Workflow & Planning
Add structure to how the agent plans, reviews and tests, so it stops jumping straight into code.
superpowers skills Software development methodology built from composable skills: the agent asks what you want, writes a spec and implementation plan, then works through tasks with subagents, TDD and review. gstack slash commands Garry Tan's set of slash-command skills that turn Claude Code into a virtual engineering team: office hours, review, QA in a real browser, security audits and a ship command that opens the PR. ponytail plugin Plugin that makes the agent ask whether code needs to exist at all and pick the smallest complete change before writing anything. vibe-wise plugin Claude Code plugin that asks for your approach first, helps you weigh tradeoffs, and writes code only after you approve the design, then explains what it changed. mattpocock/skills skills Mattpocock's everyday engineering skills, including grill-me (relentless plan interview), grill-with-docs (builds glossary and ADRs) and improve-codebase-architecture. addyosmani/agent-skills skills Production-grade engineering skills covering define, plan, build, verify, review and ship, driven by nine slash commands such as /spec, /plan and /build. planning-with-files skill Skill that keeps task_plan.md, findings.md and progress.md on disk, with hooks that reinject the plan so it survives /clear and context resets. beads plugin An open-source distributed graph issue tracker (bd) that gives Claude Code persistent task memory. skills skill An MIT-licensed set of Claude Code skills from HumanLayer for research, planning, implementation and pull request workflows. Some skills create pull requests. elephant-goldfish-model plugin Plugin for designing and reviewing software specs: an "elephant" drafts with your project context, then "goldfish" reviewers check it with fresh eyes. Needs a session that can start subagents. Jeffallan/claude-skills plugin 67 skills for languages, frameworks, DevOps and security, plus 9 epic workflow commands that need an Atlassian MCP server. n8n-mcp MCP MCP server that gives Claude n8n node docs, templates and workflow tools. Telemetry is on by default (opt-out); with your n8n key it can edit and run workflows.Multi-Agent & Orchestration
Run several agents or models at once and keep them coordinated.
ruflo plugin Agent meta-harness for Claude Code that adds multi-agent swarms, workflows, memory and cost tracking through installable plugins. Large project; some plugins can call cloud agents that cost API money. oh-my-claudecode plugin Multi-agent orchestration for Claude Code with team-style workflows and commands. Runs many sessions, so token use is high. claude-octopus plugin Runs Claude, Codex, Gemini and other models on the same task and cross-checks them. Setup installs the Codex and Qwen CLIs globally; other providers need their own keys and can cost money. orca app Desktop app that runs Claude Code and other coding agents in parallel, each in its own git worktree. MIT licensed; collects anonymous usage data unless you opt out. CLI-Anything plugin Claude Code plugin marketplace that generates agent-usable command-line interfaces for existing software so Claude can drive it. Runs commands on your machine.Setup, Rules & Guides
Configure Claude Code, learn how it works, and drop in rules files.
andrej-karpathy-skills CLAUDE.md A single CLAUDE.md with four principles (think before coding, simplicity first, surgical changes, goal-driven execution) based on Karpathy's observations on LLM coding mistakes. claude-codex-settings plugins + CLAUDE.md Marketplace of 40+ plugins plus a shared CLAUDE.md for Claude Code and Codex. Some plugins can change cloud infrastructure; others need third-party keys. claude-code-best-practice list Guide to Claude Code concepts, workflows, skill and agent collections, and tips. claude-code-guide guide + skills Community guide to Claude Code setup, commands, hooks, MCP and agents, with a 70+ skill and agent pack. everything-claude-code (ECC) plugin MIT-licensed Claude Code plugin with 71 agents, 314 skills, hooks, rules and memory for planning, testing, review and verification. cc-switch app Desktop app that switches API providers and manages MCP servers, skills and prompts for Claude Code, Claude Desktop and other agents. Rewrites your agent config files and holds API keys; the README carries paid relay-service ads. learn-claude-code tutorial 17-lesson Python tutorial on building a Claude Code-style agent harness from scratch. Needs your own Anthropic API key; the lesson agents run shell commands. MCPJungle MCP Self-hosted MCP gateway: register your MCP servers once and connect Claude to a single endpoint. Holds the credentials for those servers. i-have-adhd skill Small skill that makes Claude lead with the next action, number its steps and drop filler, installable as a Claude Code plugin.Code Review & Safety
Review code and stop the agent from doing damage.
open-code-review CLI + plugin Reads your git diff and has an LLM write line-level review comments; has a Claude Code plugin. Needs your own LLM key unless you use delegate mode, and sends the diff to that model. cc-safety-net hook Hook that blocks destructive shell commands and reads of secrets such as SSH keys and `.env` before the agent runs them. Not a sandbox. SkillSpector skill An Apache-2.0 security scanner that checks agent skills, including Claude Code skills, for prompt injection, data exfiltration and supply-chain risks before install.Token Saving & Cost
Spend fewer tokens or dollars, and see where they go.
