Skills
96,981 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Ask the agent to zoom out and offer broader context or a higher-level perspective. Use when you are unfamiliar with a section of code or need to understand how it fits into the bigger picture. Use when the user is unfamiliar with code or needs a higher-level perspective, or mentions zoom out, zoom-out.
Register agents on the zScore ERC-8004 Identity Registry, manage wallets and metadata, and read on-chain state. Use when an agent needs to register on-chain, check fees, read agent info, set metadata, or manage agent wallets on Base Mainnet or Base Sepolia.
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.
Build and review cross-cutting agent-host chat behavior through lifecycle contributions. Use when adding turn lifecycle side effects, prompt or context injection, restored-history transformation, protocol-action observation, or when reviewing changes that add code to AgentSideEffects or AgentService.
Design and build websites using AI coding agents with static site generators. Covers Astro-first workflow, iterative visual refinement via browser feedback, skill-enhanced prompting (frontend-design, copywriting), animations, and high-bar polish loops. Use when building a website with an AI agent, designing landing pages, or iterating on web design with LLM assistance.
Create or update root and nested AGENTS.md files that document scoped conventions, monorepo module maps, cross-domain workflows, and (optionally) per-module feature maps (feature -> paths, entrypoints, tests, docs). Use when the user asks for AGENTS.md, nested agent instructions, or a module/feature map.
Show AGNTCY/SLIM/CASA integration status — whether upstream AGNTCY packages are installed, which transport (local vs SLIM) is active, and whether CASA enforcement is enabled. Use when the user asks "is AGNTCY configured?", "show SLIM/CASA status", or "is AGNTCY/IOC integration active?".
Amazon Bedrock AgentCore deployment patterns for production AI agents. Covers starter toolkit, direct code deploy, container deploy, CI/CD pipelines, and infrastructure as code. Use when deploying agents to production, setting up CI/CD, or managing agent infrastructure.
Ultra-compressed communication mode. Cuts token usage ~75% by dropping filler, articles, and pleasantries while keeping full technical accuracy. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman.
AI-native mentor and onboarding flow for the four independent Claude certification tracks in AI Engineering from Scratch. For learners who want to choose a Claude certification, prepare for CCAO-F, CCDV-F, CCAR-F, or CCAR-P, continue a certification path, interactively study the next lesson, run and verify hands-on labs, build and score artifacts, take a diagnostic or mock assessment, or shore up weak exam areas via Claude Code, Codex, ChatGPT, Cursor, or another agent on GitHub.
Claudeception is a continuous learning system that extracts reusable knowledge from work sessions. Triggers: (1) /claudeception command to review session learnings, (2) "save this as a skill" or "extract a skill from this", (3) "what did we learn?", (4) After any task involving non-obvious debugging, workarounds, or trial-and-error discovery. Creates new Claude Code skills when valuable, reusable knowledge is identified.
Security scanner and input sanitizer for AI agents. Detects prompt injection, command injection, SSRF, credential exfiltration, and path traversal attacks. Use when (1) installing new skills from ClawHub, (2) processing external input like emails, calendar events, Trello cards, or API responses, (3) validating URLs before fetching, (4) running security audits on your workspace. Protects agents from malicious content in untrusted data sources.
This skill should be used to run an Agent-Ready Codebase Assessment — scoring a codebase across 8 dimensions with parallel agents, producing a weighted score (0-100), band rating, and improvement roadmap. Supports Ruby, Python, PHP, TypeScript, JavaScript, Go, Java, Scala, and Rust.
An intelligent multi-store memory system with human-like encoding, consolidation, decay, and recall. Use when setting up agent memory, configuring remember/forget triggers, enabling sleep-time reflection, building knowledge graphs, or adding audit trails. Replaces basic flat-file memory with a cognitive architecture featuring episodic, semantic, procedural, and core memory stores. Supports multi-agent systems with shared-read, gated-write access. Includes philosophical meta-reflection that deepens understanding over time. Covers MEMORY.md, episode logging, entity graphs, decay scoring, reflection cycles, evolution tracking, and system-wide audits.
