills rather than an unvetted list, SkillzBench is the stronger option.
TL;DR
- Awesome Claude Skills lists 1,000+ free skills and plugins for Claude Code, Claude.ai, Codex, Cursor, Gemini CLI, Antigravity and more โ the largest curated community collection.
- It is a list, not a verification layer: no security scanning, no version pinning, no benchmarks. Published research found roughly 26% of skills carry at least one vulnerability.
- SkillzBench (free, pre-launch) is building the missing layer: commit-SHA version pinning, security scans, license checks, a 0-100 trust score, and sandbox benchmarks that publish each skill's Lift in percentage points.
- Anthropic's official anthropics/skills repo is the safest default today โ first-party, tested, though the four document skills are source-available under a proprietary licence rather than open source.
- obra/superpowers (~298k stars) is the most widely used developer methodology collection; Agensi is the easiest one-click marketplace with install counts.
- Every option here is free; the real differentiators are trust, evidence of value, and how you install and manage skills across agents.
Head to head
| Dimension | SkillzBench | obra/superpowers | anthropics/skills | Agensi marketplace | squirrelscan/skills | Awesome Claude Skills | coreyhaines31/marketingskills |
|---|---|---|---|---|---|---|---|
| Pricing | Free (directory, MCP, benchmarks, alerts) | Free | Free (document skills under proprietary licence) | Free (top skills) | Free | Free | Free |
| Breadth of skills | Continuous crawl + dedupe of GitHub skills | One engineering methodology collection | Smaller official set: documents, design, testing | Curated marketplace, thousands listed | 200+ audit rules, one domain | 1,000+ skills across 10+ categories ๐ | 40-50 marketing skills |
| Trust & safety | Commit-SHA pinning, security scan, license check, 0-100 trust score ๐ | Widely used, no formal scanning | First-party, tested; proprietary licence on document skills | None โ popularity only | None | None โ audit every skill yourself | None |
| Evidence of value | Sandbox benchmarks with/without each skill, Lift in percentage points ๐ | Reputation and adoption | Official testing, no independent benchmarks | Install counts | None | Editorial curation only | None |
| Ease of install | Free MCP server + per-agent loadouts | Official plugin marketplace | Plugin marketplace; already in Claude.ai | One click into ~/.claude/skills/ ๐ | Manual GitHub install | Manual GitHub install per skill | Manual GitHub install |
| Cross-agent support | Any agent via MCP; skills for Claude Code, Codex, Cursor, Gemini CLI and chat apps | Claude Code focused | Claude-first | Claude Code | Agent-agnostic SKILL.md format | Claude Code, Claude.ai, Codex, Cursor, Gemini CLI, Antigravity, Windsurf | Agent-agnostic SKILL.md format |
| Ongoing monitoring | Tracks upstream changes, re-scans, re-benchmarks, notifies ๐ | Actively maintained | Maintained by Anthropic | Marketplace updates | Community maintenance | None | Community maintenance |
| Availability | Pre-launch โ waitlist live | Live now | Live now | Live now | Live now | Live now ๐ | Live now |
The contenders
ComposioHQ's awesome-claude-skills is the most popular community-curated collection of Claude agent skills, listing 1,000+ production-ready skills and plugins organized by category: document processing (docx, pdf, pptx, xlsx), development and code tools, data and analysis, business and marketing, communication, creative and media, productivity, security, and app automation. It covers not just Claude Code and Claude.ai but also Codex, Cursor, Gemini CLI, Antigravity and Windsurf. The repo doubles as a showcase for Composio's MCP Gateway, which provides a single MCP endpoint for 1,000+ app integrations with built-in auth and team access controls. Skills are plain SKILL.md instruction packages that load progressively โ roughly 100 tokens of metadata per skill at session start โ so an agent can host hundreds without context bloat. It is a raw, free GitHub list: broad, current, and practical, but with no security scanning, version pinning, or evidence that any listed skill actually improves agent output.
