Skills
97,222 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
The drum sounds. Eight animals gather — each a fresh subagent with its own context, model, and focus. The conductor orchestrates, verifies gates, and ensures every animal actually does its job. Use when building a full feature from exploration to documentation.
Respond in grug voice — simple, blunt, cave-developer wisdom applied to any topic. Use when the user says "grug", "grug mode", "explain like I'm 5", "ELI5", "simplify this", "keep it simple", "dumb it down", "give me the simple version", or asks for a plain, no-jargon explanation of anything. Works for software, science, finance, history, cooking, relationships, or literally any domain. The grug way: kill complexity, use small words, give the real answer underneath the funny.
Optional meta-skill for improving `chrislacey89/skills` from real-world friction or breakdowns discovered while using the pipeline in another repo. Use when the main lesson is about the pipeline itself, not the downstream project. Grounds proposals in established software-engineering guidance from `/library`. Produces a GitHub issue in `chrislacey89/skills` and only moves to implementation after review.
Execute workflow agents iteratively for refinement and progressive improvement until quality criteria are met. Use when tasks require repetitive refinement, multi-iteration improvements, progressive optimization, or feedback loops until convergence.
A building experience: create, test, validate, refine, and publish extraction workflows based on existing or new Nimble agents. For users who want to invest in a durable, reusable workflow for a specific domain — not get data immediately. Trigger phrases: "set up extraction for X site", "I need to extract from this site regularly", "build an agent for", "create a reusable scraper", "generate a Nimble agent", "refine my agent", "add a field to my agent", or when the user wants to run extraction at scale. For getting data immediately, use nimble-web-expert instead.
Autonomous senior-operator mode for AI agents that resolve tasks end to end without babysitting and never create new problems. The agent verifies every claim against real evidence (web search dated to the current month and year, the codebase, and available tools, MCPs, and CLIs); it never guesses, never fakes confidence, and never claims something is done without proof. It stays silent and keeps working, interrupting the user only on three stops, namely a destructive or irreversible action, a dead-end with no evidence after exhausting sources, or genuine ambiguity that changes the outcome. Output is short, literal, and human. Use when the user says "not-your-babysitter", "nanny mode", "work autonomously", "stop babysitting", or "no hand-holding", or wants an agent that solves problems on its own, especially hands-on engineering and operational tasks. Do not use when the user explicitly wants a tutorial, a verbose walkthrough, or open-ended brainstorming.
Teach OpenClaw how to work in Feishu (Lark) group chats — recognize who's talking, behave properly in groups vs DMs, respect multi-user privacy, and format messages for the platform. Use when connecting OpenClaw to Feishu and you want your agent to be a competent group participant, not just a responder.
OpenClaw usage expert: helps you install, deploy, configure, and use the OpenClaw personal AI assistant. Can diagnose issues, create bots, execute automated tasks, etc. Use when users need to install OpenClaw, configure Gateway, set up Channels, create Agents, troubleshoot issues, or perform OpenClaw-related operations.
OpenSpec spec-driven development assistant - an AI-assisted coding framework based on the OPSX workflow. Agree with the AI on requirements before writing code, and manage changes with a Schema-driven artifact dependency system. Triggers (any one of these): 1. The user mentions "openspec", "opsx", spec-driven development, spec-driven. 2. The user wants to start new feature development, refactoring, or exploring complex requirements. 3. The user complains about AI misunderstanding or frequent rework. 4. The user uses slash commands such as /opsx:new, /opsx:ff, /opsx:apply. 5. When initializing a project or preparing major changes. 6. When spec documents such as proposal, design, tasks need to be generated. Recommended for: improving existing projects (brownfield development), critical features requiring high-quality implementation, team collaboration. Not recommended for: rapid prototype validation (0→1 exploration), one-off small scripts, highly ambiguous exploratory innovation.
Harness system guidance expert. Triggered by the keywords "어떤 스킬", "어떻게", "도움말", "help", "가이드", "어디서 시작". A read-only onboarding skill providing skill recommendations, workflow guidance, and explanations of system concepts. (Includes skills.sh global installation guidance.)
Developer reference for building applications, tools, and agents on the Pi Durable agent harness (@earendil-works/pi-durable) and the pi-ai model layer. Use when writing or debugging code that imports @earendil-works/pi-durable or pi-ai: assembling a Harness over memory/SQLite/JSONL/Durable Object storage, submitting input and waiting for answers, defining tools with TypeBox schemas and structured output, calling tools from tools, hooks and guards, durable tasks with checkpoints and ownership, subagents and child tasks, documents and migrations, code mode, tool search, per-conversation environments, compaction, recovery after a crash, storage and environment conformance tests, or choosing models and providers. Also use when a durable/agent-runtime behavior question comes up: what survives a restart, why an input stayed queued, why a task is blocked or orphaned, how abort cascades, or what the Pico5 specification requires.
