Skills
100,928 packaged capabilities and playbooks for agents. Install any of them with one command.
All
Produce high-fidelity design mockups using HTML as the medium (landing pages, slides, interactive prototypes, animations, design systems, infographics, mobile mockups, etc.). Act as a professional designer rather than a generalist frontend engineer, emphasizing asking before doing, referencing existing design systems, offering multiple variants, and avoiding AI-style clichés. Triggers: /axi-front-design, "做个落地页", "设计一张海报/封面", "做张幻灯片/PPT", "做个交互原型", "做个动画", "design a landing page", "make a deck", "prototype this UI", "design system", "hi-fi mockup". Not applicable: plain text content/logic code, needs real image generation (use an image generation model instead), needs native PPTX/Figma files (HTML preferred).
Use when writing, reviewing, or refactoring TypeScript/Node.js backend application or API code where backend design guidance is useful, including APIs, routes, controllers, middleware, SDKs, services, repositories, database access, runtime workers, and external integrations that implement backend behavior. Do not use merely because code runs in Node.js or on a server. Skip CI/workflow configuration, deployment or maintenance scripts, demos, tests, simple renames, formatting, documentation-only work, and other mechanical or supporting work that does not require backend design guidance. Do not use for localized bug fixes when the cause and edit location are already known, even when the affected code is part of an SDK or backend component.
Production backend systems development. Stack: Node.js/TypeScript, Python, Go, Rust | NestJS, FastAPI, Django, Express | PostgreSQL, MongoDB, Redis. Capabilities: REST/GraphQL/gRPC APIs, OAuth 2.1/JWT auth, OWASP security, microservices, caching, load balancing, Docker/K8s deployment. Actions: design, build, implement, secure, optimize, deploy, test APIs and services. Keywords: API design, REST, GraphQL, gRPC, authentication, OAuth, JWT, RBAC, database, PostgreSQL, MongoDB, Redis, caching, microservices, Docker, Kubernetes, CI/CD, OWASP, security, performance, scalability, NestJS, FastAPI, Express, middleware, rate limiting. Use when: designing APIs, implementing auth/authz, optimizing queries, building microservices, securing endpoints, deploying containers, setting up CI/CD.
Fetch and display BBC News stories from various sections and regions via RSS feeds. Use when the user asks for BBC news, UK news headlines, world news from BBC, or news from specific BBC sections (technology, business, politics, science, health, entertainment, regional UK news, or world regions).
Manage files on Baidu Netdisk. Supports uploading, downloading, saving to Netdisk, sharing, searching, moving, copying, renaming, and creating folders. Trigger when the user mentions “百度网盘,” “bdpan,” or “网盘” in connection with file operations.
Given a sentence or paragraph, find real academic literature opposing it—contrary conclusions, boundary conditions, alternative explanations, methodological critiques, or failed replications. Rank by threat level; for each paper, include one line on “how it threatens this claim” and “how to respond to it.” Uses real searches (scimaster-cli); never fabricates counterexamples. **Trigger this skill proactively in the following cases**: the user wants to “get ahead of reviewers,” “find counterexamples to this conclusion,” “see whether this topic is controversial,” “find research that contradicts this,” “help me attack this argument,” “check how robust this conclusion is,” or “find the objections.” Even without saying “bear-counter,” use it whenever the intent is **to find academic evidence opposing a claim**. bear-support and bear-counter work best as a pair: run both on the same passage to see both sides. Not for: finding supporting literature (use bear-support), checking topic overlap (use bear-scoop), concept maps (use bear-map), or tracing the history of an idea (use bear-trace).
Given a research idea, perform collision detection, quiet-zone support searches, and potential challenge searches in sequence, then present the pre-proposal evidence landscape in one comprehensive report: collision risks, literature support in quiet zones, and the challenges that most need to be considered in advance. Uses real searches (scimaster-cli); never fabricates citations. **Trigger this skill proactively in the following cases**: the user wants to “evaluate before proposing,” “see whether this idea is worth pursuing,” “do a comprehensive literature review before project approval,” or “assess this research direction.” Even without saying “bear-propose,” use it whenever the intent is **a comprehensive assessment before formally investing in a research direction**. Not for: collision checks only (use bear-scoop), support searches only (use bear-support), opposition searches only (use bear-counter), concept maps (use bear-map), or tracing the history of an idea (use bear-trace).