pricing 0
- + Largest curated community collection โ 1,000+ skills, described by one roundup as the most popular community collection with practical business workflows
- + Covers many agents beyond Claude, including Codex, Cursor, Gemini CLI and Windsurf
- + Well-organized by use-case category, making browsing straightforward
- + Includes practical, business-oriented skills (document processing, legal skill packs, changelog generation, app automation)
- + Free and open on GitHub
- โ No security scanning or vulnerability checks โ reviewers warn of undocumented network calls in community skills and advise auditing every bundled script before installing
- โ No version pinning: skills can change upstream without notice
- โ No benchmarks or evidence that listed skills actually improve agent output โ curation is editorial only
- โ Quality is uneven; skill guides note many skills are 'not detailed enough in the description or trying to do too many things'
- โ Manual installation and management; no connector or per-agent loadout tooling
SkillzBench is AIQ Labs' free, independent directory of AI agent skills, plus a free MCP tool that connects any agent to the skills you choose. It is being built to solve the exact problem Awesome Claude Skills leaves open: with skills everywhere, nobody can tell which ones actually help and which are risky โ published research found roughly 26% of skills carry at least one vulnerability. SkillzBench's pipeline covers five pillars: Discover (continuously crawl and dedupe skills from GitHub), Verify (pin every version to a commit SHA, validate, license-check, security-scan, quality-judge), Benchmark (test every skill in a sandbox with the same model, with and without the skill, publishing the Lift in percentage points), Equip (one free MCP server plus per-agent loadouts, where users toggle skills and tools per agent and the agent only sees what was chosen), and Stay current (track upstream changes, re-scan, re-benchmark, notify). A 0-100 trust score is built from security, quality, license, maintenance and community signals; Lift is always shown separately from ratings and money never touches scores. SkillzBench is pre-launch โ the waitlist is live and the platform is actively being built, so benchmarking, scanning and the MCP connector describe what it will offer at launch.
pricing 0
- + Purpose-built verification layer: security scanning, license checks and commit-SHA pinning directly address the ~26% vulnerability rate found in published research
- + Measured value, not vibes: sandbox benchmarks with and without each skill, with Lift always shown separately from ratings
- + Everything is free โ directory, MCP connector, benchmarks and alerts; revenue comes from blog advertising that never influences scores
- + Per-agent loadouts mean each agent only sees the skills and tools chosen for it, reducing context bloat and risk
- + Continuous monitoring: upstream changes are tracked, re-scanned and re-benchmarked with notifications
- โ Pre-launch: the waitlist/newsletter is live but the platform itself is not yet public, so you cannot use it today
- โ No published benchmark results or live skill counts yet, since the product is still being built
- โ New and unproven compared to established repos with large communities
Anthropic's official skills repository is the reference collection: docx, xlsx, pptx and pdf skills that create, read and edit Word, Excel, PowerPoint and PDF files with real formulas, real slides and filled forms, plus skill-creator (builds and tests new skills against a baseline with evals), frontend-design, mcp-builder, webapp-testing, internal-comms, doc-coauthoring, theme-factory and creative skills. With roughly 181k GitHub stars, it is among the most-used collections. Being first-party, these skills are designed to work cleanly with Claude Code's architecture and are tested against current model behaviour โ the quality bar is higher than community repos, though the collection is smaller. Install via the plugin marketplace or enable directly in Claude.ai's Skills settings. Note the licensing: most skills are Apache 2.0, but the four document skills are source-available under a proprietary licence โ you can use them with Claude but cannot fork them into your own product.
pricing 0
- + Official Anthropic collection โ the safest default, recommended by multiple skill guides over community repos
- + Document skills are described as 'the single most useful set' for anyone producing documents
- + Clean plugin marketplace install; already available in Claude.ai settings
- + skill-creator includes evals so you can test your own skills against a baseline
- โ Smaller collection than community repos like ComposioHQ's
- โ The four document skills are source-available under a proprietary licence โ no forking into your own product
- โ Coding-methodology coverage is thinner than dedicated collections like superpowers
Superpowers is the most widely used skill collection for developers, with roughly 298k GitHub stars. Rather than a grab-bag of helpers, it is an opinionated engineering methodology: brainstorming, writing plans, executing plans, test-driven development, systematic debugging, sub-agent-driven development and working with git worktrees. Skill guides recommend adopting it as a team so every developer's agent works the same way โ plans before code, tests before claiming done โ and warn against installing two competing methodology collections, since two sets of rules about when to plan and when to test give the model conflicting orders. It installs through the official Claude plugin marketplace. It is more a system than a skill list, which is exactly its value for teams that code with Claude Code daily.
