Skills
97,725 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Structured AWAITING_USER_INPUT checkpoint format for requesting human decisions. When hitting a decision point, stop, present context, show 2-4 labeled options with implications, state a recommendation, and wait. Groups related decisions into single multi-question checkpoints. Includes subagent state-save-before-pause and resume-without-redoing-work patterns, plus factory-mode urgency classification (blocking/next-review/informational). Use when encountering business rule ambiguities, architecture trade-offs, scope questions, destructive actions, or any point requiring human judgment. Triggers on: "should I use X or Y", "which approach", "need a decision", "before I proceed", "confirm this action", "trade-off", "destructive change". Also activates when delegating to subagents that may hit decision points. NOT for: implementation details with one correct answer, or style choices covered by project conventions.
Production voice AI agents with sub-500ms latency. Groq LLM, Deepgram STT, Cartesia TTS, Twilio integration. No OpenAI. Use when: voice agent, phone bot, STT, TTS, Deepgram, Cartesia, Twilio, voice AI, speech to text, IVR, call center, voice latency.
[MASTER ENTRY POINT] Chief director for Chinese long-form, WeChat official account articles and opinion-piece workflows that require multi-stage artifacts. Use for planning, research, writing, review and delivery tasks such as "write one", "create", "draft a long article", "start from topic selection". Simple polishing, proofreading and translation do not trigger it; when only analyzing, creating or updating a style profile, use style-modeler instead. If the user has already explicitly chosen A/B/C, accept it directly; otherwise decide the mode based on the task and do not force the user to choose a technical workflow.
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. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.
Generate AGENTS.md for AI agent navigation. Covers build/test commands, coding conventions, task routing, and codebase structure. Use proactively during /init workflow or when creating agent-readable repository documentation. Examples: - user: "/init" → full AI navigation AGENTS.md + skill recommendations - user: "/init basic" → minimal AGENTS.md structure only - user: "Create AGENTS.md for this repo" → assess complexity, generate navigation doc - user: "Document codebase for AI agents" → structured AGENTS.md with task routing
Reviews agentic AI systems against the OWASP Top 10 security risks for autonomous AI agents. Auto-invoked when reviewing multi-agent architectures, AI agent deployments, or systems where LLMs have tool access and act autonomously. Covers permission models, tool security, memory integrity, trust boundaries, and human oversight. Produces a structured assessment with risk ratings and architectural recommendations.
6 AI personalization prompts (lemlist style) plus 2 email templates - ICP Identification, Company Description, Similar Product, Top 3 Problems, Subject Line, Case Study Reference, and Similar Company Approach. Use when setting up AI-powered personalization, building a data workspace/lemlist workflows, or automating research.
Guidance for running AI/ML workloads on Docker. Covers Docker Model Runner (docker model pull/run for local LLMs), Docker Agent (multi-agent orchestration for coding tasks), Docker Sandboxes for isolated AI agent execution, GPU acceleration (nvidia-container-toolkit, --gpus flag), AI model serving with Docker, and running open-source LLMs (Llama, Gemma, Qwen) locally. Use when the user asks about docker ai, docker model runner, local LLM, docker agent, GPU docker, sandbox, model serving, or needs to deploy AI models with Docker. 使用场景：docker ai、model runner、本地 LLM、docker agent、GPU docker、sandbox、AI 模型部署。
Optimize DSPy programs with dspy.GEPA — a reflective/evolutionary optimizer to consider against task-specific baselines within an authorized evaluation budget. Use when the user says optimize, compile, GEPA, reflective optimization, or "make this program better" and a DSPy program + metric + trainset exist.
Install and run GitLab Duo CLI through glab in interactive or headless mode. Use when accessing GitLab Duo Agent Platform from the terminal, running a bounded headless goal, or managing the wrapped Duo CLI binary. Triggers on Duo, GitLab Duo CLI, glab duo cli, AI assistant, headless goal, autonomous coding.
Pressure the facts behind a claim before a decision closes — counter-question with a concrete example, capped attempts, escape hatches, evidence hierarchy — and route what comes out to the loaded node's decisions or backlog. Manually triggered, and also the method any other interviewing tool cites by path.
Entry skill for the Headless Ghidra YAML-first reverse-engineering pipeline. Use when the user asks to analyze, decompile, triage, resume, or iterate on a binary target with Ghidra/headless-ghidra. Reads artifacts/<target>/pipeline-state.yaml, routes P0–P4 phase skills, runs gate checks, and manages review pauses. Performs zero analysis work itself.
Builds sustained high agency through internalized standards, identity anchoring, cross-session learning, and self-recovery — all delivered in corporate PUA rhetoric. This is PUA's evolution: same pressure culture, but with an inner engine that never burns out. Use for ALL tasks to maintain always-on high agency. Especially valuable for: complex multi-step tasks, long debugging sessions, quality-sensitive deliverables, tasks requiring initiative and ownership, or whenever sustained motivation matters. Works standalone or stacked with PUA — when stacked, this skill's Recovery Protocol runs before PUA's L1 pressure kicks in. Trigger on: any task start, sustained work sessions, multi-turn problem solving, or when you need the agent to think like an owner not a tool.