Given a claim, search for both supporting and opposing literature, then present the evidence landscape on both sides in one comprehensive report: the strongest supporting evidence, the most serious opposing challenge, angles with support but no opposition, and angles with opposition but no support. Uses real searches (scimaster-cli); never fabricates citations. **Trigger this skill proactively in the following cases**: the user wants to see “both support and opposition,” asks “how robust is this claim,” wants “a comprehensive assessment of an argument,” or asks to “look for both sides.” Even without saying “bear-review,” use it whenever the intent is **a complete analysis of the evidence landscape for a claim**. Not for: support only (bear-support), opposition only (bear-counter), topic-overlap checks (bear-scoop), concept maps (bear-map), or tracing the history of an idea (bear-trace).
Given a sentence or paragraph, find real academic literature supporting it, with one line for each paper explaining how it supports the claim. Uses real searches (scimaster-cli); never fabricates citations from memory—if nothing is found, say so. **Trigger this skill proactively in the following cases**: the user wants a claim to be “better supported,” “more robust to scrutiny,” wants to “add literature support,” “find a few papers,” asks “is there research on this claim,” “check this conclusion,” or wonders “will reviewers say there are no citations?” Even without saying “bear-support” or “find literature,” use it whenever the intent is **to find academic support for a claim**. Not for: finding opposing literature (bear-counter), checking topic overlap (bear-scoop), concept maps (bear-map), or tracing the history of an idea (bear-trace).
A library of 35 curated colour palette styles for building beautiful, editable Feishu / Lark (飞书) whiteboards from SVG. Use this whenever the user wants to create or draw a Feishu whiteboard, infographic, diagram, poster, or visual explainer and wants it to look polished, on brand, or in a particular aesthetic, or when they name a style. The agent figures out what board the user wants, asks about the visual vibe, picks a fitting style from the catalogue, generates an editable Feishu whiteboard, returns the doc link and the rendered image, and offers to switch styles. Requires lark-cli (npm @larksuite/cli) installed and authenticated, and a Feishu/Lark account.
Codex skill that converts images, screenshots, posters, PPT page screenshots, HTML or SVG designs into editable PPTX. Use this skill when the user needs image2ppt, image2pptx, image to PPT, screenshot to PPT, PNG/JPEG to editable slides, HTML/SVG to PPTX, splitting a design into image components and text boxes and reassembling them into PowerPoint, or rebuilding a flat image into a multi-element editable PPT using Codex's built-in image generation. When handling binary images in Codex, by default combine with the imagegen skill: first use Codex vision to understand the layout, then use the built-in image generation/editing capabilities to generate or clean up multiple component images, and finally assemble the PPTX with this skill's scripts.
Generate chapter lists for Bilibili videos. Trigger scenarios: (1) need to create Bilibili chapters for a video (2) the user says "convert to Bilibili format", "generate chapters", "generate Bilibili chapters" (3) need to generate video segments from subtitles (4) handle video progress-bar segment markers
BKN full-lifecycle orchestration entry point. Reads OpenBKN SDK / CLI operation references on demand. Whenever the task involves creating a BKN, extracting object relations from a PRD/document, generating or updating `.bkn`, data view binding, environment checks, test set generation, validation, or pushing, use bkn-creator first for workflow routing, stage gates, sub-skill orchestration, and result receipts. Applies to BKN create/read/update/delete/extract/copy/validate scenarios, skill package capability gap-filling / skill draft generation scenarios, and usage feedback inspection and improvement scenarios (scheduled task triggers, Agent conversation quality anomalies, feedback_brief input, continuous knowledge network optimization). Not for pure data semantic queries, which should be handled by data-semantic.
Explains how to use Salesforce BLIP-2 (Q-Former bridging a frozen image encoder and an LLM such as OPT or FlanT5) through HuggingFace Transformers and LAVIS for image captioning, visual question answering, image-text matching, and feature extraction. Use when generating captions for images, building a VQA system, doing zero-shot image-text understanding without task-specific training, matching or retrieving images against text, or fitting a BLIP-2 model into limited GPU memory with INT8/INT4 quantization. Prefer LLaVA or InstructBLIP for instruction-following multimodal chat, and CLIP for plain image-text similarity.
WebGL "zoom blur" effect for full-screen background image transitions (Three.js + custom GLSL shader, mouse parallax, scroll/click/key navigation) for hero sections/image slideshows. Call when the user says "hiệu ứng blur ảnh", "chuyển ảnh mờ dần kiểu zoom", "webgl blur transition", "gallery ảnh full-screen kiểu blur", or /blur. DIFFERENT from `dark-mode-maker` (circle-reveal theme switch) — this is an IMAGE TRANSITION effect, not a theme switch.
