Browse agent skills

面向企业级 AI Coding 的轻量 SDD 脚手架，以规格驱动交付，让开发经验持续沉淀与复用。

Browser tabs, element annotations and MCP/CLI-first desktop app control for DeepSeek Harness

Analyze writing locally, share style profiles in SVG metadata, and create deterministic style-guided drafts without an external AI service.

Generate royalty-free music of an exact length for videos, apps and games with the Mubert API

Isomorphic web extension polyfill, mock runtime, and headless runner bundler for browser automation

古人说唱 Battle：一句话做一期 AI 历史人物说唱对决视频的 Claude Code Skill（Suno + Midjourney + MiniMax H3 对口型）

Success codes can lie. An agent skill that makes coding agents verify each step's real effect and stop for review before the next one. Works with any agent.

Create XRechnung and EN 16931 e-invoices from your data, and read incoming XRechnung, ZUGFeRD and Factur-X invoices into fields.

Copy original RFC822 messages between IMAP mailboxes in n8n, then archive and expire the source copies

n8n Nodes für die doc2doc API: Dokumente zwischen Formaten konvertieren (PDF, DOCX, Markdown, ...)

n8n nodes for ConnectUC: call, SMS, CDR and voicemail triggers plus contact, DND, SMS and call actions

Coding agent for VS Code with any model. Use one Abacus AI plan, your ChatGPT subscription, your own API keys, or local models like Ollama.

The inference-money USDC bridge between Ethereum and Aztec: protocol SDK, contract artifacts, L1 ABIs, the testnet deployment, and a deploy library for instances of your own.

Connect coding agents (Claude Code, Codex) to a shared room

Runs a person's coding agents on their own computer for a NocoBase application's agent sessions, over the Agent Client Protocol, and the wire protocol between that bridge and the application

MiOpIIk deployment probe: reports runtime/deployment facts and audits every visible tool schema against the function-calling contract the upstream API enforces (object-root parameters, no null type). Born from an incident where one malformed third-party t

Optional S3-compatible file, native artifact and Metro cache provider for Stim.

Sprag speech and language models for Pipecat: Rhythm STT, Symphony LLM, and Chorus TTS

LlamaIndex tools and a reader for EconDash time series across ten categories: macroeconomics, financial markets, labor, demographics, real estate, energy, transport, healthcare, education and technology.

Token Meter TypeScript SDK — in-process usage capture for OpenAI, Anthropic, Gemini, Bedrock, Azure OpenAI, and the Vercel AI SDK.

Agent Skill for HarmonyOS NEXT / 鸿蒙 app development with Claude Code, Codex & AI coding agents: ArkTS errors, ArkUI state, hvigor/DevEco CI, Hypium tests, AppGallery release, APMS/AppFreeze. Grounded in Huawei's official docs.

Inline skill and prompt-template name completion for Pi, without send-time expansion

Embeddable chat SDK for Iris personas — text/voice chat over the Phoenix channel + ElevenLabs/Fish Audio TTS wire protocols.

Streaming text-to-speech that runs entirely in the browser (Kokoro 82M on WebGPU or WASM).

Start Whenever workflows from n8n and read their run results

AI KB knowledge retrieval tools for DeepSeek Harness (dsh): semantic search, document reading and RAG/agent answers over your own knowledge bases.

Prompt Compiler 插件：把粗糙 draft 编译成结构化 prompt（DSH 插件 host 半，B 组最小可运行版）。

DSH 插件：读对比文件（说明书部分，中文专利或外文专利中文译文），依据涉案专利、无效请求书、意见陈述书，在每个公开了涉案专利技术特征的段落左侧画红色大括号 + 红字标签（独立权利要求按拆分特征标「权1-特征1-1」，照抄意见陈述书的特征编号；从属权利要求标「权2-附加特征」；同段多特征合并为「权1-特征1-1,1-5,1-6,1-7；权2-4-附加特征」）；老公告文本无 [00xx] 段号时用「页+段序」（p10-2）或原文片段定位，与请求书「第10页第二段」同口径；「对应关系说明」（标题 + 四张表）

A Playwright harness for a web app developed specification first: backends stood in for by their contract, records of what the app can request and what its cases make it send, and the rules of its Rule #11 modes, for the plugin that enforces them.

Xingular Harness — a Pi agent harness built on recursive harnessing (Harness²): versioned eight-component harness trees, refinement from evidence, multi-account sessions, a compiled prompt, project notes, tools, and MCP