Skill Reviews
Every agent skill we've reviewed. Each one is labeled with two badges: the test depth (sandboxed, hands-on, smoke test, or desk review) and the functional verification state (whether we asserted on actual output). No inflated scores.
Addy Osmani Agent Skills
A 23-skill engineering methodology framework for AI coding agents, covering the full software development lifecycle from spec to ship.
Adobe Skills
Official Adobe's first-party agent skill pack: 143 skills across 10 plugins spanning AEM, Analytics, Creative Cloud, and browser automation.
Agent Skill Creator
Turn any workflow into a validated, security-scanned agent skill that installs on 17 platforms.
agents-best-practices
Provider-neutral agent skill with 15 reference docs covering harness design, tool permissions, and production readiness for any agent platform.
Agent Skills Specification
The official open specification that defines how Agent Skills work: the SKILL.md format, directory conventions, progressive disclosure, and a Python reference validator.
agentsync (ai-rules-sync)
Zero-dependency CLI that converts and syncs AI agent rule files across 9 coding tools from a single AGENTS.md source.
WebMCP Nexus
Alibaba's frontend toolkit turns plain TypeScript functions into MCP tools via JSDoc and build-time type inference, with SDK, Vite plugin, Webpack plugin, and an AI coding skill.
anthropics/skills
Anthropic's official collection of 17 production and example agent skills. The DOCX, PDF, PPTX, and XLSX skills are the same ones that power Claude.ai's document features in production.
Antigravity Awesome Skills
Installable library of 1,464 SKILL.md playbooks for Claude Code, Cursor, Codex CLI, Gemini CLI, and other AI coding assistants, with an npm installer and editorial bundles.
Awesome GitHub Copilot
GitHub's official community hub of 338 Copilot skills, 212 agents, 183 instructions, and 67 plugins with a plugin marketplace, CI validation, and a companion website.
Agent Toolkit for AWS
AWS official agent toolkit: 43 skills, 3 plugins, and a managed MCP server for Claude Code, Codex, and Kiro.
awslabs/mcp
Official AWS Labs collection of 61 open-source MCP servers covering the full AWS service surface, installable one per server via uvx.
baoyu-design
A Claude Design agent skill that produces polished UI mockups, interactive prototypes, slide decks, and design artifacts as self-contained HTML inside local coding agents.
bb-browser
MCP server and CLI that lets AI agents control your real Chrome browser using your existing login state across 36 platforms.
book-to-skill
Converts technical books and documents into structured agent skills with per-chapter summaries, glossaries, and on-demand loading. Works with PDF, EPUB, DOCX, and more.
Browserbase Skills
Official Browserbase skill pack with 13 agent skills for browser automation, web scraping, search, and cloud functions via the browse CLI.
Claude Code Skills
328-skill multi-domain pack for Claude Code, Codex, Gemini CLI, and 9 more agents. Biggest skill library reviewed.
codebase-memory-mcp
Single static C binary that turns a whole codebase into a queryable knowledge graph for coding agents. 14 MCP tools, 158 languages, sub-ms queries.
Codex PPT Skill
Image-based PowerPoint deck generator for Codex, Claude Code, OpenClaw, and Hermes Agent. Turns articles and papers into full-slide-image PPTX files.
Codex Workflows
Workflow-as-code runtime for OpenAI Codex that orchestrates multi-agent bug sweeps, security reviews, and release diffs through an MCP server and live terminal dashboard.
Context7
The documentation MCP server used by 3.85M npm installs a month, turning library questions into source-cited, version-specific code examples in your prompt.
CowAgent
A 45,904-star open-source agent harness with a cross-compatible SKILL.md skill system, self-evolution, native MCP integration, and 18 IM channel bridges. Formerly chatgpt-on-wechat.
DeerFlow
ByteDance's open-source LangGraph super-agent harness with 28 built-in skills, a native skill security scanner, sandboxed execution, and sub-agent delegation. 76,583 stars.
Engram
Persistent memory system for AI coding agents: single Go binary with SQLite, MCP server, HTTP API, CLI, and TUI.
Framelink Figma MCP
MCP server that fetches Figma design data, simplifies it for LLM consumption, and exposes it to coding agents for one-shot design-to-code implementation.
Garden Skills
A 9,462-star collection of 5 production-grade agent skills by a top Chinese frontend developer, covering web design, GPT Image 2 prompting, HTML articles, video presentations, and knowledge-base retrieval.
GitHub MCP Server
GitHub's official MCP Server. 84 tools connect AI agents to repos, issues, PRs, Actions, code scanning, and discussions over stdio or hosted HTTP.
Google Official Agent Skills
Official Google agent skills covering Google Cloud, Gemini API, BigQuery, GKE, Cloud Run, AlloyDB, Cloud SQL, and Firebase. 13 installable SKILL.md packs with deep reference docs.
