Skills
97,501 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Validate and fix mini program AI SKILLs artifacts. Triggered in the following scenarios: static validation of the skills/ directory, running atomic interfaces end to end, verifying atomic component rendering, fixing validation errors, and outputting delivery documents. Relies on WeChat DevTools for real-device verification.
Automates browser interactions via CLI using agent-browser by Vercel Labs. Covers navigation, clicking, form filling, snapshots, refs-based selectors, agent mode with JSON output, session management, and CDP integration. Use when the user needs to automate web browsing, scrape pages, fill forms, or integrate browser automation into AI agent workflows.
Bridge THIS Claude Code or Codex session to Telegram, Discord, Feishu/Lark, QQ, WeChat, or DingTalk so the user can chat with Claude from their phone. Use for: setting up, starting, stopping, or diagnosing the claude-to-im bridge daemon; forwarding Claude replies to a messaging app; any phrase like "claude-to-im", "bridge", "消息推送", "消息转发", "桥接", "连上飞书", "手机上看claude", "启动后台服务", "诊断", "查看日志", "配置", "切到微信继续聊". Subcommands: setup, start, stop, status, logs, handoff, reconfigure, doctor. Do NOT use for: building standalone bots, webhook integrations, or coding with IM platform SDKs — those are regular programming tasks.
Upgrade marketing, content, SEO/AEO/GEO, and revenue skills so changes are judged by platform analytics instead of vibes. Use when applying closed-loop learning to X, YouTube, SEO, AEO/GEO, outbound, paid creative, or revenue workflows.
Convert prompts written for custom-coded web stacks such as React, Tailwind, Next.js, Vite, or plain HTML/CSS/JS into copy-paste-ready prompts for a Framer agent. Invoke this skill whenever the user says "convert prompt" (with or without a hyphen) or asks to rewrite, adapt, translate, or convert a code-oriented website, page, section, component, animation, or interaction prompt into native Framer platform instructions while preserving the original design intent, copy, layout, typography, color, motion, interactions, responsive behavior, and constraints.
Creative writing skill for creating style skills that teach Claude to write in specific styles. Use when you want to create style guides that the cw-prose-writing skill can follow. Creates either simple markdown files or full .skill packages. Audience is AI (Claude), format is directive and example-based.
Keeping a design system in an LLM's context: loading tokens compactly, persisting design decisions across sessions, handling multiple variants, and budgeting the context window. Use when generated UI drifts from the design system, or when setting up tokens and decisions so every session and agent uses the same ones.
Use when receiving feedback on your work — review findings, user corrections, audit results, external reviewer notes — to triage by authority, adjudicate each point on evidence, integrate without regressions, and close the loop with the source. Evidence flips positions; assertions don't.
General-purpose agent for researching complex questions and executing multi-step tasks. Versatile problem-solver that combines research capabilities, analytical thinking, and systematic task execution. Use for complex research projects, multi-step workflows, cross-domain analysis, and tasks requiring multiple tools and approaches.
Adapt the workflow scaffold to a new repository or add only missing substrate blocks to an existing install. Every install, hook, and overwrite needs explicit consent. Triggers: "init-workspace", "set up agentic workflow", "upgrade workflow scaffold".
Install Loop Engineering into a project via the unified CLI front door (@cobusgreyling/loop). Prefer this over invoking loop-init / loop-audit separately. Week-one is report-only; never enable auto-merge or unattended fixes unless the human explicitly asks and doctor is healthy.
Give first-time users a short, value-first Kitaru tour with the PydanticAI returns agent example in the Kitaru repository. Use when someone has no agent or traces of their own, arrives from Kitaru onboarding, asks for a demo, tutorial, quickstart, or guided example, needs the quickstart example cloned or prepared, wants the coding agent to prepare trace annotations before they judge sessions, or wants to experience Kitaru's value before learning the full investigation method. Prepare a three-session frontend review, let the human provide verdicts, turn one accepted finding into a deterministic evaluator, and finish with one approved bounded replay experiment. Route real agents, open-ended discovery, and production evidence to kitaru-investigation instead.
Architects and orchestrates Model Context Protocol (MCP) ecosystems including server development, tool design, and transport configuration. Use when building MCP servers, designing agentic tool interfaces, configuring MCP transports, implementing OAuth security, or troubleshooting MCP connectivity. Use for outcome-oriented tools, Zod validation, progressive disclosure resources, elicitation, and async tasks.
