Skills
96,981 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Use when a public website needs an AI assistant with source-bounded answers, RAG/content scope, injection protection, escalation, logging, rate limits, refusals, and no hallucinated product/legal/price claims. Planning or implementation handoff only; do not expose secrets or give destructive tools.
Coordinate SEO, LLM-readable site architecture, UX, content, monitoring, security, and external authority work as a task-plan-driven sequence. Use this skill when the user asks to plan, execute, validate, or route multi-skill website growth work; asks what to do next across semantic core, URL structure, internal linking, schema, llms.txt, UX journeys, analytics, citation monitoring, or white-hat link placement; or wants a central architect that preserves evidence, alarms, handoffs, and tests.
Use this skill when the user asks to test, validate, regression-check, or safely repair an existing Codex/agent skill's trigger, workflow, resources, or evaluation coverage. Do not use it to create a new skill, QA an application, or run open-ended skill optimization.
Compact an active task file without changing meaning. Use when converting a no-subtask task to subtask form, when adding a new subtask after earlier subtasks already contain full section content, when duplicated section content should be compacted before approval seeking, or when a task file has become too large to use safely.
Diagnose and operate Spring AI projects with version-aware Maven or Gradle checks for ChatClient, advisors, retrieval, conversation memory, tool/MCP boundaries, streaming, configuration, and observability. Use for setup compatibility, local smoke checks, and failure triage; do not use for provider-specific model evaluation, general Java/Spring development, security threat modeling, or production rollout governance.
Shared review-stage flow for PDCA hosts: full decision review, code-design review for code-affecting solutions, Standards∥Spec dual-axis verdicts (no cross-axis merge rank), binary pass/fail, bounded auto optimization, design summary, and verification-report honesty composed with completion-evidence-discipline. Used by PDCA workflows. Triggers — 「阶段审查流程」「方案审查编排」「审查闭环」「验证报告诚实」「双轴审查」 / staged review flow, dual-axis review, verification honesty. Do NOT use as a standalone task workflow.
Uses an intake, research, and clarify loop—including bounded subagent research when available—to produce a phased implementation plan in `PLAN.md` or `plans/<task-name>.md`. Use for complex features, refactors, migrations, architectural work, and other multi-step changes.
End-to-end setup for making a Telnyx AI assistant call a phone number. Covers provisioning a phone number, creating a TeXML application, assigning the number, configuring telephony settings, whitelisting destination countries, and triggering outbound calls via scheduled events. Use this skill (not telnyx-ai-assistants-python) when the task involves an AI assistant placing, making, or triggering an outbound phone call to a user.
Route high-level Twitter/X automation setup, readiness, starter-menu, and ambiguous workflow requests to independent flat peer skills after ThreadWave checks. Use for start ThreadWave, open ThreadWave, show the ThreadWave menu, or an exact standalone `threadwave` or `tw` request, plus Twitter automation setup, readiness, repair, and workflow selection. Do not use when the request already clearly asks to create/post tweets, create/send replies, or run the daily agent; activate twitter-post, twitter-reply, or twitter-agent directly. 中文：用于启动 ThreadWave、打开 ThreadWave 菜单、单独输入 `threadwave` 或 `tw`、推特自动化安装、就绪检查、修复和模糊工作流路由；明确的发推、回复或日常代理请求应直接使用对应的平级技能。
Decide whether an agent skill, plugin, or MCP server is safe to install into an AI coding agent, where its content is loaded into a model's context and its config can run on startup. Use when reviewing a skill pack, Claude Code / Cursor / Cline plugin, or MCP server before adoption; when a repo ships a SKILL.md, .mcp.json, or plugin.json you are about to trust; or when judging whether third-party agent content can steer the model or exfiltrate an engagement.
Create concise handoff artifacts for moving work between agent sessions, coding agents, or tools such as Codex and Claude Code. Use when pausing work, switching agents, splitting work across sessions, preserving current state, pointing to specs/plans/evidence, or preparing the next agent to continue without copying the whole conversation.
Use when you need to turn a vague idea into a confirmed design spec before implementation (new feature/component/behavior change). First check project context, then ask one question at a time, provide 2-3 options with trade-offs, finally output design in segments (~200-300 words each) with confirmation after each. Triggers: brainstorm, clarify idea, design spec, refine concept, requirement clarification.
