Skills
97,501 packaged capabilities and playbooks for agents. Install any of them with one command.
All
Drive external CLI coding agents (Claude Code, Codex, OpenCode) as first-class AG2 Agents over the Agent Client Protocol (ACP): point an Agent at a ClaudeCodeConfig / CodexConfig / OpenCodeConfig preset from ag2.acp and ask()/run() it like any agent, with its thinking, tool calls, plans and permission prompts externalized onto AG2's event stream. Use when orchestrating, observing, or gating CLI coding agents from Python — permission_policy ask/auto/deny HITL gating, fs_root confinement, turn timeouts, in-process testing via fake_acp_config. To expose an AG2 agent to other systems instead, see ag2-a2a or ag2-mcp; for approval plumbing see ag2-hitl.
Browser automation for AI agents via inference.sh. Navigate web pages, interact with elements using @e refs, take screenshots, record video. Capabilities: web scraping, form filling, clicking, typing, drag-drop, file upload, JavaScript execution. Use for: web automation, data extraction, testing, agent browsing, research. Triggers: browser, web automation, scrape, navigate, click, fill form, screenshot, browse web, playwright, headless browser, web agent, surf internet, record video
Bootstrap and manage the .ai/ shared context directory for multi-agent projects. Use when initializing agent handoff, bootstrapping .ai/ files, refreshing stale handoff state, creating or repairing session files, or when the user says "agent handoff", "bootstrap", "initialize", "set up handoff", "what did the last agent do", "continue where X left off", or "resume".
Guides development workflow, design system, scripts, and component inventory for @ant-design/agentic-ui. Use when onboarding, adding new components, running build/test/docs, or asking about project structure, conventions, or component list. Triggers on agentic-ui, development workflow, 开发流程, 组件清单, 设计系统, 新组件, build, test, docs, AGENTS.md.
ALWAYS load before finalizing ANY written prose, and whenever text is suspected of being machine-written - "does this sound like AI", "make this sound human", "clean up the AI-isms", "did a bot write this", "check this draft before I send it", "review this for AI tells", "this reads like ChatGPT", "review this for mannered prose", "is this too writerly", "is this student paper AI-generated", or before handing back any draft, memo, email, or article you wrote. Use proactively even if the user never mentions AI writing.
Use this skill when a user wants to report, collect, normalize, classify, or template feedback about problems encountered while using AI tools in any scenario, including coding agents, chatbots, office assistants, search, writing, data analysis, or internal AI systems. The skill converts free-form problem descriptions into a structured feedback report with explicit labels, problem category, severity, scenario, impact, and follow-up questions.
Use this skill when reviewing AI-generated unit, functional, API, or end-to-end tests for false confidence, weak assertions, missing risks, or unsafe test behavior; triggers include AI 生成测试评审, AI-generated test review, and functional test review
Direct ALGOPACK HI2 API workflows for Herfindahl-Hirschman market concentration, hhi_* metrics, daily EQ/FO/FX concentration rows, raw hi2 endpoints, curl, JSON/CSV handoff, simple HTML output, ISS columns/data normalization, start pagination, and interpretation bands.
Direct ALGOPACK Mega Alerts API workflows for market anomaly alerts, alert_type thresholds, reference JSON, documented EQ/FO alert endpoints, curl, JSON/CSV handoff, simple HTML output, ISS columns/data normalization, start pagination, and observed abnormal volume or price signals.
Alibaba Cloud LoongCollector / SLS installation, collection onboarding, Pipeline config management and validation, machine groups, permission troubleshooting, and Lens queries. HARD RULE: for matching requests, the first tool MUST load this skill before any SSH probe, directory setup, checklist/file write, or cloud read. Triggers: "安装 LoongCollector", "ECS 安装采集器", "自建 Linux 主机安装 LoongCollector", "ACK 安装 loongcollector", "自建 K8s 部署采集", "从安装到能查到日志", "SLS 日志采集接入", "SLS 日志采集接入相关的事", "修改采集配置", "改采集配置", "采集配置校验", "validate_pipeline.py", "SLS 机器组", "新建 Logtail Pipeline 采集配置", "Logtail Pipeline", "ClusterAliyunPipelineConfig", "SLS Lens 查询", "无数据排查", "心跳异常", "SLS 采集权限排查", "SLS 权限排查", "阿里云 CLI 凭证没有 SLS 操作权限", "Logtail", "iLogtail", "AgentSight", "Agentloop", "input_agentsight", "eBPF Runtime", "ebpf-event".
Alibaba Cloud WAF Bot Management automated configuration assistant. Automatically completes bot protection policy evaluation, configuration, verification, and tuning via OpenAPI/CLI. Supports scenarios such as LLM API protection, retail promotion anti-scalper, general anti-crawling, and academic research platforms. Triggers when the user mentions Bot management, crawler protection, anti-crawling, WAF Bot, BOT2.0, LLM API abuse prevention, retail anti-scalper, promotion protection, SMS bombing, script attacks, or automated traffic protection.
devtools usage for alova. Use this skill whenever the user mentions alova openapi configuration, @alova/wormhole, API code generation, OpenAPI/Swagger with alova integration, alova devtools, or the alova VSCode extension. Trigger even for questions like "how do I use OpenAPI with alova" or "how do I generate API code with alova".