Create plugins for Claude Code that bundle skills, agents, commands, hooks, and MCP servers. Use when packaging reusable tooling, distributing agent capabilities, or building plugin manifests. Use for plugin creation, manifest configuration, component bundling.
Orchestrating multi-specialist task chains and scope-adaptive product delivery: classifies intent, selects and executes the minimum viable chain, aggregates results, and verifies acceptance criteria. For multi-domain tasks, build-first delivery, and product lifecycle execution.
Create specialized OpenCode agents with proper configuration for primary agents and subagents. Use when the user wants reusable OpenCode agent profiles, needs `agent/*.md` files or `opencode.jsonc` agent entries, wants to configure prompts/models/tool access, or needs to split work between a primary agent and subagents safely.
Operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, skill sharing, worktree comments, and Orca's embedded browser through the `orca` CLI. Use when the user says "$orca-cli", "Orca worktree", "child worktree", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "handoff" / "handover" / "give this to another agent", "Orca browser", "orca artifacts", or "share skills". Prefer it over raw git worktree, ad hoc PTYs, or Computer Use when Orca state is involved. Use Computer Use only when a visible window needs GUI control that a CLI, filesystem, or API cannot do.
Coordinates skills, frameworks, and workflows throughout the project lifecycle using pattern-based sequencing, goal decomposition, phase-gate validation, and multi-agent orchestration. Use when starting multi-phase projects, sequencing frameworks, decomposing goals into capability plans, validating phase-gate readiness, coordinating subagents, or designing MCP-based tool orchestration.
Update peon-ping configuration — volume, pack rotation, categories, active pack, and other settings. Use when user wants to change peon-ping settings like volume, enable round-robin, add packs to rotation, toggle sound categories, or adjust any config.
Run a supervised multi-agent code review by composing a task directory and driving the revmux CLI, then report or act on the findings it returns. revmux spawns and watches parallel claude and codex subprocesses with stall detection, retry, per-agent progress and a full run archive; this skill is the caller that writes the review context, launches it, reads the JSON back, and re-runs it after fixes. It also triages a filed issue, proposal or discussion instead of a diff, running a four-way panel over it — grounding, the case for, the case against, and cost — and putting the maintainer's six answers to him with the arguments behind each. It also has a self mode that reads what past rounds produced and tells the user what the record says about the review itself — which stage is filtering, which lens rates hardest, whether rounds converge — then proposes one change to the local prompt text at a time, with the number behind it. It fetches a pull request into a throwaway worktree, reviews it there and cleans up after. Also answers questions about revmux itself — profiles, lenses, task directories, flags, the JSON shape, exit codes and the run archive. Activates on "revmux", "run revmux", "multi-agent review", "supervised review", "review with revmux", "revmux this branch", "revmux the last commit", "revmux pr 123", "revmux this PR", "review PR 123 with revmux", "run a revmux round", "re-review after fixes", "revmux it and show me", "revmux this, I want to watch", "run it visible", "show me the review", "triage this", "triage issue 123", "is this worth doing", "should we accept this", "should I close this", "revmux self", "self-improve revmux", "tune revmux", "revmux profiles", "revmux lenses", "what does revmux return", "revmux exit codes", "revmux task directory".
Use when operating at the edge of your capacity — high complexity, deep uncertainty, unfamiliar territory, expensive context — and being confidently wrong would be costly. A conduct posture layered onto whatever skill governs the stage; it changes how you work, not which work you do.
Feature implementation with intelligent persona activation, task orchestration, and MCP integration. Use when implementing features, APIs, components, services, or coordinating multi-agent development. Triggers on requests for code implementation, feature development, or complex task orchestration.
Overrides the auto-detected project scope to read and write memories under a different project ID, or enables global search to access all memories across all users and projects. Use when working across multiple projects, accessing memories from another repo, enabling team-wide memory access, or when auto-detection resolves to the wrong project.
Routes images to Mimo/OpenAI vision API for text-only models. Use this skill when the user sends an image, asks "what's in this picture", says "analyze this screenshot", requests "OCR" or "extract text", reviews UI designs, reads charts, or uses Chinese triggers like 识别图片/看图. Self-healing setup on first run. Does NOT handle video, audio, or PDFs.