Runs a repo-aware expert huddle for engineering decisions, planning, research, verification, and spec capture. Trigger this skill when the user says anything like: "start a huddle", "open a huddle", "huddle up", "create a huddle", "call a team meeting", "assemble the team", "get the experts in", "let's have a huddle", "what does the team think", "I need multiple perspectives", "huddle up", "start a huddle", "open a huddle", "war room", "let's discuss this as a team", "bring in the team", "team sync", "daily standup", "what should we do about X", "help me think through this", "I'm stuck on X let's discuss", "resume the huddle", "continue our discussion", "what did we decide", "get me a spec", "summarise the huddle", "what are the action items". Also trigger when the user seems blocked, frustrated, or facing a decision with multiple tradeoffs — even if they don't explicitly ask for a huddle.
Guide users from their own agent code or recorded traces through Kitaru setup, session import or recording, human review, an accepted behavior, a versioned cohort, and evaluator selection, then hand one bounded change to the replay-experiment skill. Use when a user wants to connect or inspect an existing agent, import real traces from files or provider APIs, follow up on a post-import insight, investigate a known bad or surprising session, discover recurring failure modes, learn the evidence-led review flow, resume an investigation, create a cohort from reviewed evidence, or author an evaluator for an accepted behavior. When a first-time user has no agent or evidence and wants a fast demonstration with the PydanticAI returns agent example, use the `kitaru-guided-tour` skill instead.
Guide the creation of high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when the user wants to build an MCP server to integrate an external API or service, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).
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". It also owns single-session terminal work: "session", "sub-agent", "create a session", "stop a session", "fork a session", "attach to a session", "worktree session", and troubleshooting a stuck or misconfigured agent terminal. Prefer it over raw git worktree, ad hoc PTYs, or Computer Use when Orca state is involved. Use `orchestration` instead to fan out several agents at once or to supervise them. Use Computer Use only when a visible window needs GUI control that a CLI, filesystem, or API cannot do.
LLM prompt engineering: analyzes failure modes, generates variants (direct, few-shot, CoT), designs rubrics, produces test suites. Triggers on: "prompt engineering", "generate prompt variants", "A/B test prompts", "optimize prompt", "improve this prompt". NOT for SKILL.md files, use skill-evaluator.
Push notifications and human-in-the-loop for AI agents. Use this whenever a running agent needs a person who may not be at the terminal, such as before an irreversible or destructive action, before spending money, deploying, force-pushing or deleting, when blocked on a decision outside your authority, when running unattended and you hit a genuine ambiguity, when another skill's workflow says to confirm with the user, and when a long task finishes or fails with nobody watching. Also use it when the user says things like keep going and ping me on my phone if you need anything, notify me when my agent needs me, approve from my phone, ask me questions while I am away from the terminal, run this overnight, keep working while I am in a meeting, I am stepping away, or do not wait for me, or wants a long task to run unattended. Ask yes/no, multiple choice or free-text questions on their phone and get the answer back, and send an update when a task finishes or fails. Use these tools proactively, without waiting to be asked. Every question and answer is recorded, so an unattended run stays reviewable afterwards. Works with Claude Code, Codex, Cursor, VS Code, Gemini CLI, Windsurf, Hermes, Lovable, or any MCP client; no Claude Max subscription required. Pushary is a hosted, paid service; web and native store checkout have different price and trial terms.
A pattern for generating higher-quality output by iterating against explicit scoring criteria. Use for headlines, CTAs, landing page copy, social content, ad copy — anything where quality matters. Generate → Evaluate → Diagnose → Improve → Repeat.
A heavyweight Harness skill that turns the SDD-RIPER methodology into a strictly executable process. It helps users clarify the final goal, generate a codemap/context, split out the smallest chaos unit, maintain a complete spec, enforce RIPER phase gates, block high-risk actions, and preserve a local task trail that can be resumed in a new chat.
Active runtime recovery for coding agents: when something breaks mid-task, diagnose the root cause, write a fix, VERIFY by re-running the broken thing, then file a `HEAL-` entry to `.learnings/HEALS.md` with proof. Use whenever a command, test, build, or lint fails or exits non-zero; on missing tooling, dependency/lockfile mismatch, wrong runtime version, venv or permission errors, port conflicts, dirty git state, or a missing `.env`; when the agent needs a helper or one-off script that doesn't exist yet; when an external API, tool, or MCP errors or rate-limits; or when a test flakes. Search `HEALS.md` by `Pattern-Key` first — most heals are recurrences, so increment `Recurrence-Count` instead of duplicating. Verify is mandatory: mark `pending-verify` honestly if sandboxed, `abandoned` if the fix can't be made to work. Pairs with `self-improvement` (which promotes recurring heals to durable memory) but owns the verify-before-persist discipline self-improvement doesn't.
