Skills
Instructions, scripts, and references that teach an agent how to complete a task.
Two independent plugin ecosystems
DeepSeek Harness and Agent Plugins 1.0 use different formats and installation paths. Start with the ecosystem you actually run.
Discover community plugins for coding, automation, and AI agent workflows—with source and installation context before you adopt.
Browse packages built around plugin.json, reusable Skills, and optional MCP configuration.
Every plugin is verified as compliant with Agent Plugins 1.0.0.
Editor's shortlist
A portable Agent Plugin for retrieving current, version-specific library documentation and code examples through Context7 MCP.
Expert skills for building n8n workflows with the n8n-mcp MCP server
An Agent Plugin package connecting MCP clients to Exa for web search, code search, content fetching, crawling, and multi-step research.
A portable design-lead workflow for building production-ready interfaces that feel specific to the product instead of assembled from defaults.
A portable Playwright workflow for verifying local web apps, debugging UI behavior, and capturing visual evidence.
A portable test-driven workflow for authoring, validating, evaluating, packaging, and improving Agent Skills.
A portable end-to-end guide for planning, implementing, testing, and evaluating MCP servers.
A portable art-direction workflow for original posters, visual compositions, and polished PNG or PDF output.
A portable creative-coding workflow for original p5.js generative art and interactive exploration.
A portable brand-styling workflow for consistent colors, typography, hierarchy, and visual accents across artifacts.
A portable implementation guide for Claude API applications across Python, TypeScript, Java, Go, C#, PHP, Ruby, and curl.
A portable writing workflow for concise, audience-aware internal updates in recurring company formats.
A portable theming workflow with ten curated palettes, font pairings, a visual showcase, and support for custom themes.
A portable toolkit for complex interactive web artifacts that need components, state, routing, and reliable bundling.
A portable Python toolkit for creating polished, compact animated GIFs that meet common Slack emoji and message constraints.
End-to-end guidance for full-stack applications, from focused context, requirements, and planning through architecture, data, APIs, source-verified implementation, React and frontend delivery, and launch.
Recently reviewed
A portable Agent Plugin for retrieving current, version-specific library documentation and code examples through Context7 MCP.
Agent Skills for academic paper work: training-pipeline diagnosis, drafting and review, publication figures, and getting a manuscript through a journal — source, packet, and response to reviewers.
Send transactional emails through the Resend remote MCP server with OAuth.
Structured data manipulation with jq, yq, dasel, qsv, and mlr for JSON, JSONL, YAML, TOML, XML, and CSV processing
Configure Claude Code Remote Control server mode and choose between the current computer, Claude web, an existing host, and new paid hosting.
Safely configure and verify a Terminalbox SSH host for the official Codex app.
Configure GitHub Copilot CLI remote control and choose between local, GitHub cloud sandbox, and persistent-host execution.
Read-only repository audit for choosing between a rebuildable remote-agent task and a persistent development workspace.
Independent discovery hub / Agent Plugins 1.0
Explore Agent Plugins through the portable package standard, then compare public Plugin and Skill repositories using transparent GitHub activity.
The standard defines the package. Clients still control installation, permissions, and runtime behavior.
The simple model
Agent Plugins give reusable agent capabilities a predictable home. The package can contain instructions, tool connections, or both.
Read the official specificationInstructions, scripts, and references that teach an agent how to complete a task.
Server definitions that connect an agent to data, APIs, and external systems.
The portable directory that gives Skills and MCP configuration a shared, predictable structure.
Start from the source
The 1.0 ecosystem is new. We separate official references, conformant packages, and client-specific formats instead of inflating the catalog.
agentplugins / agent-plugins-example
The canonical example demonstrates a valid manifest and a migration Skill. It is a reference package—not an official marketplace pick.
Agent Plugins 1.0 is kept separate from client-only formats.
Every ranked repository points back to a source you can inspect.
Stars are attention—not installs, compatibility, or security.
Repository rankings
Compare Plugin and Skill repositories by GitHub stars and transparent seven-day growth snapshots.
View rankingsDaily snapshots start with the dataset.
No estimated growth and no made-up install counts. Rankings appear when verified source data is ready.
Learn the standard
Understand the open 1.0 package format and what makes a plugin portable.
See where instructions, tools, and the portable package fit together.
Inspect, install, verify, update, and remove plugins without assuming one universal client command.
Agent Plugins guide
Learn how portable agent capabilities are structured, what to check before installation, and where Skills and MCP fit into the package.
Agent Plugin = Skills + MCP
Instructions explain the work. Tool connections make the work possible. The package gives both a portable structure.
Agent Plugins are portable packages for reusable agent capabilities. Instead of scattering prompts, setup notes, scripts, and tool configuration across several repositories, a plugin gives them a predictable home. A manifest identifies the package, while optional components describe what an agent should know and which external systems it may use. This structure helps people inspect a capability before deciding whether it belongs in their workflow.
An Agent Plugin is not a model and it is not an autonomous agent. It is a package that a compatible client can read. The client still decides how installation works, which permissions are allowed, and how tools run. That separation is important: the package describes a capability, while the client controls its execution and security boundaries.
Skills provide the instructional layer. A Skill can include task guidance, repeatable procedures, scripts, examples, and reference material. It teaches an agent how to approach a specific job with more consistency. MCP provides the connection layer. An MCP server can expose data, APIs, files, databases, or actions that the agent needs in order to complete that job.
Agent Plugins bring those layers together without making every package identical. One Agent Plugin may contain only a focused Skill, another may configure an MCP connection, and a larger package may include both. The manifest provides the shared entry point, so clients and users can understand the package contents before enabling them.
Teams use Agent Plugins to make useful agent behavior easier to share, review, and maintain. A portable package can reduce repeated setup work, keep operational knowledge close to the scripts that support it, and make changes visible through source control. It also creates a clearer review boundary than copying instructions from an unknown page directly into an agent session.
Portability does not guarantee universal compatibility. Different clients may support different installation flows, runtime options, or permission models. Treat every Agent Plugin as source material that must be evaluated for your chosen client and environment. Repository activity can help with discovery, but stars are not a substitute for code review, documentation, or security checks.
Agent Plugins can support focused workflows without locking useful knowledge inside one conversation or one client.
Package repeatable discovery, analysis, and source-checking workflows.
Combine coding instructions with tools for repositories and services.
Share documented procedures for support, reporting, and routine tasks.
Start with the source repository and inspect the package before you install it. Useful Agent Plugins explain their purpose, list their components, document external access, and keep instructions narrow enough to review. For any Agent Plugin that connects to outside systems, check credentials, requested permissions, network access, and the actions its tools can perform.
Use the learning guides to understand the package model, then use the rankings to find public repositories worth inspecting. This gives Agent Plugins a useful discovery path without confusing popularity with safety, quality, or client support.
Frequently asked questions
An Agent Plugin is a portable package that describes a reusable agent capability. It can include a manifest, task instructions, supporting files, scripts, and connections to external tools.
No. A Skill is the instructional part that teaches an agent how to perform work. A plugin is the wider package that can contain Skills, MCP configuration, or both.
No. MCP is optional. An Agent Plugin can contain only a Skill when the task needs instructions but no external tool connection. Other packages use MCP to reach APIs, data, or services.
Inspect the source, manifest, instructions, scripts, permissions, credentials, and network access. Confirm that your client supports the package, and never treat repository popularity as proof of safety.