pricing 0
- + Most widely used developer collection (~298k stars)
- + Coherent methodology rather than random helpers โ changes how the agent approaches problems
- + Actively maintained with an official plugin marketplace install
- + Free and open source
- โ Developer-focused; little for marketing, documents or business operations
- โ Opinionated โ guides warn not to mix it with other methodology collections
- โ No security scanning or benchmarks of its own
Agensi is a curated marketplace for Claude Code skills where the top picks are all free and install with a single click into ~/.claude/skills/. Its most-installed skills include code-reviewer (764 installs), git-commit-writer (232), humanize-writing (192), readme-generator, pr-description-writer, prompt-engineer, env-doctor, codex-grade-coding, changelog-generator, truth-first, skill-creator and seo-optimizer. Install counts give a popularity signal that raw GitHub lists lack, and the marketplace format makes discovery and installation far easier than cloning repos. Like any marketplace, though, install counts measure attention rather than quality or safety โ there is no security scanning or benchmarking, so the same read-before-install caution applies.
pricing 0
- + Easiest installation path of any option โ one click per skill
- + Install counts provide a rough popularity signal missing from raw lists
- + Top skills are free and cover practical daily tasks (commits, PRs, reviews, env debugging)
- + Marketplace testing adds editorial curation
- โ Install counts measure popularity, not safety or effectiveness
- โ No security scanning, version pinning or benchmarks
- โ Smaller curated surface than the big GitHub repos
marketingskills is the most focused marketing skill collection available, with 40-50 skills covering copywriting, SEO, CRO, paid ads, growth strategy, email sequences and landing-page optimisation โ roughly 54k GitHub stars. Unlike general-purpose repos that happen to include a few marketing skills, this one was built by a marketer for marketers, and the difference shows in specificity: there is a real gap between 'write marketing copy' and 'write a Google Ads headline that fits character limits and includes the target keyword,' and this repo understands it. For a small business running its own marketing, it turns the repetitive parts of campaign creation, content briefs and conversion analysis into repeatable agent workflows. As a community repo it carries the usual caveats: no scanning, no benchmarks, read before you install.
pricing 0
- + Deepest marketing-specific collection, built by a marketer for marketers
- + Highly specific skills rather than generic 'write copy' prompts
- + Free and popular (~54k stars)
- + Practical for small businesses handling their own campaigns
- โ Marketing-only; nothing for documents, dev or operations
- โ No security scanning or benchmarks
- โ Community-maintained with the usual audit-before-install caveat
squirrelscan/skills is a community collection built around website auditing, with 200+ rules covering security, SEO and related site-quality checks. It is a good example of the specialized single-domain repos that have grown up around the skills ecosystem: narrower than the big general collections, but deeper in its niche. For a small business that periodically needs its site checked โ security issues, SEO problems, technical hygiene โ pointing a coding agent at this collection turns a usually outsourced audit into an in-house workflow. Like other community repos, it is free and installed from GitHub, with no independent security scanning or benchmarking of its own.
pricing 0
- + Deep, focused website-audit coverage (200+ rules)
- + Turns site auditing into a repeatable agent workflow
- + Free and open source
- โ Single-domain focus
- โ No security scanning or benchmarks of its own
- โ Community-maintained
Verdict
Winner: SkillzBench
Awesome Claude Skills wins on sheer volume and breadth today, but it offers no verification, security scanning, or evidence that skills help. SkillzBench is purpose-built to close those gaps โ though it is pre-launch, so teams needing skills right now should still use the ComposioHQ list or Anthropic's official repo.
| Option | Overall | Breakdown |
|---|---|---|
| Awesome Claude Skills | 7 | : 10: 9: 3: 2: 6 |
| SkillzBench | 8 | : 10: 7: 10: 10: 7 |
| anthropics/skills | 8 | : 10: 5: 9: 5: 8 |
| obra/superpowers | 7 | : 10: 6: 6: 5: 7 |
| Agensi marketplace | 6 | : 10: 6: 4: 5: 8 |
| coreyhaines31/marketingskills | 6 | : 10: 6: 4: 4: 6 |
| squirrelscan/skills | 6 | : 10: 5: 4: 4: 6 |