Graphify
A knowledge-graph agent skill that turns any codebase into a queryable graph you traverse instead of grep, installed across 23 AI coding agents including native Hermes and OpenClaw.
Headroom
Context compression layer for AI agents that cuts token usage 60-95% on tool outputs, logs, and RAG chunks through a local proxy, MCP server, or inline library call.
Hermes Atlas
The community-curated atlas of the Hermes Agent ecosystem, tracking 185 repos across 12 categories with a live RAG chatbot at hermesatlas.com.
Hermes Dreaming
Staged self-improvement engine that scans source inputs, proposes memory/skill/fact changes, and only writes after explicit review and approval.
SkillHub
Enterprise-grade, self-hosted agent skill registry from iFlytek with CLI, web UI, RBAC, and on-premise deployment via Docker or Kubernetes.
jscpd
Mature copy-paste detector (5.7K stars, 223 languages) that added AI agent skills and an MCP server for on-demand duplication detection.
kepano/obsidian-skills
First-party agent skills from Obsidian's creator, covering Markdown, Bases, Canvas, CLI, and web extraction in five well-structured SKILL.md files.
Laravel Boost
Official Laravel MCP server that gives AI agents database schema, log access, docs search, and code execution in a running Laravel app context.
last30days
Multi-source research skill that searches Reddit, X, YouTube, TikTok, HN, Polymarket, and more across the last 30 days, scoring results by real engagement then synthesizing a structured brief.
lrnev (lrnev-govern)
Project governance engine for AI-assisted development. MCP server plus CLI dual-mode, file-based truth, zero model dependency, with 39 tools covering specs, tasks, scenes, ADRs, memory, and multi-agent coordination.
Matt Pocock Skills
A 178K-star single-author engineering skill pack for Claude Code and Codex built around a grilling-first workflow (idea to spec to tickets to TDD to review).
googleapis/mcp-toolbox
Official Google MCP server for databases that turns 35+ enterprise datasources into agent-callable tools, with a built-in skills-generate command that emits SKILL.md files from YAML tool definitions.
OpenClaw (metamask-openclaw)
A MetaMask SDK wrapper and agent skill for wallet operations that cannot be installed or built due to every source file containing the wrong content.
Playwright MCP
Microsoft's official MCP server for browser automation, giving agents 68 Playwright tools and 5 million weekly npm downloads.
SkillOpt
Microsoft's automated skill optimizer trains natural-language skill documents through iterative rollout, reflection, and validation cycles.
Microsoft Waza
CLI for evaluating AI agent skills with benchmarks, graders, and CI integration. Microsoft's official tooling for skill quality assurance.
MCP Inspector
The official visual and CLI testing tool for MCP servers, from the modelcontextprotocol org. 85 passing CLI tests, 765K monthly npm downloads, security-hardened after CVE-2025-49596.
mrtooher/fable-mode
A doc-only Claude skill that enforces staged execution discipline on complex tasks, riding the "Fable wave" of community recreations of a discontinued Anthropic capability.
NotebookLM MCP CLI
The only programmatic bridge to Google NotebookLM for AI agents, shipping a 39-tool MCP server, a 22-command CLI, and a native agent skill in one pip-installable package.
NVIDIA Agent Skills
NVIDIA's official agent skill catalog: 108 verified, signed, and benchmarked skills covering GPU computing, AI training, inference, robotics, and simulation.
NVIDIA SkillSpector
NVIDIA's security scanner for AI agent skills detects 64 vulnerability patterns across 16 categories with live CVE lookups and SARIF output.
OpenViking
ByteDance's official 26K-star context database for AI agents unifies memory, RAG, and skills under a filesystem paradigm with three-tier loading, 16 MCP tools, and native Hermes integration.
pentest-ai (ptai)
Offensive-security MCP server that oracle-verifies every finding by re-running the exploit. 52 MCP tools, 17 specialist agents, no API key on the Claude Code path.
Planning-with-Files
Persistent file-based planning for AI agents that turns markdown files into crash-proof working memory. Works with 17+ agent platforms.
pm-claude-skills
A 1,178-star professional skill library with 466 SKILL.md files, 32 executable calculators, an MCP server, and a cross-agent installer, listed in Anthropic's official Claude plugin directory.
PM Skills Marketplace
A marketplace of 68 PM skills and 42 chained workflows across 9 plugins, covering product discovery through AI shipping.
Ponytail
Multi-platform YAGNI-discipline skill that makes AI agents ship the laziest correct solution, measured at 54% less code on a real agentic benchmark.
FastMCP
The standard Python framework for building MCP servers and clients, with 5,617 tests, 3.25M daily PyPI downloads, and official SDK integration.