caveman skill + CLI Skill and CLI that make the agent answer tersely in caveman style and shrink logs and tool output before the model reads them, to cut tokens. TokenEconomizer CLI Microsoft Research tool that learns from past sessions in a repository to cut coding-agent cost (micro-skills, tool pruning, failure-loop intervention); reports 18.4-30.9% lower mean cost on SWE-Bench. 9router router Local router that connects Claude Code and other coding tools to 40+ AI providers with auto-fallback and token trimming of tool output. codegraph MCP server Local code knowledge graph exposed over MCP that auto-syncs on every file change. gemini-mcp-tool MCP server MCP server that lets Claude call the Gemini CLI to use Gemini's large context window for analyzing big files. CodeBurn CLI Free, open-source, local-first tracker of AI coding token usage and cost across 42 tools, broken down by model, project and task. context-mode plugin A free local Claude Code plugin and MCP server that keeps tool output out of the context window using a sandbox and SQLite full-text search. rtk app Rust CLI proxy that compresses command output before it reaches Claude to cut token use. Installs a Claude Code hook; the CLI is Apache-2.0 but its engine folder is BUSL-1.1 (source-available). headroom app Compression layer that shrinks tool output, logs and files before they reach Claude, usable as a wrapper, proxy or MCP server. Telemetry can be turned off; a paid Teams offering is separate. token-optimizer plugin Plugin that finds wasted tokens, checkpoints sessions across compaction and shows cost on a local dashboard. PolyForm Noncommercial: free for personal use and businesses under 5 people.Memory & Context
Stop the agent forgetting everything between sessions. Obsidian tools are under [Productivity](#notes--knowledge).
engram MCP server + CLI Persistent memory for Claude: a single Go binary with SQLite and FTS5, exposed through MCP, CLI, HTTP API and a TUI, usable with Claude Code and other MCP agents. claude-mem plugin Plugin that automatically captures tool usage during sessions, generates semantic summaries and injects relevant context into future sessions. base CLI basemode: a Rust binary (base) that maps your workspace into a knowledge graph and wires it into Claude Code's hooks to brief the agent with relevant memory. codegraph MCP server Local code knowledge graph exposed over MCP that auto-syncs on every file change. Understand-Anything plugin Claude Code plugin that analyzes a project with a multi-agent pipeline, builds a knowledge graph of files, functions and dependencies, and offers an interactive dashboard to explore it. graphify skill An open-source /graphify skill for Claude Code that turns code, docs, schemas and PDFs into a queryable knowledge graph. mempalace MCP A free, local-first memory system that stores conversation history verbatim and retrieves it by semantic search through Claude Code hooks and an MCP server. GitNexus MCP Indexes a repo into a knowledge graph and exposes it to Claude Code through MCP, skills and hooks. Free for noncommercial use only (PolyForm Noncommercial); commercial use needs a paid licence. ai-memory MCP Shared long-term memory layer that lets Claude Code sessions keep and hand off context across tools. Stores session memory locally. floorplan plugin Claude Code plugin that has Claude read your codebase and draw an interactive plain-English map of its parts, files and connections. Runs locally on your existing Claude plan. codebase-memory-mcp MCP Indexes your repo into a local knowledge graph so Claude can query structure instead of re-reading files. The installer runs a curl or PowerShell script. serena MCP Toolkit that gives Claude IDE-style semantic code search, editing and refactoring. GPL-3.0; an optional JetBrains plugin is paid. book-to-skill skill Turns a technical book PDF into a chaptered skill Claude can consult on demand. Don't redistribute skills made from copyrighted books. cognee plugin Memory layer that turns documents, code and chats into a knowledge graph agents can search across sessions; installs as a Claude Code plugin and runs locally. Using an LLM key sends data to that provider; Cognee Cloud is an optional paid service.Browser & Computer Control
Let the agent read, search and browse the web, or drive a desktop app or a phone.
Agent-Reach skill + CLI Skill and CLI that give an agent read and search access to platforms such as YouTube and Bilibili subtitles, GitHub, V2EX, web pages, Xiaohongshu and Twitter/X (experimental, needs login). playwright-mcp MCP server Microsoft's MCP server that gives an LLM browser automation through Playwright, using structured accessibility snapshots instead of screenshots. agent-browser CLI Native Rust CLI that lets an agent drive a persistent browser from the shell: snapshot a page as text with element refs, then click, fill and read again. obscura CLI Open-source headless browser engine written in Rust for AI agents and scraping; runs JavaScript via V8, speaks Chrome DevTools Protocol, and can replace headless Chrome with Puppeteer or Playwright. BrowserMCP MCP MCP server plus a Chrome extension that lets Claude control your own browser. crawl4ai MCP Open-source web crawler and scraper with a hosted MCP server; the hosted cloud is optional, the core is self-hostable. stagehand SDK MIT-licensed SDK for building browser agents that navigate sites, act on pages and extract data. Needs your own LLM API key (Anthropic for Claude); the Browserbase cloud is optional and paid. chrome-devtools-mcp MCP Google's MCP server that lets Claude debug, automate and profile a live Chrome browser. Usage stats are on by default (opt out with a flag) and it exposes page content to the agent. mobile-mcp MCP MCP server that lets Claude control iOS and Android simulators, emulators and real devices. Telemetry is on by default; optional paid cloud devices. agent-desktop CLI Apache-2.0 CLI from Vercel Labs that lets an AI agent see and operate desktop apps on macOS, Windows and Linux through the accessibility tree. Works with any agent that can run shell commands; it doesn't mention Claude by name. Snapshots can contain private desktop content. cua app Open-source infrastructure for computer-use agents (desktop-app driver, local VMs, sandboxes), with a Claude Code integration. Freemium: the hosted Spaces app has paid tiers. The agents control desktop apps, so review before running.Docs & Databases
Give the agent current documentation and a way to query your databases.
context7 MCP server + plugin Pulls up-to-date, version-specific library documentation and code examples into the agent's context so generated code matches the library versions you use. pg-aiguide plugin + MCP Postgres best-practice skills and search over the Postgres and TimescaleDB manuals. The MCP is hosted by TigerData, so queries go to them. dbhub MCP A minimal, token-efficient MCP server that lets Claude explore and query PostgreSQL, MySQL, MariaDB, SQL Server, Oracle and SQLite databases. Runs SQL against the databases you point it at; read-only mode is available.