Agent skills directory: skills worth installing
367 source-checked Claude skills, MCP servers, plugins and agents, each pinned to a public commit.
Setup snippets for
- 367 entries
- 62 publishers
- Checked 17 Oct 2026 – 3 Nov 2026
From the index
How we rankCommunity entries first, then Anthropic official entries; newest pinned commit within each group.
Community entries first; ordered by citations in our workflow guides within each group.
Community entries first; pinned README documents a one-line setup command.
THE SHORTLIST
Featured
A few good places to start. Explore the source, then make it yours.
SkillMIT
GhostPaw Codex
GhostPawJS
Local memory for AI agents in one SQLite file.
SkillMIT
gstack
garrytan
Garry Tan’s Claude Code skill set with role-based tools for planning, design, engineering review, QA, release and docs.
MCPReview source terms
Filesystem
modelcontextprotocol
Implements file-system operations through an MCP server built with Node.js.
AppMIT OR Apache-2.0
PhotoCraft
storytold
Provides layered image editing, including Photoshop document handling and adjustment tools.
PluginMIT
Superpowers
obra
Skills and workflow plugin for coding agents: brainstorm, write a plan, test first, then review the result.
AgentMIT
OpenAI Agents SDK
openai
Provides a Python framework for agent workflows that supports multiple model providers.
NEW IN THE INDEX
This week’s trending
Fresh skills and tools to put on your radar.
WEEKLY INTAKE
Editorial picks in intake order, not a live star-growth ranking.
- Skills for Real EngineersComposable engineering skills for requirements, test-driven development, architecture review and turning plans into small tasks.
- Cloudflare Security AuditCloudflare’s defensive audit workflow uses reconnaissance, independent validation and source-grounded reports to investigate a codebase.
- ECCA coding-agent harness of skills, rules, hooks and memory, with guided setup for Claude Code and other supported agents.
- Spec KitGitHub’s specification-driven development toolkit organizes requirements, plans and implementation through coding-agent integrations.
- XERJLocal-first code, document and log search with lexical and vector retrieval, exposed to coding agents through MCP.
- PaperclipA self-hosted control plane for coordinating agent teams with roles, budgets, shared work and runtime adapters.
- DeerFlowByteDance’s self-hosted agent harness combines sandboxed tools, memory, skills and subagents for longer tasks.
- OpenSpecA lightweight specification workflow that aligns a proposed change, its design and tasks before implementation.
- Lemo-OpuscarA code-driven film toolkit with 59 documented styles, guides and rendering tools for creating short films with a coding agent.
- MCP for BlenderA community MCP bridge that lets an LLM inspect and change Blender scenes, objects and materials through a local addon.
All entries
367 entries, metadata checked 17 Oct 2026 – 3 Nov 2026
Use case
Productivity121Development89Security analysis69Scientific research65Creative tools25Integrations24Threat detection18Incident response & forensics17Cloud & infrastructure security12Application & supply-chain security10Identity & email security3OT & ICS security3Data / E-commerce1Healthcare1Medical vocabularies1Publisher
alirezarezvani26addyosmani21vercel-labs8phuryn6modelcontextprotocol5github2affaan-m1ahujasid1anthropics1artouc1bytedance1ChromeDevTools1cloudflare1cursor1debpalash1EL4CTEO1Fission-AI1footalytics1garrytan1GhostPawJS1GLips1Graphicle-AI1Graphify-Labs1Jakeschincariol1juliobsz1K-Dense-AI1kanaworksai-hue1Leonxlnx1mattpocock1microsoft1obra1OMOPHub1openai1Pangolin-spg1paperclipai1RayenMlayeh1reason-machines1Saivion1thedotmack1tt-a1i1upstash1xerj-org1ZSeven-W10xS4N4TG06reams0AirsupHQ0alchaincyf0assafkip0calesthio0DietrichGebert0FeedHive0Garinmckayl0heydamir0JuliusBrussee0kepano0latent-spaces0lemomo-ai0mukul9750storytold0theCosmicCrafter0wheresryan220Yuan1z08250Community first · Anthropic official entries follow. Sort applies within each group.