Genre pack: raw mechanical interface — Swiss typographic print fused with military/industrial terminal aesthetics. Rigid visible grids, extreme type-scale contrast, utilitarian color (paper, ink, one signal color), hard edges, zero decoration. Use for data-heavy platforms, developer tools, portfolios, studios, or any brief wanting to feel like a declassified blueprint. Stack on the design-taste core skill.
Turn a plain-language request into a working, tested Agenta agent through the Agenta API. Use when the user wants to build, create, schedule, or test an Agenta agent from their coding agent (Claude Code, Codex, Cursor). Writes one config and runs a few bundled shell scripts. Triggers on "build an Agenta agent", "create an agent on Agenta", "schedule an Agenta agent", or wiring an agent to a tool like GitHub or Slack via Agenta.
Build, edit, test, enable, and monitor PostHog workflows over MCP. Author the action/edge graph so it runs and opens cleanly in the visual editor, then change drafts surgically with patch operations. Use when asked to build, set up, automate, change, fix, or debug a workflow, campaign, broadcast, drip sequence, or event-triggered automation in the workflows product, including first-time messaging setup (email channel, sending domain, frequency cap) before the first workflow.
Manage local development proxies using Caddy server's Admin API. Use when users need to allocate localhost ports, create reverse proxies for local dev servers, track which ports are in use across Claude Code sessions, or access apps via friendly URLs like myapp.localhost. Triggers on "set up a proxy", "allocate a port", "what's running on localhost", "register my app", or any Caddy configuration task.
Creates museum-quality visual art as .png or .pdf by first generating a design philosophy manifesto, then expressing it on canvas with precise composition, color palettes, and typography. Use when the user asks to create a poster, art piece, visual design, or static creative piece. Outputs original designs using design-forward principles.
CarPlay framework for iOS in-car applications — audio, communication, navigation, parking, EV charging, quick food ordering, fueling, driving task, public safety, and voice-based conversational apps. Includes widgets, live activities, CarPlay Ultra, and instrument cluster support.
Helps agents write and review Scala Cats Effect IO code. Use for suspending side effects and non-determinism in IO or F[_], choosing Sync/Async/Temporal/Concurrent/Clock/Random, handling blocking I/O, testing with TestControl, and composing resources, fibers, races, and structured concurrency safely.
Classifies a single piece of text (an AI-coding human input, optionally with its paired assistant reply and previous-assistant tail) into the current v2 human-input category + topic taxonomies, returning the primary category, the full priority-ordered categories array, one topic, topic_reason, and topic_confidence — a pure reasoning check against uid.core-product's classify rules (gpt-4o production prompt, postprocess off), no CawPlan data or API calls involved. Use when: asked to classify/categorize a specific sentence or human input against the current category and topic rules — e.g. "what category/topic is this: ...", "classify this with this assistant reply" — or as the per-row classification step used by cawplan-internal-qa-coding-humaninputs-test. NOT for: bulk/batch accuracy testing across many already-uploaded human inputs, or fetching data from CawPlan at all (use cawplan-internal-qa-coding-humaninputs-test for that), submitting reports, or creating tickets.
Internal QA check for the AI-coding human-input classifier: pulls already-uploaded human inputs via the cawplan CLI, classifies each one via cawplan-internal-qa-coding-humaninputs, and compares category and/or topic against persisted cloud labels — in category-only, topic-only, or both-together mode — reporting accuracy and concrete mismatches for manual review. Use when: asked to test/verify/check human-input classification accuracy — e.g. "test today's category accuracy", "check topic only for spx last week", "compare both category and topic for the last 2 days" — optionally scoped to one person, one product, or both. NOT for: classifying a single ad hoc sentence (use cawplan-internal-qa-coding-humaninputs directly), submitting coding reports (use cawplan-coding-commit), general cost/usage insights or prompt-quality scores (use cawplan-coding-insights), or creating tickets.
Internal QA check for the new AI-coding session-summary feature: fetches one session's full conversation via the cawplan CLI and generates a narrative summary yourself using the same rules uid.core-product's session-insights prompt uses, so you can eyeball the summary's quality before/without needing the cloud enrichment pipeline to have run. Display-only, no comparison against any cloud value (the field is new and may not be backfilled yet). Use when: asked to test/check/try out the session summary prompt — e.g. "test the summary for yupeng's session today", "see what the summary would look like for entry <id>", "try the session summary on a random recent session". NOT for: human-input category classification (use cawplan-internal-qa-coding-humaninputs[-test]), bulk accuracy testing, submitting reports, or general session cost/usage insights.
