Skills
96,770 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Design and build a CLI that an AI agent can operate safely and a human can supervise. Use when the user wants to build a CLI, wrap an API in a command-line tool, build a CRUD or local-first tool over a database or filesystem the user already owns, add --json or --dry-run to an existing CLI, design a trust ladder or approval gate for risky commands, wrap an async API so agents do not write their own poll loop, or decide how to distribute a CLI (npm, native binary, source). Runs on its own when the contract is yours to define; follows a surface-recon report when the target is someone else's service
Builds and deploys remote Model Context Protocol (MCP) servers on Cloudflare Workers in TypeScript with @modelcontextprotocol/sdk: McpAgent class patterns for tools, resources, and prompts, all four authentication patterns (token validation, remote OAuth with DCR, OAuth proxy via workers-oauth-provider, and a full OAuth provider), GitHub/Google/Azure OAuth, Durable Objects for per-session state, WebSocket hibernation for cost savings, and dual SSE plus Streamable HTTP transports. Ships templates for basic servers, OAuth proxy, stateful Durable Object servers, and wrangler.jsonc configs, and prevents 15+ errors including McpAgent export mistakes, OAuth redirect URI mismatches, WebSocket state loss, Durable Objects binding errors, and CORS misconfiguration. Use when creating, authenticating, debugging, or deploying an MCP server on Cloudflare Workers.
Create detailed implementation plan using GPT-5.4 with xhigh reasoning. Use when the user wants to plan a feature, refactor, bug fix, infrastructure change, or any multi-step coding task before implementation. 觸發詞：/codex-plan、plan this、implementation plan、幫我規劃、寫實作計畫、規劃功能
Use when the user explicitly asks to run an iterative Codex review cycle on a non-empty git diff, including working-tree changes, current branch vs base, explicit commit/tag/branch refs, code diffs, or Markdown planning documents. Do not use for one-shot reviews, auto-hardening, background checks, plan drafting, or empty review targets.
Configures Letta agent compaction settings and custom summarization prompts. Use when a user asks to change an agent's compaction prompt, improve summaries after context eviction, tune sliding-window or all-message compaction, or design companion/coding-agent continuity summaries.
Use for persistent, one-question-at-a-time consultation on an ambiguous repository change: inspect supplied territory, save decisions, produce an approved plan, run a scored bounded adversarial review with explicit targets and automatic-review limits, resolve or accept human-owned risks, and hand off to a fresh builder without implementing. Trigger on explicit /consultant or requests for durable consultation, plan approval, adversarial review, or builder handoff; exclude ordinary plan editing, standalone review or brainstorming, PRD writing, implementation, testing, and generic technical advice.
Compatibility entry point for the task-generation portion of roadmap planning. Builds a complete dependency DAG, obtains task-boundary approval, and materializes AI-executable task files in bounded topological batches while locking consumed tasks. Use when the user says /create-tasks or has an approved roadmap/design ready for implementation decomposition.
Write decision-free refactor specs/plans that a local small model (e.g. Qwen3.6-27B) can execute mechanically. Use when asked to plan a refactor for small-model execution, "make specs for X", or to set up/extend a specs/ program in a repo. Language-agnostic core; per-language inventory tools in languages/ (TypeScript, Python, Rust, Flutter/Dart).
Disable a hub-installed community skill without removing its files. Use when the user wants to temporarily deactivate a community skill ("disable [skill]"), keep configuration but stop its hooks from firing, or re-enable a previously disabled skill.
Set up a persistent AI agent gateway on macOS with Redis event bridge, heartbeat monitoring, and multi-session routing. Interactive Q&A to match your intent — from minimal (Redis + extension) to full (embedded daemon + Telegram + watchdog). Use when: 'set up a gateway', 'I want my agent always on', 'event bridge', 'heartbeat monitoring', 'agent notifications', or any request to make an AI agent persistent and reachable.
Breaks a feature, PRD, or requirement into TDD implementation tasks with task 0.0 as feature branch creation (MUST be first), each task uses TDD quadruplet (RED test→run fail→GREEN impl→run pass→REFACTOR), includes mandatory public API docs task, update docs task, and code review task, output with `Guidance Used` and `Relevant Files` sections saved to `tasks/tasks-[name].md`, auto-detects test command/source dir/test dir from project conventions. Ruby-first but language-agnostic. Trigger words: tdd task list, tdd tasks, generate tasks, tdd breakdown, implementation tasks, task breakdown, feature tasks, quadruplet pattern.
Create and maintain GitHub Agentic Workflows (gh-aw) for Prowler. Trigger: When creating agentic workflows, modifying gh-aw frontmatter, configuring safe-outputs, setting up MCP servers in workflows, importing Copilot Custom Agents, or debugging gh-aw compilation.
Invoke for complex multi-step tasks requiring intelligent planning and multi-agent coordination. Use when tasks need decomposition, dependency mapping, parallel/sequential/swarm/iterative execution strategies, or coordination of multiple specialized agents.
Build a whole feature by breaking a plan into tasks and running implement-and-review on each. Given a plan or feature, decompose it into tracer-bullet vertical-slice tasks (via the to-tasks skill — acceptance contracts + gate flags + dependencies), then execute the task DAG (independent tasks in parallel) by running implement-and-review for each per-task FE/BE cross-reviewed TDD build, integrate tasks in dependency order with per-task acceptance, and finish with a feature-wide full-review, leaving tests/build green. Opus orchestrates breakdown, scheduling, integration, and final review; per-task building is implement-and-review's job. Use to ship/build a multi-task feature end-to-end. For a single scoped task call implement-and-review directly. Distinct from models-consensus (deliberation only, produces no code) and from ship — ship is the idea→PR conductor, including the interactive spec/plan phases and PR delivery, while implement-feature is the plan→built-feature DAG executor with no spec phase and no PR.