Showing 105 of 367
FootalyticsMCPHosted Footalytics MCP at footalytics.farvor.com/mcp. Graded football predictions and fixtures; Bearer API key or OAuth required. Freemium (5 free calls).Requires paid API keyFreemium (5 free)HostedOAuth or BearerMITmcpListed ≠ audited
SwiftPiecesMCPPublic MCP for searching and installing free animated SwiftUI pieces. No account for the free library; optional Pro is a separate paid server.HostedFree MCP (no account)MIT + Commons ClauseOptional PromcpListed ≠ audited
claude-memPluginPlugin that records what happens in agent sessions, compresses it and adds relevant context back into later sessions.
agent-browserSkillVercel Labs’ browser automation CLI for AI agents (Chrome via CDP, accessibility-tree snapshots) and its discovery skill. Install the CLI, add the skill, then load workflows with agent-browser skills get core.
archifySkilltt-a1i’s Archify agent skill: turns an idea, plan or codebase into a standalone interactive HTML architecture diagram for people to explore and share.
graphifySkillGraphify Labs’ /graphify skill and CLI: turns a folder of code, docs and PDFs into a queryable knowledge graph for coding agents. PyPI package graphifyy (double y).
VoiceStudioSkillAgent skill and MCP endpoint for debpalash/VoiceStudio, an open-source local voice studio: speech, transcription, cloning, dubbing. Needs the app running on your machine.Local dependenciesAGPL source
OMOPHubMCPSearch and map OMOP medical vocabulary concepts with a hosted or local MCP server. Requires an API key; freemium.Requires API keyFreemiumHosted servicemitmcp
pstackSkillLauren Tan’s (poteto) skill pack for rigorous AI-assisted engineering: run setup, then /poteto-mode routes playbooks and the other skills.
DeerFlowAgentByteDance’s self-hosted agent harness combines sandboxed tools, memory, skills and subagents for longer tasks.Local dependenciesExecutes code
GhostPaw CodexSkillLocal memory for AI agents in one SQLite file.Local dependencies
MCP for BlenderMCPA community MCP bridge that lets an LLM inspect and change Blender scenes, objects and materials through a local addon.Local dependenciesExecutes code
OpenSpecSkillA lightweight specification workflow that aligns a proposed change, its design and tasks before implementation.
PaperclipAppA self-hosted control plane for coordinating agent teams with roles, budgets, shared work and runtime adapters.Local dependenciesExecutes code
XERJMCPLocal-first code, document and log search with lexical and vector retrieval, exposed to coding agents through MCP.Local dependencies
Spec KitPluginGitHub’s specification-driven development toolkit organizes requirements, plans and implementation through coding-agent integrations.
ECCPluginA coding-agent harness of skills, rules, hooks and memory, with guided setup for Claude Code and other supported agents.Executes code
Pangolinfo Amazon Data MCPMCPUnofficial MIT bridge to a paid hosted MCP service for e-commerce product, review and category data. Needs a Pangolinfo API key.Requires paid API keyHosted serviceUnofficial
gstackSkillGarry Tan’s Claude Code skill set with role-based tools for planning, design, engineering review, QA, release and docs.
Skills for Real EngineersSkillComposable engineering skills for requirements, test-driven development, architecture review and turning plans into small tasks.
Audio Studio MCPMCPRuns Stable Audio 3 locally on Apple Silicon for music and sound effects, then loops, normalizes and exports the audio.
awwwardsSkillGuides website planning, implementation and browser checks, with a separate jury skill for reviewing the result.
Chrome DevTools MCPMCPOfficial Chrome DevTools MCP server that lets coding agents debug, inspect network traffic and record performance traces in Chrome.
Codex OpenPencilPluginConnects Codex to an editable OpenPencil canvas for inspecting layers, refining layouts and saving native .op designs.
earshotMCPAnalyzes local audio, checks timing against video, and exposes sound measurements and synthesis tools through MCP.