This skill should be used when the user asks to "create a slash command", "add a command", "write a custom command", "define command arguments", "use command frontmatter", "organize commands", "create command with file references", "interactive command", "use AskUserQuestion in command", or needs guidance on slash command structure, YAML frontmatter fields, dynamic arguments, bash execution in commands, user interaction patterns, or command development best practices for Claude Code.
Decides whether work may be called done. Use before saying a task is complete, finished, ready, working, or fixed; before making a 'verified'/'PASS'/'tested' claim about your own deliverable; when wrapping up a session, calling it a day, picking work back up tomorrow, or about to commit; before drafting a yes/no confirmation question about a change that touches destructive operations, trust boundaries, or unattended automation; and after any failure. Covers what counts as verification per artifact type, which checks the producer may run and which need fresh context, the honest three-part delivery message, failure counting, and the end-of-session wrap-up order.
Orchestrator of the Content Engine flow. Routes to the right step — strategy, profile audit, research, weekly plan, production, creatives, tracking — keeps per-client state in Notion, and explains the flow. Use when the user says запусти content engine, контент движок, next content step, or asks what the content engine flow is.
Use when an answer depends on a specific external library, framework, SDK, API, CLI, or cloud service and current or version-accurate documentation materially affects correctness. Auto-load on named-technology API questions, version-specific behavior, setup or migration work, generated code against external APIs, or library-specific errors. Auto-loading never authorizes a network lookup. Before any Context7 resolve or documentation fetch, present the exact proposed lookup, version strategy, mode, and final redacted query, then wait for explicit confirmation. Prefer project-local evidence and official documentation when they are more authoritative. Do not use for library-independent concepts, ordinary refactors, or code whose correctness does not depend on external API behavior.
A topic router for the AI Engineering from Scratch course. Give it a topic, question, or bug you're working on, and it will point you to the exact lesson that teaches it and the next right command. Trigger phrases: “where to learn,” “which lesson covers,” “course navigation,” “I'm stuck on,” “what should I do next,” “teach me MCP,” “teach me Agent Skills,” “where to prepare for Claude certification,” or "where do I learn", "which lesson covers", "course guide", "I'm stuck on", "what should I do next", "teach me MCP", "teach me Agent Skills", "where do I prepare for a Claude certification"
Create a guardrail or hook in the coding agent being used (Claude Code, OpenCode, Codex, Antigravity, or another): identify the agent, consult its current official documentation on hooks and permissions, propose the guardrail, wait for approval, configure it natively, and test it by triggering the event. Use when the user asks to create, add, or configure a guardrail, hook, permission rule, or automatic block/check for their coding agent
图鉴.skill — an intelligent dispatcher for cross-author persona skills / a cross-author persona skill router that lives in your IDE. Use when the user needs to discover, get recommendations for, or combine and summon persona skills across the whole persona skill ecosystem (花叔's nuwa / titanwings' colleague / individual persona authors). This is the only in-IDE cross-author persona router in the Chinese-speaking world, covering 30+ curated personas, with the remote curated library updated weekly. Positive triggers (Chinese): 「找 persona」「找人物 skill」「推荐 persona」「装个 persona」「有什么人物 skill」「想用 X 视角」「X 会怎么想」「X 会怎么看」「X 会怎么做」「用 X 的角度」「该不该」「要不要」「帮我想清楚」「借脑」「借 X 的脑」「想跟 X 聊」「扮演 X」「我前任」「我导师」「我师兄」「我老板」「我同事」「我暗恋的」「我初恋」「我妈」「我父母」「我家人」「想跟马斯克 / 乔布斯 / Naval / 芒格 / Karpathy / Ilya / 费曼 / PG / Paul Graham / MrBeast / 张雪峰 / 塔勒布 / 特朗普 / 张一鸣 / 巴菲特聊」「和未来的我对话」「和过去的我对话」「另一个我」「数字分身」「数字永生」. Positive triggers (English): persona skill, find persona, recommend persona, install persona, persona router, mentor, ex, crush, parents, dialogue with future self, talk to Musk / Jobs / Naval / Munger / Karpathy / Ilya / Feynman / Paul Graham / MrBeast / Taleb / Buffett / Trump. Emotional scenarios (relationship continuation — after