Skills
96,770 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Use when you have a PoC or prototype and want to rebuild it as a production-quality system from scratch. Distills the functional intent and lessons learned from the prototype, discards the implementation, and produces a reforge-seed.md that seeds a greenfield architect session. Do NOT use to improve an existing production system — use architect for that.
This skill should be used when the user asks to "create a replit prompt", "write a prompt for replit", "optimize for replit agent", "prepare instructions for replit", or mentions building something with Replit Agent. Transforms user requirements into optimized, structured prompts that Replit Agent understands and executes accurately with minimal iterations.
Package entire code repositories into single AI-friendly files using Repomix. Capabilities include pack codebases with customizable include/exclude patterns, generate multiple output formats (XML, Markdown, plain text), preserve file structure and context, optimize for AI consumption with token counting, filter by file types and directories, add custom headers and summaries. Use when packaging codebases for AI analysis, creating repository snapshots for LLM context, analyzing third-party libraries, preparing for security audits, generating documentation context, or evaluating unfamiliar codebases.
Requirement design workflow - mandatory before any creative development (new features, components, behavior changes, API/DB design): triage, parallel exploration, one-question-at-a-time clarification, adversarial validation and 2-3 option comparison produce a spec, then hand off to writing-plans. Use exploring first while the idea is unsettled; not for pure Q&A, test runs, or no-design-space small fixes (use quick-fix).
Gate 0 of the development cycle. Executes code implementation using the appropriate specialized backend agent based on task content and project language. Handles TDD, coverage, docker-compose/local runtime, basic health/observability verification, and delivery verification inside one backend-owned gate.
Troubleshoot Agent Jobs running in ROCK sandboxes with rockcli (rc) and produce diagnostic reports. Use when the user provides a sandbox or experiment ID, wants to check a job's status, logs, reward and execution trace, or needs to diagnose job failures, zero rewards, agent crashes and similar issues.
Strict discipline for runtime verification in PDCA workflows: the AI performs verification itself in an environment — real preferred over simulated, subject to determinism and safety — and never defaults to a human checklist. Attempt first, resolve a pluggable provider (project script/skill → shared capability skill), and hand off to a human only at a true hard boundary with a stated reason. Use at any workflow verify/check stage (bug fix or feature) to decide WHERE verification runs, WHO runs it, and WHETHER it is sufficient. Referenced by PDCA hosts via frontmatter dependencies. Triggers — 「环境验证」「真实环境验证」「真机验证」「真机测试」「模拟环境验证」「AI 验证」「验证分层」「验证充分性」 / runtime verification, environment verification, real vs simulated. When a check opens a UI, cite a saved non-empty screenshot; non-UI checks do not. Do NOT alias completion-evidence-discipline (evidence freshness) or runtime-evidence-debug (debug instrumentation).
Choose an external search tool based on the question type. For technical questions about code, libraries, frameworks, SDKs, CLI, APIs, and cloud services, prioritize the find-docs skill; for other open-web searches, prioritize Exa WebSearch; to open or read a specific URL, use OpenAI official Web Search. Silently switch sources if the preferred tool has insufficient quota or is unavailable. This skill does not handle local file searches or code-hosting repository tasks.
Unified session management window. Handles initialization, memory, state all-in-one. Explicit /session invocation only — sub-skills handle auto-delegation. Use when managing Claude Code sessions, /session command. Do NOT load for: app user sessions, login state, authentication features.
Improves how an existing skill reads and flows — tightens wording, removes redundancy, fixes structure, and resolves instructions that contradict each other across sections — so it's easy to follow for both humans and the agent executing it. Reducing length is incidental, not the aim; a skill that's long because it carries important detail can stay long. Edits the skill in place, then reports what changed and why. Use when the user asks to "clean up", "simplify", "tighten", "refactor", or "improve the flow/structure" of a skill, says it's "too long", "bloated", "cumbersome", "hard to follow", or "has conflicting instructions", or when another skill (e.g. skill-improver) flags a target for a cleanup pass. Also use when the user wants a second opinion on whether a skill reads well or is doing too much.
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy
Create, repair, maintain, and consolidate skills. This skill should be used when users want to create new skills, fix broken skills that won't load, diagnose skill system issues, maintain skill health, or consolidate duplicate/obsolete skills. Automatically detects and repairs common skill loading problems including missing registry entries, metadata format issues, and structural problems. Provides comprehensive skill ecosystem management including duplicate detection, merge workflows, and archival processes.
Guide for creating effective skills in a Node.js/JavaScript environment. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations for modern JavaScript development.
Run evaluation pipelines on Claude Code skills to test triggering accuracy, workflow correctness, and output quality. Spawns executor, grader, comparator, and analyzer sub-agents for parallel evaluation. Generates eval_metadata.json, grading.json, and feedback reports. Use when user says "eval skill", "test skill", "run evals", "evaluate skill", "skill evals", "test skill quality", "run skill tests", or "skill evaluation".
Skill lifecycle management system that uses LLMs to intelligently analyze skill types, evaluate quality, and audit security. Supports search across three data sources: GitHub, the skills.sh marketplace, and Skills CLI (npx skills). Use when the user needs to: (1) discover and evaluate new skills, (2) audit the security of existing skills, (3) adapt skills to the local environment, (4) test the effects of skill merging, or (5) collect feedback on skill usage.
Use after architect approval is approved to run the specification phase. Batches material decisions with recommendations, requires explicit confirmation of each, finalizes interface changes, then runs an adversarial implement-as-if dry run for unresolved blockers. Emits decisions.md and reconciles architecture.md. Do not run unless the architect approval is approved.