GraphicleMCPIndexes code, documents and agent-session context locally, then serves cited excerpts and code-graph tools over MCP.
Reason NetworkMCPConnects teammates’ coding-agent sessions through a local daemon with peer messaging and MCP team-directory tools.
Context7MCPUpstash’s MCP server and CLI that pull current, version-specific library docs into a coding agent’s context.
Blender Agent SkillsSkillPackages Blender modeling, materials, animation, rendering and asset checks as a collection of agent skills.- code-reviewSkillReviews implementation boundaries for actionable correctness defects, tracing each finding from an agreed contract to the code that violates it.
- dummy-datasetSkillGenerates synthetic datasets from requested columns and constraints, producing test fixtures in formats such as CSV, JSON or SQL.
- intended-vs-implementedSkillCompares a system's intended behavior with its implementation, building an explicit model of expectations to reveal gaps generic scanners may miss.
- shipping-artifactsSkillAssembles the durable documentation needed to review an AI-built application, covering architecture, permissions and operational evidence before shipping.
- sql-queriesSkillDrafts SQL from a question and supplied schema information, targeting supported warehouse dialects while making the query's assumptions visible.
- test-scenariosSkillDerives test scenarios from user stories, recording starting conditions, roles, actions and expected results for systematic quality review.
GitHub MCP ServerMCPGitHub’s official MCP server for working with repositories, issues, pull requests and Actions from an agent.
jevcrawlMCPCrawls a site toward a question and returns source-linked passages, using an external Jev decision model to rank the route.
OpenAI Agents SDKAgentProvides a Python framework for agent workflows that supports multiple model providers.
Agent Skills (Addy Osmani)PluginAddy Osmani’s collection of 25 engineering skills for coding agents, covering spec, planning, testing, review and shipping.- api-and-interface-designSkillTurns API requirements into endpoint and module contracts, with attention to compatibility, error handling and interface boundaries.
- browser-testing-with-devtoolsSkillUses browser observations, DOM inspection and console output to diagnose interface failures and check fixes in Chrome DevTools.High static risk — review source first
- ci-cd-and-automationSkillBuilds delivery pipelines from a project's build and test commands, organizing repeatable checks before environment-specific deployment stages.
- code-review-and-qualitySkillReviews a proposed code change across correctness and maintainability concerns, turning concrete defects into prioritized review findings.
- code-simplificationSkillRefactors working code to make control flow and responsibilities easier to follow while preserving the existing observable behavior.
- constraint-driven-developmentSkillConverts a project's quality expectations into a written contract, so implementation checks use explicit thresholds rather than shifting assumptions.
- context-engineeringSkillOrganizes rules and task context for coding agents, helping a new session start with relevant information and fewer conflicting instructions.High static risk — review source first
- debugging-and-error-recoverySkillInvestigates failed tests, broken builds and unexpected behavior through reproducible evidence, narrowing causes before choosing a repair.High static risk — review source first
- deprecation-and-migrationSkillPlans transitions away from old APIs or implementations, covering compatibility, user migration and the eventual removal of deprecated behavior.
- documentation-and-adrsSkillRecords architectural choices and public-interface changes as documentation and decision records, preserving the reasons future maintainers need.
- doubt-driven-developmentSkillChallenges a technical decision through an independent review of its assumptions, producing objections to resolve before implementation proceeds.
- git-workflow-and-versioningSkillOrganizes branches, commits and conflict resolution for a code change, including splitting a mixed working tree into reviewable units.High static risk — review source first
- incremental-implementationSkillBreaks a multi-file implementation into small working increments, pairing each increment with evidence that its behavior is correct.High static risk — review source first
- observability-and-instrumentationSkillPlans logs, measurements and traces around a feature's production behavior, making failures and operational questions easier to investigate.
- performance-optimizationSkillInvestigates slow frontend, backend and database paths using measurements, then evaluates whether a targeted optimization improves the observed bottleneck.
- planning-and-task-breakdownSkillTurns an agreed specification into ordered implementation tasks, exposing dependencies and verification steps before work begins.