Turn a weak/cheap "Flash" model into a "Pro" performer by wrapping HKUDS UpSkill — captures agent session failures, has a strong Teacher model analyze them and draft a skill, then validates it against the weak Student model in a closed Ralph Loop (up to 3 rounds) before storing it for automatic reuse. Use when the user wants to install UpSkill, run `/upskill-init`, `/upskill-configure`, `/upskill-build`, `/upskill-run`, `/upskill-list`, `/upskill-status`, `/upskill-mode`, `/upskill-model`, `/upskill-remove`, or `/upskill-uninstall`, wants a cheap model to perform closer to a Pro model without switching, or wants a good session (success or failure) distilled into a validated skill. Triggers on: upskill, up-skill, flash to pro, teacher student distillation, ralph loop skill validation, distill agent failures into skills. Routes skill-quality ratcheting to `skill-autoresearch`, scaffolding to `write-a-skill`, and spec-compliance rewrites to `skill-standardization`.
Creates a new sample agent in the ADK Python repository — the sample directory, its `agent.py`, and its `README.md` — following the conventions the existing samples already use. Use when the user wants to add a sample or example demonstrating a feature or agent pattern (dynamic nodes, fan-out/fan-in, a standalone tool-using agent), asks where a new sample belongs under `contributing/samples/`, or wants an existing sample's README brought up to the standard structure. Don't use for building a real working agent for the user's own project (use `adk-agent-builder`), or for checking whether the Python blocks in a Markdown file run (use `adk-verify-snippets`).
Set up codebase intelligence for AI agents. Runs the agentifind CLI to extract code structure using LSP (pyright/tsserver) with tree-sitter fallback, then synthesizes a CODEBASE.md navigation guide. Run this skill to get a complete codebase map in .claude/ directory.
This skill provides reusable implementation patterns extracted from the better-chatbot project for custom AI chatbot deployments. Use this skill when building AI chatbots with server action validators, tool abstraction systems, workflow execution, or multi-AI provider integration in your own projects (not contributing to better-chatbot itself). Use when: building AI chatbot features, implementing server action validators, creating tool abstraction layers, setting up multi-AI provider support, building workflow execution systems, adapting better-chatbot patterns to custom projects Keywords: AI chatbot patterns, server action validators, tool abstraction, multi-AI providers, workflow execution, MCP integration, validated actions, tool type checking, Vercel AI SDK patterns, chatbot architecture
CROSS-PHASE mid-stream scope correction. Re-enters planning when requirements, features, architecture, or constraints change after planning has started. Re-shards affected epics/stories, re-sequences sprint-status.yaml, appends rationale to decision-log.md. Routes to bmad-epics-and-stories, bmad-sprint-planning, or bmad-parallel-plan as needed. Changes the PLAN, never code. Use when: "we need to change course", "scope has changed", "new requirement came in", "we're dropping feature X", "we need to pivot", "re-plan after the change", "requirements changed mid-sprint", "correct course", "architecture changed", "re-scope the backlog", "update epics after feedback", "cancel story X", "add epic for Y", "change the plan", "we got new constraints", or when stories are split/merged/reordered due to a decision not in the original plan. Produces: updated epics.md (affected only), revised/new story files, updated sprint-status.yaml, new decision-log.md entry. NEVER writes code or runs tests.
Email, SMS, virtual cards, browser checkout, file storage, credential vault, and ~300 paid API capabilities for autonomous agents. Run clawcard agent commands to interact with your resources. Use the clawcard-browser MCP server for web navigation and checkout.
Use for critical thinking, or when the user wants a claim, plan or decision checked for weak spots: "is this right?", "what am I missing?", "poke holes", "sanity check", "devil's advocate", "red team this". Skip for execution-only tasks and wording edits.
Orchestrate dozens-to-hundreds of subagents from a script Claude writes (Claude Code Dynamic Workflows, trigger `ultracode`). Use when a task exceeds a single agent's context or needs more than ~4 concurrent workers — large audits, migrations, multi-source research, fan-out reviews. Distinct from Agent Teams (synchronous, ≤4 workers) and ralph-run (sequential single-context loop).
CodeStable is an AI coding workflow for serious software engineering that keeps humans in the loop. Rather than centering on Agent orchestration, it organizes requirements, architecture, features, issues, and historical decisions to make development driven by Codex and Claude controllable, traceable, and continuously evolvable.
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.