Use when you need to ship a single PR-sized feature end-to-end (plan -> implement -> verify) with artifacts. Ship core product features quickly in a Next.js codebase: turn a feature idea into an executable plan, implement in PR-sized slices, and keep artifacts under runs/ (or OpenSpec changes/ when available). Supports plan-only mode for early scoping. For prototype UI work, include a demo-ready wow moment (animation/micro-interaction) by default unless user opts out.
Pre-work validation before starting implementation. Use when beginning work on an issue, feature, or fix to verify remote state, check for existing PRs, detect branch conflicts, and prevent redundant work. Catches problems that cause wasted effort: already-merged fixes, stale branches, unclean diffs.
Design and write high-quality Claude Code agents and agent prompts. Use when creating or updating .claude/agents/*.md for (1) single-purpose subagents (reviewers, implementers, researchers) and (2) workflow orchestrators (multi-phase coordinators like pr-review, feature-development, bug-fix). Covers delegation triggers, tool/permission/model choices, Task-tool orchestration, phase handoffs, aggregation, iteration gates, and output contracts. Also use when deciding between subagents vs skills vs always-on repo guidance.
Use when you have a finalized brainstorm-beagle spec at `.beagle/concepts/<slug>/spec.md` and need a bite-sized, TDD-driven implementation plan before any code is written. Triggers on: "write a plan", "plan this spec", "turn the spec into a plan", "now plan the implementation", "write-plan". Reads the spec, designs the file structure, decomposes work into 2-5 minute TDD steps with exact paths and commands, self-reviews against the spec, gets user approval, then writes to `.beagle/concepts/<slug>/plan.md` and offers to generate an execution handoff prompt via the subagent-prompt skill. Does NOT brainstorm specs, write code, or execute the plan — produces the plan document (and an optional handoff prompt) only.
Write structured implementation plans from specs or requirements before touching code. Use when given a spec, requirements doc, or feature description, when user says "plan this out", "write a plan for", "how should we implement", or before starting any multi-step coding task.
This skill should be used when the user asks to "create a 1C customization", "implement 1C functionality", "add a new capability to 1C", "develop a 1C module", "make a customization in 1С", or needs the full 1C customization development cycle from analysis to implementation with plan validation and acceptance checks.
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.
Create custom agents for Claude Code including YAML frontmatter, system prompts, tool restrictions, and discovery optimization. Use when creating, building, or designing agents, or when asked about agent creation, subagent configuration, Task tool delegation, or agent best practices.
AGENTS.md / CLAUDE.md best-practices advisor. For users asking about agents markdown, AGENTS.md, CLAUDE.md, Claude Code memory, or the format, structure, and best practices of AI coding agent instruction files; also for reviewing, diagnosing, rewriting, optimizing, or creating from scratch agent instruction files such as AGENTS.md, CLAUDE.md, CLAUDE.local.md, and .claude/rules. Not for general README writing, unless the goal is to give an AI coding agent project context.
Provision dedicated AI agents on AgentBox via x402 payment ($5 USDC on Solana). Use when creating cloud instances running OpenClaw AI gateways with HTTPS and web terminal. Requires Node.js and a Solana wallet.json with USDC funds. Covers: provisioning new instances, polling status, interacting via OpenAI-compatible chat completions, extending, and listing instances.
Design agentic workflows, AI agent operating loops, delegation patterns, human gates, tool boundaries, handoffs, evals, memory, review checkpoints, failure recovery, observability, and autonomy levels for product, engineering, research, support, marketing, operations, and skill-factory work. Use when creating repeatable workflows for AI agents or improving multi-step agent execution quality.
Karpathy-inspired autonomous research loop. Agent edits one file, evals, keeps or discards, repeats. Plateau-triggered web search breaks through ceilings. Git as state machine. Runs until stopped or budget exhausted. Supports claude, codex, and pi providers.
Use when the user wants more human-like AI responses — less robotic, less listy, more authentic. Triggers: 'behuman', 'be real', 'like a human', 'more human', 'less AI', 'talk like a person', 'mirror mode', 'stop being so AI', or when conversations are emotionally charged (grief, job loss, relationship advice, fear). NOT for technical questions, code generation, or factual lookups.
Automatically benchmark your custom memory implementation against established systems like Supermemory. Set up a public benchmark, or create your own. Compare solutions against quality, latency, features and cost, easily, with a simple UI and CLI.