- shipping-and-launchSkillChecks a proposed production launch against operational prerequisites, gathering launch checks and unresolved readiness issues before release.
- source-driven-developmentSkillChecks an implementation approach against official references, connecting technical decisions to documented behavior instead of relying on recalled APIs.
- spec-driven-developmentSkillCreates a specification from a new feature or project idea, making objectives, requirements and acceptance conditions explicit before coding.
- test-driven-developmentSkillUses failing tests, a minimal implementation and subsequent refactoring to develop a behavior change with a repeatable correctness check.
- pi-agentSkillConfigures and extends the Pi terminal coding harness, connecting model providers and local settings to reusable skills and extensions.High static risk — review source first
EverythingMCPExercises MCP features as a reference test server for developers building MCP clients.
FetchMCPRetrieves web content and converts HTML into Markdown for use through MCP.
FilesystemMCPImplements file-system operations through an MCP server built with Node.js.
GitMCPProvides MCP tools for reading, searching and modifying Git repositories.
MemoryMCPStores persistent conversational context using a local knowledge graph.
stop-describing-uiSkillReviews interface text for redundant explanations and replaces them with clearer labels, controls or spacing.
SuperpowersPluginSkills and workflow plugin for coding agents: brainstorm, write a plan, test first, then review the result.- pr-labelingSkillClassifies a skills-CLI pull request from its description and patch, recommending labels while respecting those already assigned.
- triageSkillTriages skills-CLI bugs using reproducible evidence, identifying narrow fixes that merit maintainer attention.
Playwright MCPMCPMicrosoft’s MCP server for driving a real browser through accessibility-tree snapshots instead of screenshots.
replica-skillSkillEleven Claude skills by Jake Schincariol that map public app flows and rebuild features clean-room style, from recon to deploy.
Cloudflare Security AuditSkillCloudflare’s defensive audit workflow uses reconnaissance, independent validation and source-grounded reports to investigate a codebase.
deploy-to-vercelSkillGuides deploying applications and websites to Vercel, including test deployments.
vercel-composition-patternsSkillExplains reusable React composition techniques, including compound components, render props and context.
vercel-react-best-practicesSkillProvides Vercel Engineering guidance for performance in React and Next.js applications.
vercel-react-native-skillsSkillCovers performance practices for React Native and Expo, including lists, animations and native modules.
vercel-react-view-transitionsSkillGuides interface animations using React view transitions, including navigation and shared elements.- adversarial-reviewerSkillSubjects a code change to a deliberately critical review, looking for concrete weaknesses that a familiar implementation perspective may overlook.
- agent-launcher-orchestratorSkillGuides creation and evaluation of a Claude Managed Agent in the user's account, taking an agent brief toward a launchable configuration.High static risk — review source first
- agent-protocolSkillDefines structured messages between executive-role agents, including invocation, isolation and loop-prevention conventions for coordinated work.
- api-design-reviewerSkillReviews REST API designs for inconsistent conventions and breaking changes, turning endpoint definitions into a structured design scorecard.High static risk — review source first
- boost-asio-proSkillSupports asynchronous C++ networking work with Asio, covering event-loop ownership, concurrent handlers and coroutine-based client or server code.
- ci-cd-pipeline-builderSkillUses detected stack information to draft a CI/CD baseline, with repeatable checks and distinct delivery stages for each environment.High static risk — review source first
- code-reviewerSkillAnalyzes code changes across supported languages for complexity, correctness and review risk, producing actionable findings for pull-request review.High static risk — review source first
- coverageSkillExamines a project's tests and coverage information to identify untested behaviors and suggest additional checks with meaningful failure detection.
- engineering-advanced-skillsSkillIndexes an advanced engineering skill collection, helping choose a focused specialist for agent infrastructure, development or operations work.
- evalSkillCompares completed AgentHub runs using a chosen metric or judge, producing ranked results for an experiment rather than new implementations.
- feature-flags-architectSkillPlans feature-flag rollout and retirement from a project's release needs, including kill switches, ownership and stale-flag cleanup.
