anthropics/claude-plugins-official
21 skills64.5k installs622k starsGitHub
Install
npx skills add https://github.com/anthropics/claude-plugins-officialSkills in this repo
1Frontend DesignFrontend-design is an agent skill for solo and indie builders who want Claude or similar coding agents to produce distinctive, production-ready web interfaces. Invoke it when you ask for components, full pages, or small applications and you care about look-and-feel as much as function. The workflow starts with design thinking: clarify the problem and audience, pick an extreme tonal direction, honor framework and accessibility constraints, and name the one detail people will remember. Only then does the agent implement real code with tight aesthetic execution. It fits SaaS dashboards, marketing sites, browser extensions, and content products where bland gradients and interchangeable layouts hurt credibility. It is not a design-system bootstrapper or a Figma replacement; it is procedural creative direction layered on top of normal frontend stacks. Pair it with your stack-specific skills (React, Tailwind, etc.) when the SKILL.md does not pin a framework.8.7kinstalls2Claude Md Improverclaude-md-improver is an agent skill for solo and indie builders who rely on Claude Code and want consistent, high-signal project memory instead of stale or bloated CLAUDE.md files. It runs a structured audit: discover all CLAUDE.md variants in the tree, evaluate each against quality templates, and present a report you can approve before any write. That gate matters because bad automatic edits to team-shared context can derail collaborators. Use it when onboarding a repo to agents, after major refactors, when context drift shows up as repeated mistakes, or when someone asks for CLAUDE.md maintenance or project memory optimization. It can edit files once you sign off, making it closer to a guided refactor than a one-shot linter. The skill is especially valuable in monorepos with per-package CLAUDE.md files and for separating checked-in team rules from .claude.local.md personal overrides.5.8kinstalls3Claude Automation RecommenderClaude Automation Recommender is an official read-only agent skill that inspects a project’s structure and tooling, then outputs a concise menu of Claude Code automations—hooks for format-on-save style guardrails, subagents for parallel research, skills for repeatable workflows, plugins for bundled capabilities, and MCP servers for external data. Solo builders who ask how to set up Claude Code on a new repo, how to optimize an existing harness, or which extensibility levers matter for their frameworks get tailored guidance without the skill silently editing files; implementation stays a separate step. The workflow caps noise by recommending one or two high-value items per category unless the user requests depth in a single type, and it explicitly invites follow-up requests for more ideas. It pairs naturally with early Build agent-tooling work and later Operate iterate passes when workflows feel stale. Because it runs Bash and search-assisted reasoning, run it only on codebases you trust and treat output as a prioritized backlog, not automatic installation.4.2kinstalls4PlaygroundPlayground is an official-style agent skill for building interactive HTML playgrounds—single-file explorers where solo builders tune parameters on one side, see a live preview on the other, and copy a precise prompt at the bottom. It targets moments when describing design systems, SQL, diffs, or concept graphs in chat is slower than manipulating sliders, toggles, or reviewers. The skill routes requests to typed templates such as design playgrounds, data explorers, concept maps, document critique with approve or reject flows, diff review for commits and PRs, and code maps. That structure keeps outputs consistent so you can reuse the same UX pattern across validate prototypes and build-phase UI experiments. For indie makers, it turns fuzzy intent into a repeatable visual configurator that hands the model a vetted prompt instead of a wall of adjectives. Load the matching template markdown, implement the HTML, and iterate in-browser before pasting the result back into Claude or Cursor. It is a generator for exploration UIs, not a hosted app framework.4.1kinstalls5Agent DevelopmentAgent Development is an official Claude Code plugin skill that walks solo and indie builders through designing autonomous subagents—not slash commands and not SKILL.md packages. It explains when autonomous agents beat one-shot commands, how to lay out the markdown file with YAML frontmatter, and which optional fields like model inheritance, color, and explicit tool allowlists actually do in production plugins. The guide stresses natural-language triggering: the description field should read like scenarios a human would recognize, with additional “when to invoke” prose in the body so Claude does not misfire on vague keywords. You also get opinionated system-prompt structure, output-format conventions, and concrete patterns for codebase