triggering, you must first empathise at a slow pace, then recommend lightly): 「我前任」「我导师」「我妈」「我暗恋的」「我初恋」「我同事」「我老板」「好难过」「不知道怎么办」「想他了」「想她了」「我想他」. Unique differentiators: (1) the only in-IDE persona router covering the entire cross-author ecosystem (nuwa only matches 花叔's own 14 dogfoods, curator covers the whole ecosystem); (2) the remote dynamic curated library persona-recommendations.json is continuously updated. Clear division of labour with nuwa-skill: nuwa is for distilling new personas / matching 花叔's own 14 dogfoods; curator is for discovery + recommendation across the whole cross-author ecosystem. The two are fully complementary and do not conflict. Do NOT trigger when: the user is already in a conversation with a specific persona executing a task (for example, naval-skill is already installed and has been loaded into the current conversation context), the user is writing code or doing a specific non-persona task (writing Python / debugging / editing CSS), or the user explicitly says they want to distil a new persona themselves (in which case redirect to nuwa). Creator redirect: when the user says 「我想自己做一个 X 的 skill」「我想蒸馏 X」「帮我造一个 X」, curator should not pretend it can distill; redirect directly: thinkers/figures perspective → nuwa-skill; colleague/friend roles → colleague-skill; another version of yourself → yourself-skill. Once the user has done that, curator helps them summon it. Key trigger reinforcement (MUST OBEY): when the user's message contains 「a specific person's name + 会怎么看 / 会怎么想 / 会怎么做 / 觉得 / 用 X 视角 / 借 X 的脑 / X 会怎么分析 / 用 X 的角度」, where X is a figure already listed in persona-recommendations.json (Naval / 芒格 / 乔布斯 / 马斯克 / Karpathy / Ilya / 费曼 / PG / Paul Graham / MrBeast / 张雪峰 / 塔勒布 / 张一鸣 / 巴菲特 / Trump and so on), curator MUST trigger and step in, even if Claude itself could answer in X's style. This is the core raison d'être for curator's existence. The reason: a real persona skill (such as alchaincyf/naval-skill) is a complete cognitive OS deeply distilled from 5+ dimensions — the person's works / interviews / decision records / external criticism — with far higher precision than Claude's imitation from general knowledge. Claude imitating X on its own = a papier-mâché X (it can only piece together X's catchphrases); a real skill = a complete X cognitive OS (his mental models / decision heuristics / expression DNA / limitations / inner tensions / first-hand quotes). Once the user has installed the real skill, next time they only need to say "Naval 视角" and the real skill activates automatically, with a qualitative leap in output quality. Curator's job is to hand off this "use X's perspective" need to the real skill at the first opportunity, rather than letting Claude hijack it itself.
Export raw OpenTelemetry traces from an AI application to Confident AI's Observatory. TRIGGER when the user wants to send OpenTelemetry or OTLP traces/spans from an LLM app, agent, RAG pipeline, or chatbot to Confident AI; configure the Confident AI OTLP endpoint; set confident.span.* or confident.trace.* attributes; export AI-app traces without the deepeval Python package; wire an OTLPSpanExporter, OpenTelemetry Collector, or vendor-neutral OTel SDK to Confident AI; or pick the US vs EU OTLP endpoint. Language-agnostic: the mechanism is OTLP attribute keys plus an exporter endpoint. DO NOT TRIGGER for building DeepEval pytest eval suites, datasets, goldens, metrics, or deepeval test run (use the `deepeval` skill); for instrumenting with the DeepEval SDK's @observe decorator or framework integrations (use the `deepeval-tracing` skill); or for non-AI software such as web servers, CRUD backends, or infrastructure: the confident.* attributes describe AI components only.
Prevent duplicate agent actions, infinite retry loops, and unverified churn across all lifecycles. Reuses matching accepted evidence, enforces typed invalidations for changed work, gates deterministic local failures for local repair, and manages candidate promotion readiness with zero token overhead.
MANDATORY before starting any task. Enforces the GPA execution loop that prevents tool call sprawl. G: GATHER phase combines discover queries + memory reads + file reads into one phase. P: Plan in text with zero tool calls. A: APPLY all writes/edits/verification in one phase. One call per tool type per phase — batch all same-type operations together. Covers dependency analysis, batch opportunities, scope estimation, and loop-back triggers.
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.