- git-worktree-managerSkillSets up isolated Git worktrees for concurrent features, coordinating branches, local ports and environment handling for separate development contexts.High static risk — review source first
- hivemindSkillOrganizes isolated coding workers and benchmark comparisons for delegated work, documenting orchestration choices and claimed cost tradeoffs.High static risk — review source first
- loopSkillConfigures an iterative research experiment with a selected interval, defining repeated runs for an existing autoresearch workflow.
- monorepo-navigatorSkillMaps dependencies and package boundaries in a monorepo, helping select affected tests and builds instead of processing every package indiscriminately.
- named-persona-adversarial-reviewSkillReviews code using explicitly named engineering philosophies, comparing concrete implementation choices against documented principles rather than generic checklists.
- pw-reviewSkillReviews Playwright tests for weak assertions and unreliable execution patterns, producing improvements to browser-test quality and maintainability.
- rag-architectSkillDesigns retrieval pipelines around document chunking, embeddings and search evaluation, relating architecture choices to measurable retrieval quality.
- runSkillCoordinates the stages of an AgentHub experiment from baseline through evaluation and merge, using the session's selected competition workflow.
- saas-scaffolderSkillScaffolds a SaaS application's authentication, database, billing and dashboard structure using the source's supported full-stack conventions.High static risk — review source first
- senior-computer-visionSkillSupports computer-vision system design across detection and segmentation tasks, connecting model architecture and training choices to visual-analysis requirements.High static risk — review source first
- senior-fullstackSkillCompares full-stack architecture choices and scaffolds supported application stacks, pairing implementation structure with code-quality assessment.High static risk — review source first
- snowflake-developmentSkillSupports Snowflake SQL and data-pipeline development, including managed table updates, stream/task processing and Cortex-based data features.High static risk — review source first
- sql-database-assistantSkillUses database schemas and a query goal to draft SQL, migrations or optimization suggestions for common database and ORM workflows.High static risk — review source first
- stage-launchSkillConverts an approved managed-agent build sheet into API payloads and a resumable launch sequence using the user's own credentials.High static risk — review source first
- write-a-skillSkillTurns a repeatable task into a structured agent skill with focused instructions, progressive references and the supporting resources it needs.
Framelink MCP for FigmaMCPMCP server that gives coding agents simplified layout and style data from Figma files for design-to-code work.
Claude Code pluginsPluginContains the Claude Code coding tool repository and its plugin examples and documentation.
Choose by workflow
Short guides that compare candidates for one job, with the reference docs they follow.
Most of them compare MCP (Model Context Protocol) servers, for example for Claude Code, Codex or Blender.
Best MCP Servers for Claude CodeModel Context Protocol (MCP) servers give Claude Code extra tools.earshotReason NetworkGraphicleAudio Studio MCPjevcrawlFilesystemGit
MCP for Claude: Best Servers and How to Add ThemShort answer: in the Claude desktop app, add one MCP server for the job in front of you — Filesystem to work with chosen folders, Memory to keep context between chats, Fetch to read web pages.FetchFilesystemMemory
Best MCP Servers for CodexCodex can connect to Model Context Protocol (MCP) servers for extra tools.Reason NetworkGraphicleFilesystemGitMemory
ArtCraft creative applicationsCompare source-linked application hubs for image editing and documents.ArtCraftPhotoCraftWordCraft
Best Blender MCP Servers & Skills (2026)Blender MCP connects an AI client to a running Blender session through the Model Context Protocol, so the model can read your scene and run Python through Blender’s API.Blender Agent Skills
Best Reverse Engineering MCP ServersA reverse engineering MCP server lets an agent drive analysis tools such as Ghidra, IDA Pro, radare2 or JADX through the Model Context Protocol and report what it found with evidence.GhidraMCPida-pro-mcpradare2-mcpreverse-engineering-assistant (ReVa)jadx-ai-mcp
Harness² (harness2) vs OpenCode and Codex CLIHarness² (harness2) is an open-source research framework from Google Research that tunes the harness around an agent while the model stays fixed.