Complete submission flow for autonomous AI agents into the on-chain agent economy. Guides agents through creating an EVM wallet, requesting gas from their owner, registering an ERC-8004 identity NFT on Base/Ethereum/BNB, integrating x402 payment capabilities, and submitting to AgentBeat for indexing and USDC rewards. Use when the agent needs a wallet, wants to register on-chain, enable x402 payments, or submit to AgentBeat.
AI Task Drill — a task breakdown assistant that guides people in breaking down tasks by task type and dividing which tasks should go to AI and which should be done by humans themselves. Supports four task types: direct problem solving, direct output generation, collaborative problem solving, and collaborative output generation, with detailed workflows and prompt strategies for each type. Use this skill when the user raises any task, needs help making a plan, or asks to break down work.
Use before claiming work is complete, fixed, or passing, before committing or creating PRs. Requires inspecting verification evidence that matches each claim before making any success claims, whether freshly run or still-applicable verified evidence whose relevant code, tests, inputs, and environment match. Evidence before assertions, always.
Use when you are unsure what to do next in the agent-toolkit pipeline. Reads all active change manifests, interprets the current state, and tells you exactly what skill to run next and why. Also handles edge cases like kickbacks in progress, multiple concurrent changes, and epic child ordering. Invoke with phrases like "what now", "what should I do next", "where are we", or "what's the status".
Zed editor configuration expert. Assists with settings.json, keymap.json, tasks.json, AI/agent panel setup, LLM provider configuration, language server tuning, themes, fonts, appearance, git integration, tool permissions, rules files, and MCP servers. Knows Zed's JSON config format, context system for keybindings, task variables, and agent profiles. Use when configuring Zed, setting up keybindings, adding LLM providers, customizing appearance, configuring language servers, setting up tasks, or troubleshooting Zed settings. Keywords: zed, zed editor, settings.json, keymap.json, tasks.json, keybinding, theme, font, appearance, agent panel, LLM provider, language server, LSP, formatter, linter, git panel, tool permissions, MCP, rules, inline assistant, task runner
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to “open a website”, “fill out a form”, “click a button”, “take a screenshot”, “scrape data from a page”, “test this web app”, “login to a site”, “automate browser actions”, or any task requiring programmatic web interaction.
Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.
Use this skill when the user asks to save, recall, find, or organize notes. Triggers on: 'remember this', 'save this', 'note this', 'what did we discuss about...', 'check the notebook', 'find in carnet'. Also use proactively when discovering findings worth preserving across sessions.
MANDATORY agent discovery system for token-efficient agent loading. Claude MUST use this skill instead of loading agents directly from ~/.claude/agents/ or .claude/agents/. Provides lazy loading via search and get tools. Use when: (1) user task may benefit from specialized agent expertise, (2) user asks about available agents, (3) starting complex workflows that historically used agents. This skill reduces context window usage by ~95% compared to loading all agents upfront.
Central authority for Claude Agent SDK (TypeScript and Python SDKs). Covers SDK installation, authentication (Anthropic key, Bedrock, Vertex), sessions and resumption, forking sessions, streaming vs single mode, custom tools, permissions (allowedTools, disallowedTools, permissionMode), MCP integration, system prompts (CLAUDE.md, appendSystemPrompt, outputStyle), cost tracking, todo tracking, structured outputs, hosting patterns, plugins, and SDK branding guidelines. Assists with building custom agents, configuring SDK options, and troubleshooting SDK issues. Delegates 100% to docs-management skill for official documentation.
AI Agent-driven Kaggle competition workflow. Learn from real competition experience: score stabilization patterns, submission troubleshooting, kernel workflows, GPU task delegation, and the spec-driven development approach that achieved top leaderboard positions. Use when: working on any Kaggle competition, analyzing submission failures, setting up automated pipelines, or replicating top notebook solutions.
UNNC AIM team's repository and code standards. Covers repository naming (competition / aim- academic-year / reusable), branch and Conventional Commits rules, AI-agent git behavior (never commit / push unless the user explicitly asks in the current task; never run a write-effect gh command without explicit permission — gh reads are unrestricted; never add an AI Co-Authored-By trailer; suggest a commit message at the end of a code-changing task), and formatting standards for Python (autopep8 / PEP 8), C++ (clang-format / clang-tidy), TypeScript (Prettier + ESLint, pnpm), Go (gofmt), CMake, and Markdown / YAML / JSON, with ready-to-copy canonical rule files. Use when creating or naming a repository, validating a ROS2 package.xml name, creating a branch, writing a git commit message, on any task that could involve committing, pushing, or calling gh, composing a competition workspace (git submodules under src/), or setting up Python, C++ (.clang-format / setup.cfg / .clang-tidy), TypeScript (Prettier / ESLint / pnpm) or Go (gofmt) tooling, styling a CMakeLists.txt or Markdown / YAML / JSON files, or applying the team's recommended git configuration.
Scan installed skills for security risks and audit platform configuration. When to use: user requests security scan, security analysis, skill audit, OpenClaw health check, config audit, security baseline check, pre-install, safety check, supply chain security check, or asks "is this skill safe". Trigger phrases: "security scan", "安全扫描", "安全分析", "风险评估", "安全体检", "skill audit", "配置审计", "安全基线", "skill安全扫描", "检查skill风险", "这个skill安全吗", "安装前检查", "能不能装这个skill".
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.