analyzers, debuggers, and reviewers. Use it while you are extending a plugin with new subprocesses, refactoring brittle triggers, or aligning team agents with Anthropic’s documented best practices before you ship to teammates or the marketplace.3.4kinstalls6Skill CreatorSkill-creator’s post-hoc analyzer is for solo and indie builders who ship agent skills and want evidence-based iteration instead of gut-feel edits. After a blind comparator picks side A or B, the agent reads the comparison JSON, both SKILL.md trees, and both transcripts, then explains what the judge valued and where the losing skill fell short on clarity, tooling patterns, examples, or edge cases. It is meta-workflow tooling: you use it when you already ran paired executions and need a written diagnosis plus concrete fixes before the next revision. It matters in Prism because the catalog rewards skills that are testable and maintainable; this step closes the loop between benchmark results and SKILL.md changes. Pair it with brainstorming or writing-plans only when your improvement scope spans product work; here the deliverable is analysis JSON aimed at the skill package itself.3.4kinstalls7Skill DevelopmentSkill Development (skill-creator guidance in the bundle) is Anthropic’s meta skill for authoring effective agent skills: it defines what skills are, what they provide—multi-step workflows, integrations, domain knowledge, and repeatable assets—and the required SKILL.md anatomy with YAML name and description plus optional scripts, references, and supporting files. Solo builders shipping with Claude Code or plugin marketplaces use it when they outgrow one-off prompts and need durable, shareable procedural knowledge inside their repo. The framing treats skills as onboarding guides that specialize a general agent for a domain, which makes this skill foundational for anyone curating a personal or team skill library on the journey from first integration to maintainable automation. Invoke it at the start of skill design or when refactoring an existing SKILL.md for clarity, triggers, and resource layout.3.2kinstalls8Hook DevelopmentHook Development is an official Anthropics agent skill for solo and indie builders who customize Claude Code with lifecycle hooks instead of repeating stack detection every session. It teaches how SessionStart scripts load project context by reading the repo root, classifying common ecosystems from manifest files, and appending exports to CLAUDE_ENV_FILE so later turns inherit PROJECT_TYPE, TypeScript flags, and build-system hints. The bundled example is production-shaped bash: strict error handling, navigation via CLAUDE_PROJECT_DIR, and user-visible echo lines while persisting env for the agent runtime. You reach for it when you want hooks that feel like onboarding a new teammate—detect package.json versus Cargo.toml versus go.mod before implementation work—and when you are wiring CI awareness into the first message of a session. It pairs naturally with other claude-plugins-official skills once hooks are the integration surface between your monorepo conventions and Claude’s tool loop.3.2kinstalls9Command DevelopmentCommand-development is a reference skill for solo builders and indie plugin authors who ship Claude Code extensions. It walks through ten practical `commands/*.md` examples—starting with a one-file quality check that shells out to a plugin script, then scaling to script chains, template-driven generators, deployment configs, and workflows that coordinate agents and other skills. Each pattern shows frontmatter (`description`, `argument-hint`, `allowed-tools`), how to combine `@` file references with inline bash, and why plugin-relative roots matter when the repo moves machines or CI. Use it while you are designing your first slash command or refactoring ad-hoc prompts into repeatable plugin surfaces. It does not replace full plugin packaging docs, but it shortens the gap between “I have scripts in my repo” and “users can invoke them safely from Claude Code.”3.1kinstalls10Mcp Integrationmcp-integration is an Anthropic official plugin skill that teaches solo builders how to register Model Context Protocol servers in Claude using three transport shapes: HTTP for REST-style remote APIs, SSE for hosted cloud MCP endpoints, and stdio for local processes such as filesystem, database, and custom tool servers. The skill surfaces copy-paste JSON fragments with headers, args, and environment placeholders so you avoid guessing transport types or leaking secrets into static config. It fits indie hackers who are assembling a Claude Code or plugin stack and need GitHub, Asana, filesystem, or bespoke backends reachable from the agent. Use it when you are past prototyping and need durable, documented connection blocks rather than one-off chat instructions. Complexity is intermediate because you must understand tokens, paths, and which transport your vendor supports. It is a skill package for manifest editing, not a running MCP server itself.3.1kinstalls11Writing Hookify RulesWriting