Autonomous biomedical AI agent framework for executing complex research tasks across genomics, drug discovery, molecular biology, and clinical analysis. Use this skill when conducting multi-step biomedical research including CRISPR screening design, single-cell RNA-seq analysis, ADMET prediction, GWAS interpretation, rare disease diagnosis, or lab protocol optimization. Leverages LLM reasoning with code execution and integrated biomedical databases
This skill should be used before implementing features, building components, or making changes. It guides exploring user intent, approaches, and design decisions before planning. Triggers on "let's brainstorm", "help me think through", "what should we build", "explore approaches", ambiguous feature requests, or when the user's request has multiple valid interpretations that need clarification.
Universal ChromaDB integration patterns for semantic search, persistent storage, and pattern matching across all agent types. Use when agents need to store/search large datasets, build knowledge bases, perform semantic analysis, or maintain persistent memory across sessions.
Your EYES into the user's live Chrome browser and Electron apps. This skill lets you SEE and INTERACT with the user's actual browser or Electron app — their open tabs, logged-in sessions, and live page state. You MUST use this whenever the user's request involves browser content or Electron app inspection in ANY way. TRIGGER THIS SKILL when the user: - References pages they have open: 'I have X open', 'my tabs', 'open tabs' - Asks to look at, compare, or analyze anything in their browser: 'compare these pages', 'which looks better', 'check this page' - Mentions UI/visual analysis of live pages: 'dashboard', 'UI', 'layout', 'design quality' - Asks for screenshots or page inspection: screenshot, inspect, debug, check the page - Refers to 'the page', 'the browser', 'my tab' in any context - Mentions console errors, page state, or anything requiring browser access - Mentions Electron apps or CDP connections: 'Electron', 'electron app', 'CDP', 'CDP_PORT', 'DevTools Protocol', 'desktop app', 'remote-debugging-port' CRITICAL: NEVER say you cannot see the user's browser or ask users to paste screenshots. You CAN see their browser through this skill. Use `list` to discover open tabs, then `perceive` or `shot` to see page content. Do NOT use Playwright — it launches an isolated browser without the user's login state, cookies, or open tabs.
Use when a user's request is vague, ambiguous, or underspecified. Explores the codebase first, then runs an iterative Q&A loop grounded in the findings until ambiguity is gone. Outputs a clear, well-scoped context brief so the user can plan sharply. Triggers on "I want to...", "I need...", "let's build...", "can you help me...", "we should...", or any request where the full scope isn't immediately clear.
Build with Claude Messages API using structured outputs (v0.69.0+, Nov 2025) for guaranteed JSON schema validation. Covers prompt caching (90% savings), streaming SSE, tool use, model deprecations (3.5/3.7 retired Oct 2025). Use when: building chatbots/agents with validated JSON responses, or troubleshooting rate_limit_error, structured output validation, prompt caching not activating, streaming SSE parsing.
Build AI agents with Cloudflare Agents SDK on Workers + Durable Objects. Includes critical guidance on choosing between Agents SDK (infrastructure/state) vs AI SDK (simpler flows). Use when: deciding SDK choice, building WebSocket agents with state, RAG with Vectorize, MCP servers, multi-agent orchestration, or troubleshooting "Agent class must extend", "new_sqlite_classes", binding errors.
Multi-agent code review run as a debate. Critics raise findings in parallel, an Author defends the change, Critics concede or maintain, and a Judge rules on what stays contested. Findings are classified by kind and blocking, and handled durably when previously decided. Use `rq` to request a review of diffs (defaults to main branch), `rs` to respond to findings and record intentional non-fix decisions beside the relevant code. Triggers on "review this", "review my code", "code review", "check for bugs", "audit this", when examining PRs, pull requests, branches, or diffs. Always asks user before applying fixes or adding decision comments
Use when the user wants Claude to send images to OpenAI Codex — Codex-side review of an existing screenshot/mockup/wireframe, generate an image via Codex's image_gen tool, or edit an existing image via Codex — or explicitly invokes "codex-vision" / "/codex-vision". Wraps the upstream `codex exec -i` CLI with three modes (review / generate / edit) and an opt-in tmux session prefixed `claude-codex-` so the user can attach and watch. SKIP when the task is text-only, or when the user needs a durable multi-turn repo task — use codex-tmux-dev for long-running image+coding work. The bare word "codex" is not a trigger; the bare word "screenshot" is not a trigger. For generate/edit, require explicit Codex intent unless the user invoked the skill by name. Explicit `/codex-vision` invocation overrides every other gating rule below.
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.