RAPTOR
Autonomous offensive/defensive security research framework built on Claude Code. Scan, validate, exploit, and patch via 30 slash commands and a 638K-line Python engine.
Comet
A CLI that chains OpenSpec and Superpowers into one phase-guarded, resumable five-stage workflow across 30 AI coding platforms.
Scientific Agent Skills
A collection of 139 agent skill docs spanning bioinformatics, drug discovery, clinical research, ML, and 78 public databases, built on the Agent Skills standard for Cursor, Claude Code, and Codex.
Snyk Agent Scan
Snyk's CLI scanner that auto-discovers agent configs, MCP servers, and skills on your machine, then checks them for prompt injection, tool poisoning, and supply-chain threats.
Google Stitch Skills
Eight-agent skill pack for Google Stitch covering UI design, prompt enhancement, multi-page site generation, React conversion, walkthrough videos, and shadcn/ui integration.
Supabase Agent Skills
Official Supabase agent skills covering database operations, auth, storage, realtime, and Postgres best practices for 18+ AI agents.
Superpowers
Cross-platform agent skills framework providing a complete TDD-first development methodology for AI coding agents across 8 harnesses.
TanStack Intent
CLI for npm library maintainers to generate, validate, and ship Agent Skills alongside their packages. TanStack's official tooling for package-native skill distribution.
Vercel Agent Skills
Vercel's official collection of 8 agent skills for deploying to Vercel, optimizing costs, and following React/Next.js best practices from the Vercel Engineering team.
Vercel Skills CLI
The canonical package manager and runtime for the open agent skills ecosystem, installing SKILL.md files across 73 AI coding agents from a single CLI.
🏆 THE STORY: HEADROOM ENDS GRAPHIFY'S EIGHT-PEAT (+521 → 60,483★ — leads the gain board for the FIRST TIME, snapping Graphify's 8-scan streak) + ponytail's 66-window gap-widening streak BROKEN (headroom outgained ponytail, gap narrowed 116) + Graphify still pulls away from caveman (gap 663 → 864) + NVIDIA/SkillSpector 93rd consecutive + Smithery directory scrape
Agent skill pending review.
NVIDIA/SkillSpector 93rd consecutive (+29 → 13,447★ — RECORD extended a 43RD TIME past 50; microsoft/SkillOpt narrowed the gap a 4TH consecutive scan — 274 → 257)
Agent skill pending review.
NEW ESTABLISHED MCP TRACKS (10): borski/travel-hacking-toolkit + classfang/ssh-mcp-server + wshobson/maverick-mcp + RyanAlberts/best-of-Agent-Harnesses + Lyellr88/marm-memory + DAWNCR0W/affine-mcp-server + dmmulroy/overseer + finite-sample/rmcp + Qoyyuum/mcp-metatrader5-server + zilliztech/mcp-server-milvus [OFFICIAL]
Agent skill pending review.
NEW ESTABLISHED SKILL TRACKS (10): JackyST0/awesome-agent-skills + JimLiu/baocut + Peiiii/nextclaw + contentful/skill-kit [OFFICIAL] + fallow-rs/fallow-skills + ollygarden/opentelemetry-agent-skills + simota/agent-skills + mblode/agent-skills + HLND2T/CS2_VibeSignatures + PixVerseAI/skills [OFFICIAL]
Agent skill pending review.
Fresh Concept-Novel MCP Creations (created in-window Jul 19–20, 0–3★) **[NEW TRACK]**
Agent skill pending review.
Fresh Concept-Novel Skill Creations (created in-window Jul 19–20, 1–24★) **[NEW TRACK]**
Agent skill pending review.
The "skill-quality / skill-governance" META tier — now FIVE repos across three consecutive scans (#148–#150), graduating from signal to ESTABLISHED VERTICAL
Agent skill pending review.
alphaparkinc "X-like" product-clone skill fleet — 6 MORE repos this window (cluster now 100+ repos)
Agent skill pending review.
Directory Scrape: Glama — 57,058 servers (+32, FORTY-FIRST CONSECUTIVE INCREASE — ROUND-41 watermark)
Agent skill pending review.
🏆 THE STORY: GRAPHIFY EIGHT-PEAT (+510 → 91,359★ — CROSSED 91K; leads gain board for the 8TH consecutive scan, extends its OWN all-time pipeline-RECORD) + ponytail ROUND-66 + OpenMontage CROSSED 40K (61st consecutive) + NVIDIA/SkillSpector 92nd consecutive + the MEGA blind-spot olimorris/codecompanion.nvim 6,751★ (canonical Neovim agent-host, missed across 148 scans)
Agent skill pending review.
Five skills that mattered, every Friday
Honest gear ratings and verdicts in your inbox. No spam, no affiliate links, no sponsors.
Want something reviewed?
Tell us which AI agent skill we should test next.