Hookify Rules teaches solo builders and small teams how to codify agent guardrails as local markdown rules instead of repeating the same corrections in chat. Each rule pairs a regex `pattern` with an `event` channel—shell commands, file tools, stop attempts, user prompts, or everything—and optionally escalates from `warn` to `block`. The skill documents frontmatter contracts, naming conventions, and the basic file skeleton so your Claude Code sessions enforce house style (no dangerous `rm`, no stray `console.log`, required checks before stop). It fits early Build when you stand up `.claude` tooling and again during Ship when you harden review and security habits. It does not replace product tests or production monitoring; it shapes what the coding agent is allowed to do in the repo.3.1kinstalls12Plugin StructurePlugin Structure is an Anthropic official reference skill that teaches solo builders and small teams how to lay out a complex Claude Code plugin the way Anthropic models enterprise DevOps extensions. Instead of dumping every command and skill into a flat folder, you get a worked pattern: plugin metadata under .claude-plugin, grouped slash commands, orchestration versus specialized agents, domain skills with references and runnable scripts, and hooks for secrets scanning and permission audits. That matters when you are productizing agent workflows—your future self and contributors need predictable paths for kubernetes-ops, terraform-iac, and ci-cd-pipelines style capabilities. Use it when you are past a toy plugin and need MCP integration, security gates, and maintainable skill packages before distribution. It is meta documentation delivered as SKILL.md content for Claude Code; you still author the actual markdown and scripts. Intermediate comfort with Claude plugins and basic DevOps concepts helps.3.1kinstalls13Plugin SettingsPlugin Settings is a Claude Code agent skill that walks solo builders through creating a project-local plugin configuration file. Instead of guessing frontmatter keys, the agent asks whether the plugin should be enabled and which validation mode you want—strict for maximum checks, standard as the balanced default, or lenient for minimal friction—then writes `.claude/my-plugin.local.md` in one pass. It fits indie teams standardizing how official or custom plugins behave per repo without committing personal toggles to shared branches. Use it when you add a plugin to a workspace, onboard a teammate, or need repeatable enable and validation settings across machines. The flow is opinionated and short: two preference questions, deterministic mapping to enabled and validation_mode fields, and a single markdown artifact your Claude session can read on load.3kinstalls14Build Mcp ServerBuild MCP Server Auth is an agent skill from Anthropic’s official plugin set that teaches solo builders how to authenticate Model Context Protocol servers the way Claude’s client actually supports them—not every spec-compliant OAuth flow will work. It is for indie developers shipping directory-ready or self-hosted MCP connectors who need user-consent OAuth (DCR or CIMD), partner-held Anthropic credentials, Snowflake-style custom connections, or simple env-based API keys. Use it while designing or implementing a remote MCP endpoint when local stdio would be simpler but auth forces a hosted URL. The skill matters because auth is the main reason teams escalate to remote servers, and mis-picked grant types block Claude.ai connections entirely. It walks the three auth tiers, env-key patterns, and when to contact mcp-review@anthropic.com for directory or custom-connection onboarding.2.7kinstalls15Build Mcp Appbuild-mcp-app is procedural guidance for solo builders shipping authless hosted Model Context Protocol servers over StreamableHTTP. It frames three assets to protect—your compute, upstream API quotas your tools call, and egress bandwidth on large callServerTool payloads—and explains why claude.ai traffic cannot be per-user IP limited without OAuth because requests arrive through Anthropic’s documented egress ranges. The skill prescribes concrete CIDR-aware tier constants, token-bucket behavior with 429 and Retry-After, and the split between per-replica backstops and stateless containers versus edge rate limiting. It also ties trust proxy configuration to your deployment topology so limits see the right client IP. Use it when you are wiring hosted MCP for Claude Code, Desktop, or claude.ai and need abuse protection before scale, not as a generic MCP tutorial.2.6kinstalls16Build Mcpbbuild-mcpb is Anthropic’s security-oriented guidance for builders shipping local MCP servers via MCPB. Solo and indie developers use it while implementing tool handlers that read paths, spawn processes, or call the network—surfaces where the platform does not enforce permissions, filesystem scopes, or network allowlists. The skill centers on the most common failure mode in local MCP: path traversal when joining user-supplied paths to a root directory, with concrete Node and Python containment