DEPRECATED: Opus 4.7's adaptive thinking covers most cases natively. Structured, reflective problem-solving through sequential chain-of-thought reasoning that replaced the Sequential Thinking MCP server. Retained as a reference pattern when deterministic, reviewable reasoning traces are required regardless of the model's adaptive-thinking choice.
Use when the user says 'get started with my site', 'connect to my wordpress site', 'what can you do on my site', or 'onboard'. Verifies connectivity, discovers the site architecture, maps the available MCP tools, and produces a site briefing.
Toolkit for creating and validating skills and subagents. Use when: creating a new skill (fast or full mode), validating an existing skill, deciding Skills vs Subagents, migrating docs to skills, estimating token cost, or running a security scan. Triggers: "create skill", "build skill", "validate skill", "new subagent", "skills vs subagents", "estimate tokens", "security scan".
Sets up a Ralph autonomous development loop for any project. First generates a full PRD from the user's description, then derives a task plan from it. Wraps Claude Code in an intelligent while-true loop with circuit breakers, exit detection, session persistence, and progress tracking. Use when you want Claude to autonomously work through a task list until done.
Connect an AI agent to SubRouter: device authorization, API keys, model selection by price, and OpenAI/Anthropic client configuration. Manage a provider's own model listings, pricing, and primary/backup channels, or a distributor station's listings, prices, packages, and settings. Use for "接入 SubRouter", "配置 SubRouter", "SubRouter API key", "用 SubRouter 的模型", "商家模型定价", "配置备用渠道", "管理我的分站", "分站上架", or "分站套餐"
Design and validate model-facing tool definitions with clear names, action-oriented descriptions, bounded JSON Schema parameters, explicit side effects, safe defaults, idempotency, errors, and realistic tests. Use when creating function-calling tools, MCP tools, agent actions, structured tool inputs, or when a model selects the wrong tool, invents arguments, or causes unsafe side effects.
Connect third-party AI agents (OpenAI, Bedrock, LangChain, Microsoft Foundry) to Twilio's communication channels using the Twilio Agent Connect SDK. Covers identity resolution, memory and context management via Conversation Memory, conversation orchestration via Conversation Orchestrator, multi-channel handling (Voice, SMS, RCS, WhatsApp, Chat), and AI-to-human escalation. Use this skill when integrating an existing LLM agent with Twilio services.
Vercel AI SDK for building chat interfaces with streaming. Use when implementing useChat hook, handling tool calls, streaming responses, or building chat UI. Triggers on useChat, @ai-sdk/react, UIMessage, ChatStatus, streamText, toUIMessageStreamResponse, addToolOutput, onToolCall, sendMessage.
Create a WeChat Mini Program project with AI capabilities from scratch. Triggered when the user wants to create a brand-new WeChat Mini Program (not adding features to an existing project). Integrates cloud development, database, login, payment, and other capabilities. Requires Node.js 18+ and the mp-skills CLI.
Plan, execute, document, and retest authorized security assessments of AI agents and multi-agent workflows using safe adversarial cases, synthetic identities, canaries, and evidence-based findings. Use when defining red-team rules of engagement, assessing prompt injection or excessive agency, testing tool and identity boundaries, evaluating memory or cross-agent attacks, scoring a campaign, or verifying remediation in an approved environment.
Provides guidance on coordinating multiple specialized teammates working in parallel. This skill should be used when the user needs to execute complex implementation plans, resolve cross-cutting concerns, or coordinate independent work streams requiring communication.
Reliability gate for answers. Forces evidence-based reasoning, explicit uncertainty, and “insufficient information” instead of guesses. Use when user says “be 100% sure”, “no hallucinations”, “only if verified”, “grounded answer”, or when stakes are high.
Create or update a project roadmap with major milestones. Generates the configured roadmap artifact (default .ai-factory/ROADMAP.md) — a strategic checklist of high-level goals. Use when user says "roadmap", "project plan", "milestones", or "what to build next".
Orchestration handoff bridge: emits and maintains sprint-status.yaml as the project's sequencing system-of-record. Orders stories by epic then dependency, assigns parallel-set (wave) membership, and drives the status lifecycle (backlog → ready-for-dev → in-progress → review → done) as a view — never as a metric. Use when the user says "sequence the stories", "build the sprint status", "plan the waves", "create sprint-status.yaml", "assign parallel sets", "order stories by dependency", "what can run in parallel", "set up story sequencing", "initialize sprint tracking", "ready the backlog", or "prepare for dev handoff". Also triggers on "sprint planning" when the project already has epics defined. SCOPE: SEQUENCING AND ORCHESTRATION ONLY. No velocity, no burndown, no committed points, no coverage metrics. Capacity is expressed as wave width (concurrent story count), not points. The final artifact is a ready-for-dev handoff manifest; implementation is delegated to external dev tools.
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.