Transforms a rough, unpolished prompt idea into a precise, structured AI coding prompt. Automatically researches the current project context (stack, file structure, conventions, git history) before generating. This skill should be used when the user provides a vague or "dirty" prompt idea and asks to refine, improve, or rewrite it — e.g. "improve this prompt", "refine my prompt", or simply describes what they want done in rough terms. NOT when the user wants the work done or planned rather than the prompt itself — deep-plan produces the plan; just doing the task needs no prompt skill.
Initialize the Playwright trio (@playwright/test + @playwright/cli + playwright-mcp) in a pnpm monorepo, including an e2e/visual testing skeleton, MCP configuration, AI skills generation, and AI memory updates. Includes a built-in postmortem and troubleshooting guide for headless browser CPU overload incidents: it can diagnose high CPU usage (100%) in headless mode, browser hangs, leftover Chromium processes that cannot exit, and 3D WebGL rendering performance issues on large displays; it provides emergency stop-loss steps and a long-term regression checklist. Triggered when the user mentions Playwright initialization, large-display visual testing, e2e test setup, Playwright monorepo configuration, pw_core/pw_visual MCP, and so on. Also applies to troubleshooting scenarios such as headless browser performance issues, headless CPU 100% diagnosis, cleanup of leftover Chromium processes, handling browser processes that cannot exit, and performance tuning for large-display 3D rendering.
Conduct iterative interviews to clarify scope, surface constraints, and reach a "do / don't do / done" agreement before starting non-trivial work. Use when receiving an ambiguous request, a vague task description, or before entering plan mode for any non-trivial implementation. Skip for one-line concrete fixes ("rename foo to bar"), single-fact questions, or when scope, constraints, and acceptance criteria are already specified. Supports requirements definition, debugging investigation, architecture review, security review, documentation creation, and general exploration.
Use this skill for ANY question about creating test or evaluation datasets for LangChain agents. Covers generating datasets from traces (final_response, single_step, trajectory, RAG types), uploading to LangSmith, and managing evaluation data.
Debug LangChain and LangGraph agents by fetching execution traces from LangSmith Studio. Use when debugging agent behavior, investigating errors, analyzing tool calls, checking memory operations, or examining agent performance. Automatically fetches recent traces and analyzes execution patterns. Requires langsmith-fetch CLI installed.
Two-tier memory system that makes Claude a true workplace collaborator. Decodes shorthand, acronyms, nicknames, and internal language so Claude understands requests like a colleague would. CLAUDE.md for working memory, memory/ directory for the full knowledge base.
OpenClaw troubleshooting skill. Use when users encounter the following issues: 1. Gateway fails to start, cannot connect, or is unresponsive 2. Channel connection issues (WhatsApp, Telegram, Discord, etc.) 3. Model configuration issues (API Key, model switching failures) 4. Dashboard/Control UI won't open or connection fails 5. Session issues (lost sessions, cannot create new sessions) 6. Configuration errors (config file errors) 7. Node issues (pairing failures, tool call failures) 8. Browser tool issues 9. Cron/heartbeat tasks not running 10. Installation/update issues Trigger keywords: Gateway, start, connect, error, issue, fault, cannot, failure, error, Dashboard, UI, channel, model, session, node, browser, Cron, heartbeat, install, update Find solutions by consulting the official documentation as needed, and provide step-by-step instructions that beginners can follow.
This skill should be used when the user asks to "create an OpenCode tool", "build an OpenCode plugin", "write a custom tool for OpenCode", "use @opencode-ai/sdk", "use @opencode-ai/plugin", "integrate with OpenCode", "create OpenCode hooks", "define tool schema", "use tool.schema", "work with OpenCode sessions", or needs guidance on OpenCode SDK patterns, plugin development, or custom tool creation.
Use when planning, applying, validating, or archiving OpenSpec changes in this repo, or when downstream AI tools need OpenSpec JSON status/instructions. NOT for generic code review, unrelated docs edits, or replacing generated upstream openspec-* skills.
Use the unified Opper SDKs (`opperai` package for both Python and TypeScript, with built-in agent support) for AI task completion, structured output with Pydantic / Zod / JSON Schema, knowledge base semantic search, streaming, tracing, tool use, and multi-agent composition. Use this skill whenever the user is writing Python or TypeScript code that imports `opperai`, builds an Opper agent, needs to migrate code off the legacy `opper.call`, or asks how to do anything Opper-related in code — even if they don't explicitly name the SDK. Both languages live in one repo with parallel numbered examples; agents are part of the SDK, not a separate package.
Prompt optimization consultant for AI system prompts and custom instructions. Supports Claude API, OpenAI API, Gemini API, CLAUDE.md, ChatGPT, and n8n. Use when user wants to improve, review, or create prompts, system instructions, or custom instructions for any AI platform.
Use the public `orca` CLI to operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, worktree comments, and the browser embedded inside the Orca app. Use when the user says "$orca-cli", "use orca cli", "Orca worktree", "child worktree", "cardStatus", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "terminal send", "full handoff", "handover", "give this to another agent", "another worktree", "Orca browser", "orca artifacts", "share HTML/Markdown", "public artifact link", or "control the browser inside Orca". Prefer this over raw `git worktree`, ad hoc PTYs, Playwright, or Computer Use when the task touches Orca-managed state. Use Computer Use for browser windows, webviews, or desktop UI outside Orca's embedded browser.
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.