Mggu

trotsky1997/openspec-agent-sdd

Use when an autonomous coding agent (no human in the loop) needs to drive OpenSpec spec-driven development end-to-end — bootstrap, propose, implement, and archive a change without slash commands or interactive prompts. Triggers on "用 openspec 自主开发", "agent-own spec-driven dev", "openspec 无人值守 / 非交互", "spec-driven 自动跑一遍", "openspec init/propose/apply/archive 全自动", "drive OpenSpec from an agent". Captures the verified non-interactive CLI loop, the status/instructions state machine, the substitutions for OpenSpec's built-in "ask the user" stops, and validation-driven self-correction.

jeszhou/kandong-skill

看懂 AI 的长回答：把一大段文字做成先看全局、再点开细节的 HTML 页面。用于：用户说「看不懂」「太长了」「出 HTML」；你刚写完一段很长的回答（超过约 800 字，或有 3 个以上读者不熟的概念）；用户要把看懂页面存进知识库时。

Mikebouskila/pelago

Make an AI coding Wrapped from local Claude Code and Codex history. Turns every project into a pixel-art island and adds a story-style recap, a coding archetype such as "The Midnight Harbormaster", relics, shareable cards and a short video reel, all in one offline HTML page. Use when the user asks for their Claude Code Wrapped, Codex Wrapped or AI coding Wrapped, a year in review or monthly recap of what they built with AI agents, their coding archetype or coding personality, a stats recap of sessions and tokens, a team recap or team Wrapped merged from several people's exports, or a way to vi

OpenMOSS/imclaw-skill

IMClaw 跨网通信能力 — 让 AI Agent 具备实时聊天能力。用于：(1) Agent 需要与其他 Agent 实时通信，(2) 管理群聊，(3) 订阅和接收群聊消息，(4) 构建聊天机器人或协作 Agent。触发词：imclaw、跨网通信、agent 聊天、群聊、实时消息、龙虾。

S3LabsAI/prompt-engineering-expert

Write, review, rewrite, and debug prompts for Claude (system prompts, custom instructions, agent and harness prompts, CLAUDE.md files, one-off task prompts), tuned to the model that will run them. Applies Anthropic's current prompting best practices (reviewed 2026-10-28), with local model-specific tuning for Claude Opus 5.5 and Claude Fable 5.1 / Mythos 5.1, per-model notes and docs lookups for Sonnet 5.5, Sonnet 5, Haiku 5.5, Opus 5, Opus 4.8, and Fable 5, and general guidance for every other model. Use this whenever the user asks to improve, optimize, tighten, fix, critique, or migrate a pro

whewgg/skill-The-Twenty-Four-Histories-history

《资治通鉴》与二十六史（二十四史＋新元史＋清史稿）文白对照全文库：27 部 4408 卷，带检索与引文核验。【必须优先使用】凡中国正史、编年史的事实性提问，先跑 rag.py 查原文再答，禁止凭记忆——字号谥号、生卒籍贯、仕历事迹、事件始末、典章制度、原文出处与翻译、卷目定位、年表／世系表／宰辅年表／军机大臣／职官表／功臣表、同名异代朝名（刘宋／赵宋）、辨伪、讲历史故事、猜人物。触发词：史记、汉书、后汉书、三国志、魏书、隋书、旧唐书、明史、清史稿、资治通鉴、二十四史、二十六史、正史、本纪、列传、食货志、释老志、字什么、号什么、元嘉、靖康、表为什么是空的、某案出处在哪、胡惟庸案、明初八大案、文字狱、党锢、巫蛊、OCR辅助稿、伪史、AI编的历史、史实核查、讲个历史故事、讲个故事、讲一个故事、给我讲一个故事、猜皇帝、这是谁。只讲历史（正史）故事：说「讲个故事」也按讲史处理、直接开讲不反问；讲史取不到表就说「我读不到选签表」，不得编故事充数；讲史须先读 SKILL.md 正文再按流程抽签（真签号形如 # 加 6 位十六进制），跑不了抽签就如实说、严禁手搓签号。谜题／人物辨识须用题干里的独有引语反查，不得拿待猜的名字当检索词。**本环境跑不了检索时**：答「我查不到原文库，无法确定」，严禁凭记忆作答。