LLM AI Skills, Multi-AI-Agent
See also upper main item: /llm
DEFINITION 定义 #
DEF SKILL = 进阶版 Prompt,解决的就是Prompt太长的问题。
英文文档:What are skills 中文视频:手把手彻底学会 Agent Skills!小白教程 2026-02-02 秋芝2046
my-skill/
├── SKILL.md # Required: instructions + metadata
├── scripts/ # Optional: executable code
├── references/ # Optional: documentation
└── assets/ # Optional: templates, resources
Tab Coding -> Vibe Coding (氛围编程) -> Spec Coding (Kiro广告,国内唯一无门槛调用Claude) (含:产品需求、技术设计、任务清单 Tasks) -> TDD AI Coding (Sunny补充最后一个)
ref Bilibili video by 秋芝2046 轻松学会!高手都在用的AI编程大法! 2025-12-12
WARN! #
- AI 是不靠谱的,例如各 AI IDE 带有 “revert” 功能,实际经常无法还原AI的变更,一定要对每个功能自己 git commit。
- 不要使用 /share 功能,会导致远程AI服务的内容(含自己的代码)暴露在公网!
- 多Agent系统(multi-agent)、复杂架构不一定是好事:那些 Agent 神文没告诉你的事:照着做,系统只会更烂。AI agent 搭建实操指南。2025-12-21
COLLECTION OF TUTORIALS/TIPS #
claudecode/opencode frequently used commands #
/init # 对当前项目已有的代码生成一个总结 AGENTS.MD,帮助AI快速理解一个项目
/compact # 压缩之前的对话为摘要,释放上下文窗口
create our own slash-commands/agents #
commands for either build/plan mode; agents as either primary/subagent.
ref: video Bilibili by 技术爬爬虾 OpenCode详细攻略,开源版Claude Code,免费模型与神级插件
use opencode in cloud #
e.g. ref: video Bilibili by 技术爬爬虾 在Github利用Actions自动调用OpenCode修复 issues
oh my opencode (omo) #
DEF OMO = 预置常用工具 + 预置常用MCP + 预置Agent
install: #
open OpenCode, then paste:
Install and configure oh-my-opencode by following the instructions here:
https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
and enter.
ref: github
experience: #
omo使用技巧/经验分享
评论区 of 一人抵一个开发团队!OpenCode最强插件Oh My OpenCode, AI超元域 2026-01 (文本教程, OMO中级(进阶)水平)
by 硬币大仙人: 这俩我都用,简单点说claude code要自己折腾配置才好用, omo插件加持下的opencode基本上是开箱即用,就是废订阅。我也尝试过改Sisyphus的model minimaxm2.1 自我感觉不行,主力还是得靠opus4.5。 上来先用计划agent制定"agents.md",调教好以后,小任务直接两三句就交给他做,复杂点的任务再套一层ralph-loop 配合他自己内置的continue hook,两层续写,基本可以做到交代完之后 就 hands-off。 omo有一个很好的hook是移除不必要的注释。agent写代码一般都很多废话,他这个机制会让agent每次编辑文件都有强制精简注释的动作,很实用。
by carter003: 我觉得OmO还是在测试版范畴中.第一,架构非常好,通过subagent进行任务编排.这种架构要好于Zen的架构.第二,编排还不是智能的,prompt写的比较固定,所以还不够智能.第三,由于explore因为网络或者项目大小.导致background返回可能要1分钟甚至更长时间.这时候主agent就会等不及,然后自己去探索了.相当于explore agent白干了.第四,偶发边界失效,比如在spec kit中 /plan 下一步是 /task 但是因为OmO遵循Sisyphus模式,所以自动提示大模型继续检查现有工作,这时候大模型误认为是继续执行下一步. 不过OmO最近好火爆.