(NS) Verify internal consistency of requirements.md and adherence to project rules before task generation. Use after scope confirmation, when the user asks to analyze or validate requirements, or after manual edits to requirements — automatically between Gate 2 and task generation. Produces consistency-report.md; does not modify requirements. Do NOT generate tasks.
This skill should be used when starting a local Andocs demo or preview, introducing or presenting Andocs, onboarding a new user, or authoring Andocs documentation and HTML prototypes. Trigger phrases include “start the Andocs demo,” “introduce Andocs,” “představení Andocs,” “rozběhni andocs demo,” “predstavenie Andocs,” and “spusti Andocs demo.” Use it for markdown rendering, diagrams, math, html-preview, prototype.json, shared assets, managed prototype datasets and persistent data, local CLI state, screen review status and release versions, or editing a prototype through OpenDesign.
Comprehensive API design and development workflow that orchestrates API architecture, endpoint design, documentation generation, and integration patterns. Handles everything from RESTful API creation and GraphQL schema design to API gateway setup and version management.
Use this skill when you need to generate single-file HTML API documentation from TypeSpec (.tsp) files. Supports automatic generation of curl examples, parameter tables, version tags, and Markdown snippet injection. Trigger words: generate API docs, generate interface docs, api doc, typespec doc, generate HTML docs.
Advanced Product Quality Planning (APQP) — plan and track a new product launch through all 5 phases, identify deliverables per phase, run gate reviews, and ensure quality outputs are complete before Start of Production (SOP). Use when launching a new part, managing an APQP project, or auditing APQP completeness. Covers AIAG APQP 2nd edition and IATF 16949 §8.3.
Use when recording an architectural decision so future maintainers know why—phrases like "ADR", "architecture decision record", "why did we pick X", "lightweight architecture documentation", "decision log", "should this be an ADR", "tribal knowledge", "we chose this two years ago and now nobody remembers why". Defines an ADR as a short, structured document capturing what was decided, why, alternatives considered, and consequences—written at the time of the decision and kept in the repo
Use when protecting a service from a degraded or failing dependency—phrases like "circuit breaker", "bulkhead", "retry policy", "exponential backoff", "fail fast", "fallback", "cascade failure", "open/closed/half-open state", "resilience4j", "Polly", "Hystrix". Defines the circuit breaker pattern and related failure-isolation tools (bulkhead, retry-with-backoff, timeout, fallback)
Use when separating read and write models — phrases like "CQRS", "Command Query Responsibility Segregation", "read model vs write model", "denormalized read side", "materialized view", "different scaling for reads and writes", "command bus", "query bus", "should we use CQRS here". Defines CQRS as keeping reads and writes on separate models, with picking criteria that resist over-application.
Use when designing a system around asynchronous events—phrases like "event-driven architecture", "EDA", "event broker", "Kafka", "publish-subscribe at scale", "event-first design", "choreography vs orchestration", "event broker vs message queue", "broker topology vs mediator topology". Defines event-driven as a macro architectural style where components emit and react to events, not direct calls
Use when persisting state as a sequence of events—phrases like "event sourcing", "event store", "append-only log of facts", "replay events to rebuild state", "audit log as source of truth", "temporal queries", "snapshot", "should we event-source this aggregate". Defines event sourcing as storing the history of what happened as the canonical state, with current state derived by replay
Use when enforcing architectural characteristics through automation — phrases like "architecture fitness function", "evolutionary architecture", "architecture tests", "ArchUnit", "dependency rules in CI", "automated architecture checks", "guardrails", "drift detection". Defines fitness functions as automated checks that verify architectural characteristics on every change, catching drift before it accumulates.
Use when picking a software architecture style or pattern for a system — phrases like "what architecture should we use", "monolith vs microservices", "should we go event-driven", "is this a CQRS situation", "do we need a saga here", "should we use hexagonal architecture", "what's the right architectural style for this problem". Diagnostic mode that routes a stated problem to the relevant architecture skill or rules patterns out.
Use when discussing layered architecture (n-tier) — phrases like "layered architecture", "n-tier", "presentation / business / persistence layers", "should we go layered", "leaky layer", "sinkhole anti-pattern", "open vs closed layers". Defines the layered style with its strict top-down dependency rule, the open/closed layer concept, and the most common failure modes.
Use when designing or reviewing a microservices architecture — phrases like "microservices", "should we split this service", "service boundaries", "bounded context per service", "API gateway", "backend for frontend", "BFF", "service mesh", "microservices vs monolith", "database per service", "data ownership". Defines microservices as independently deployable services aligned with business capabilities — and the discipline of resisting the style until the monolith proves inadequate.
Use when migrating a legacy system incrementally — phrases like "strangler fig", "strangler pattern", "incremental migration", "monolith to microservices", "rewrite without big bang", "replace legacy in place", "facade in front of the legacy system", "anticorruption layer". Defines Fowler's Strangler Fig as the migration discipline of gradually replacing legacy functionality behind a stable facade, never doing a big-bang rewrite.
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.