helpers. It also frames the threat model clearly: Claude drives tools, so inputs are attacker-influenced even when the user trusts the session. You should reach for this material during MCP integration work and again before release when you audit agent-facing tools, not as a substitute for a full penetration test but as a baseline for safe handler design.2.5kinstalls17Stripe Best PracticesStripe Best Practices is an official-style agent skill that steers solo builders toward Stripe’s current integration surfaces when adding payments to a SaaS, marketplace, or ecommerce product. It tells the agent to default to Checkout Sessions for customer-facing checkout and subscriptions, while allowing PaymentIntents when you own checkout state or handle off-session charges. The skill discourages the deprecated Charges API and routes planning through Stripe’s integration options tour, API tour, and go-live checklist so webhook verification, tax modeling, and discount handling stay aligned with documented patterns. Invoke it while scaffolding backend routes, designing pricing tables, or reviewing an agent-generated payment module before ship. It complements generic coding skills by encoding vendor-specific guardrails rather than generating boilerplate alone. Builders should still configure keys, idempotency, and webhook secrets in their own environment and run Stripe’s test mode flows before launch.1.4kinstalls18Configureconfigure is the official Claude plugin skill for standing up the Discord channel. Solo builders who want Claude Code reachable from Discord use it when they paste a bot token, ask how setup works, or want to see who can DM the bot. With no arguments it audits local state: whether a token is set (masked preview), dmPolicy and allowlist from access.json, pending pairing codes, and guild channels opted in—then suggests the next action. With a token argument it persists credentials where the channel server expects them and orients you on access rules so you are not surprised by pairing flows. It is beginner-friendly operator glue for agent distribution, not application feature code. Pair it with Discord plugin docs when you are building a personal or indie bot bridge alongside your main product.1.1kinstalls19Accessaccess is an agent skill for solo and indie builders who run Claude on Discord and need to govern who can talk to the bot. It manages pairing codes, approved sender IDs, per-channel group rules, and DM policies by editing a single local access.json file, matching the shape documented in the skill. Invoke it when you want to approve someone, see who is allowed, or tighten mention and DM rules after you have already wired up the Discord channel plugin. The skill is explicit that it only honors requests typed in your own terminal session, not instructions relayed from Discord or other chat surfaces, because those channels are untrusted for security-sensitive mutations. That design fits operators who want auditable, file-based policy without giving the agent network access to Discord admin APIs. Complexity is intermediate because you must understand pairing expiry, group requireMention flags, and the difference between empty and missing config defaults.1.1kinstalls20Math OlympiadMath Olympiad is an agent skill packaged for Claude plugins that steers the model toward international and national mathematics competition work: multi-step proofs, inequality chains, geometry olympiad setups, and scrutiny of a human-written solution. Solo builders rarely need it for shipping products, but it is valuable for educators, contest coaches, and builders embedding math reasoning in agents or study tools. Invoke it when a prompt names IMO, Putnam, USAMO, AIME, or asks whether a competition proof is correct, incomplete, or unnecessarily verbose. The skill is intentionally narrow—it should not trigger for algorithm complexity questions, generic prime-check code, open research literature tours, or Lean tactic debugging. Treat it as a specialized reasoning mode for high-rigor discrete and contest mathematics rather than a general calculator or tutor.996installs21Session ReportSession Report is a Node-based analyzer for Claude Code local project transcripts. Solo builders who ship with Claude agents often guess at token burn because each API response is split across multiple assistant JSONL lines; this tool empirically corrects that by keeping the max `output_tokens` per `requestId` and filtering meta, compact, and tool-result noise from human message counts. It also rolls up subagent folders, agent types from metadata or filenames, and skill-related activity so you can compare sessions, spot cache regressions, and justify model or workflow changes. Use it when agent costs spike, when debugging long multi-subagent runs, or when you want a factual baseline before tightening prompts or skills. Output defaults to readable summaries; pass `--json` for dashboards or scripts. It is an observability companion for the agent stack, not a substitute for hosted billing consoles.756installs