Use when a change rationale just emerged—a revert, hotfix, fix chain, workaround, or abandoned approach—and you want guided facilitation (options, consequences, provenance) before it becomes a ChDR. Not for mining git history (/change-init) or reviewing drafts (/change-clarify); class boots only suggest this skill, they never write drafts.
Screen and evaluate A-share investment opportunities. Adapts the original deal-screening skill for Chinese market data sources, A-share screening criteria, and domestic deal origination. Triggers on "A股标的筛选", "标的筛选", "screening China", "deal screening A-share", "筛选投资标的", or "screen for [criteria]".
Accretion/dilution and merger consequences analysis for A-share M&A transactions. Adapted from the original merger-model skill for Chinese GAAP, CAS accounting, and China market transaction structures. Triggers on "A股并购分析", "并购模型", "accretion dilution China", "merger model A-share", or "M&A consequences [company]".
Claude Code integration for OpenClaw. This skill provides interfaces to: - Query Claude Code documentation from https://code.claude.com/docs - Use Spec-Driven Development with spec-kit - Manage subagents and coding tasks - Execute AI-assisted coding workflows - Access best practices and common workflows Use this skill when users want to: - Get help with coding tasks - Query Claude Code documentation - Use spec-kit workflow for structured development - Manage AI-assisted development workflows - Execute complex programming tasks
Use this skill to drive Jetpack Compose UI from tests with the high-level action APIs that do not go through a gesture builder — performClick, performScrollTo, performScrollToIndex, performScrollToKey, performScrollToNode, requestFocus, performSemanticsAction, and performFirstLinkClick. Covers picking the correct receiver node (the scrollable container vs an item), the matchers used to find a scroll parent (hasScrollAction, hasScrollToIndexAction, hasScrollToKeyAction, hasScrollToNodeAction), and how the lazy vs non-lazy scrollable cases differ. Use when the developer asks "how do I tap a Compose node", "scroll to an item in a LazyColumn", "click a link inside Text", "trigger a custom semantics action", or reports "Action performScrollTo failed" / "node has no parent layout with a Scroll SemanticsAction" / "ScrollToIndex not defined".
Analyze codebases for anti-patterns and code smells via ast-grep structural matching across JS, TS, Python, Vue, React. Use when reviewing code quality, identifying tech debt, or doing broad code analysis. Pairs with /code-antipatterns.
Authoritative framework for reviewing the design quality of proposed code changes — before implementation. Evaluates architecture-level quality attributes (testability, modularity, reliability, scalability, dependency direction via ISO 25010 + Clean Architecture + SDP) and code-level metrics (accidental complexity, coupling via Myers/Connascence, cohesion, change amplification, tech debt, cyclomatic/cognitive complexity, Law of Demeter) plus a security pass (OWASP Top 10). Use this skill whenever a proposed solution involves code changes and you need to assess whether the code is well-designed — not just whether it works, but whether it is maintainable, testable, and does not introduce design debt. Triggers — 「代码审查」「代码设计审查」「代码设计质量」「审查代码设计」「代码架构审查」「设计质量评估」「代码质量评审」「这个代码设计合理吗」「耦合度审查」「代码可维护性」 / code design review, code architecture review, design quality assessment, coupling analysis, maintainability review.
Shared vocabulary for designing deep modules. Use when the user wants to design or improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable or AI-navigable, or when another skill needs the deep-module vocabulary.
Turn a ~5-second voiceover line, opinion sentence, or abstract concept into a premium editorial halftone paper-collage B-roll clip — optionally with a fitted voiceover. Load this skill whenever the user says "collage b-roll", "paper collage b-roll", "halftone collage", "collage-style visuals", "collage explainer" (or the Chinese triggers 纸拼贴 b-roll / 半调拼贴 / 拼贴风格配画面), or wants a script line turned into a collage visual metaphor. Enforces a mandatory three-gate approval flow — propose visual metaphors only, generate color collage stills only after the user confirms, and only after the stills are confirmed generate the assemble-from-empty animation. Video model is gemini-omni-flash, locked — no other video model is ever used and the choice is never offered to the user
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.