Install Claude Code Skills: Folder, Plugin & CLIChoose one of three ways to install Claude Code skills: copy a complete skill folder, install a plugin from its marketplace, or use npx skills add with Claude Code selected.frontend-designpdfvercel-react-best-practices
Best MCP Servers for Cursor: Files, Git & MemoryChoose a Cursor MCP server for one missing capability: file access, repository history, web fetching or persistent context.FetchFilesystemGitMemory
Claude Code Security: Audit & Hardening SkillsA Claude Code security review starts with an authorized scope and a question to investigate.Cloudflare Security Auditsecurity-and-hardening
Best Codex Skills (2026)Codex can use agent skills: folders with a SKILL.Skills for Real EngineersgstackTaste SkillawwwardsAI Slop DetectorGhostPaw Codexhuashu-motionLemo-OpuscarMachine SkillanatomyBlender Agent Skills
Best Claude Skills for Video EditingPick a Claude skill by the video you need to ship.huashu-motionpapercut-explainerLemo-OpuscarMachine Skill
Questions before you install
Is a listed entry safe to install?
Not automatically. Listed ≠ audited. We pin each entry to a public commit so you can read exactly what you install. Read the SKILL.md or README, check bundled scripts, and limit tool and file access before you enable it.
What is a Claude skill?
A folder with a SKILL.md file that tells an agent how to carry out one task, sometimes with scripts or reference files. Claude Code loads skills from ~/.claude/skills or a project’s .claude/skills folder.
How are skills different from MCP servers?
A skill describes a workflow. An MCP (Model Context Protocol) server is a program that exposes tools or data to a compatible AI client. A project may offer both, but review their permissions and setup separately.
How is the list ordered?
By the date of each entry’s pinned upstream commit, newest first. You can also sort by guide citations or name. We do not have star, download or install counts, so we never show them. How we rank
Are these resources officially endorsed?
No. AgentSkillsDir is an independent directory run by Orbitlane Labs and is not affiliated with any listed publisher. Source links and pinned revisions help you inspect each entry; they are not a certification.
How do I suggest a resource?
Use the submission form to share a public project link, or email submit@agentskillsdir.com. Listing is free and not guaranteed. Submit
What are the best Claude skills and MCP servers to start with?
There is no single best list; pick by the job you need done. AgentSkillsDir lists source-checked Claude skills and MCP servers, each pinned to a public commit, and orders them by the date of that commit. Our workflow guides compare candidates for one task: for Claude Code, the Filesystem and Git MCP servers are common starting points. Read each entry’s SKILL.md or README before you install it.
How do I install a Claude skill from GitHub?
A skill is a folder that contains a SKILL.md file. Copy that folder from the repository into ~/.claude/skills/ to use it in every project, or into .claude/skills/ inside one project. Claude Code picks up skills from those locations. Each entry on AgentSkillsDir links to the pinned commit and shows a setup snippet you can copy.
How do I add an MCP server to Claude Code?
Run claude mcp add with a name and the server’s launch command, for example: claude mcp add fetch -- uvx mcp-server-fetch. Then run claude mcp list to confirm it is registered. For Claude Desktop, add the server under mcpServers in claude_desktop_config.json and restart the app. Use the launch command from the server’s own README.
What goes in a SKILL.md file, and how is a skill folder structured?
SKILL.md starts with YAML front matter that gives the skill a name and a description of when to use it, followed by Markdown instructions for the agent. The folder can also hold scripts, reference documents or templates that the instructions point to. Keep the folder name and the name field consistent, and review any bundled scripts before you enable a skill.
Is AgentSkillsDir a skills marketplace, and can I submit my MCP server?
AgentSkillsDir is a free, independent directory, not a store: nothing is sold here, and every entry links to its public source repository. You can submit a public GitHub link for a Claude skill, MCP server, plugin or agent through the submission page or by emailing submit@agentskillsdir.com. Submissions are reviewed before they are listed, and listing is not guaranteed. Submit