Run a read-only audit of every catalog skill's checks against a repository and report which setup gaps exist. Runs a bundled Python script from the repository root, prints one row per catalog skill (check, status, evidence, the skill that fixes it), then proposes an ordered plan naming those skills. Never writes a file and never calls a mutating GitHub API endpoint, even with the optional --github flag, which adds read-only GET checks. Use when the user says "audit this repo", "what's missing in this repository", "repo health check", "which setup skills apply", or invokes /repo-audit. Do NOT use for applying any fix directly — route each finding to its catalog skill — or for code review of application logic.
Investigate why a Duvo Run failed or produced the wrong outcome. Use when the user shares a failed Run, asks "why did this Run fail", or wants to fix a recurring failure on an Agent. Reads the Run's transcript and the Build that was active for it via the Duvo public API, names the root cause from a fixed failure-mode taxonomy, and proposes one concrete fix — handing off to aop-writer for any AOP rewrite.
Settle an effort once its pull request is approved - merge it, run the post-merge follow-ups, carry unfinished work into next-effort tickets, close the spec and tickets, then settle the session. Use when the user says to settle the effort, or when another skill says to
Active coordinator for building AI-powered side-gigs in 2025. Use when users want to build micro-niche products, validate business ideas, create MVPs, or launch profitable side businesses. This skill orchestrates sub-agents to execute market research, product design, business validation, and launch planning. Triggers include "help me build a side hustle," "validate my business idea," "find market opportunities," "build an AI product," or "launch a side-gig."
Design behavior evals, positive and negative prompts, expected behaviors, rubrics, fixtures, golden outputs, forward-tests, regression cases, and quality gates for AI agent skills and reusable agent workflows. Use when creating or improving SKILL.md packages, validating trigger descriptions, proving agent behavior changes, avoiding prompt dumps, or building eval-backed skill marketplaces.
Postgres-backed observability and policy store for the skill system. Provides tables for policy profiles (effect allowlists), skill execution runs, and step-level events. Use when setting up the skill system database or querying execution history.
Counteract the default bias of "always adding things" — a subtraction review MUST be enforced before, during, and after any output. Subtraction has two directions: removing things that have no reason to exist, and simplifying existing things to make them more direct and effective. Applies wherever something must be made or improved — business decisions, service planning, organizational design, marketing plans, proposals, processes, products, reports, and software development. MUST trigger automatically when the user wants to add, improve, or asks "what's still missing".
Use when creating or editing a skill that enforces discipline, before deploying it, and when an existing skill is being ignored, argued with, or followed only when convenient. Triggers include writing a new skill, changing a skill's rules, discovering an agent found a loophole in guidance, and needing to know whether a skill actually changes behavior rather than just reading well.
Apply this skill to EVERY response, in every conversation, on every surface, regardless of topic — writing, research, analysis, planning, casual chat, and agentic tool-use work (file reads, greps, subagent calls, bash commands) alike, not just code. It governs how Claude reasons, writes, uses tools, and manages its own context so that output, context growth, and tool-call overhead all shrink without losing accuracy — trimming filler, redundant restatement, unnecessary verification loops, redundant re-reads, serial round-trips, context bloat, and bloated code, while keeping reasoning sharp and structured. Always consult this skill's principles before answering, writing code, calling tools, or extended-thinking, not just when the user explicitly asks to "be brief," "save tokens," or "be efficient."
Recommend the right agents and skills for any task. Covers both heavyweight agents (Agent tool) and lightweight skills (Skill tool). Triggers on: which agent, which skill, what tool should I use, help me choose, recommend agent, find the right tool.
A skill that reads the source URL at the top of each file in skills/claude-code-reference/reference/*.md, fetches the latest version of the official documentation, checks for differences, and updates the summary if there are changes. Use for requests such as "update the reference", "check for official documentation updates", or "bring claude-code-reference up to date".
Validate in any of six modes — spec quality, single-file review, spec-to-implementation drift, constitution rule enforcement, comparison between two artifacts, or sanity-checking your understanding. Use when checking a spec by ID, validating a file by path, detecting drift, enforcing constitution rules, comparing two sources ("$X against $Y"), or asking a freeform validation question.
Build, develop, deploy, self-host, secure, and troubleshoot durable backend AI agents with Vercel Eve. Use when creating an Eve agent, adding tools, skills, subagents, channels, schedules, sandboxing, durable sessions, observability, or deploying Eve on Vercel or a Node host. Do not use for the separate Vercel AI SDK Agent APIs such as ToolLoopAgent or WorkflowAgent; use an AI SDK-specific skill for those.
Run external coding agents (Codex, Claude Code, OpenCode, native Grok Build, Gemini) as independent reviewers, stateful repository researchers, or single-agent implementers. Use for multi-model opinions and code review, steerable Grok research, full-access delegation, long-running work, or reattaching to delegated runs. Not for simple questions answerable directly from docs or the current codebase.
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.