by 沙漠白莲:我喜欢规划一个特别特别长的规划文件,然后要codex运行十几个小时一次性跑完,第2天早上一醒来,任务就做完了[打call]
by 海十Mirage: 我测试了,一个组提示词跑了一早上,然后效果并不太满意,用cc给同样的提示词,中途再加一些提示词,更快效果更好
features quick overview: #
- Agents: Sisyphus (the main agent), Prometheus (planner), Oracle (architecture/debugging), Librarian (docs/code search), Explore (fast codebase grep), Multimodal Looker
- Background Agents: Run multiple agents in parallel like a real dev team
- LSP & AST Tools: Refactoring, rename, diagnostics, AST-aware code search
- Context Injection: Auto-inject AGENTS.md, README.md, conditional rules
- Claude Code Compatibility: Full hook system, commands, skills, agents, MCPs
- Built-in MCPs: websearch (Exa), context7 (docs), grep_app (GitHub search)
- Session Tools: List, read, search, and analyze session history
- Productivity Features: Ralph Loop, Todo Enforcer, Comment Checker, Think Mode, and more



(pic ref: AI超元域)
Main agent: Sisyphus (Opus 4.5 High), which uses the tools below:
- Sisyphus’s Teammates (Curated Agents)
- Hephaestus: Autonomous deep worker, goal-oriented execution (GPT 5.2 Codex Medium) — The Legitimate Craftsman
- Oracle 先知: Design/architecture, debugging (GPT 5.2 Medium)
- Frontend UI/UX Engineer: Frontend development (Gemini 3 Pro)
- Librarian 图书馆管理员: Official docs, open source implementations, codebase exploration (Claude Sonnet 4.5)
- Explore 网络搜索: Blazing fast codebase exploration (Contextual Grep) (Claude Haiku 4.5)
- Full LSP / AstGrep Support: Refactor decisively.
- Todo Continuation Enforcer: Forces the agent to continue if it quits halfway. This is what keeps Sisyphus rolling that boulder.
- Comment Checker: Prevents AI from adding excessive comments. Code generated by Sisyphus should be indistinguishable from human-written code.
- Claude Code Compatibility: Command, Agent, Skill, MCP, Hook(PreToolUse, PostToolUse, UserPromptSubmit, Stop)
- Curated MCPs:
- Exa (Web Search)
- Context7 (Official Documentation)
- Grep.app (GitHub Code Search)
- Interactive Terminal Supported - Tmux Integration
- Async Agents
- …
ref: video Bilibili by 技术爬爬虾 OpenCode详细攻略,开源版Claude Code,免费模型与神级插件
opencode local+remote 本地+远程 mcp 配置方法 #
ref: video Bilibili by 技术爬爬虾 OpenCode详细攻略,开源版Claude Code,免费模型与神级插件
claudecode/opencode 典型项目实例(配置、项目结构) #
# ClaudeCode
project/
├── Product-Spec.md # 产品需求文档(自动生成)
├── Product-Spec-CHANGELOG.md # 变更记录(自动生成)
├── UI-Prompts.md # 原型图提示词(自动生成)
└── .claude/
├── CLAUDE.md # !!! 主控配置(必须创建) !!! 位置和名字与 OpenCode 不同!
└── skills/ # 技能包(必须创建)
├── product-spec-builder/ # 产品经理技能包
│ ├── SKILL.md
│ └── templates/
│ ├── product-spec-template.md
│ └── changelog-template.md
├── ui-prompt-generator/ # UI 提示词设计师技能包
│ ├── SKILL.md
│ └── templates/
│ └── ui-prompt-template.md
└── dev-builder/ # 全栈开发工程师技能包
└── SKILL.md
# OpenCode
project/
├── AGENTS.md # !!! 主控配置(必须创建) !!! 位置和名字与 Claude 不同!
├── Product-Spec.md # 产品需求文档(自动生成)
├── Product-Spec-CHANGELOG.md # 变更记录(自动生成)
├── UI-Prompts.md # 原型图提示词(自动生成)
└── .opencode/
└── skills/ # 技能包(必须创建)
├── product-spec-builder/ # 产品经理技能包
│ ├── SKILL.md
│ └── templates/
│ ├── product-spec-template.md
│ └── changelog-template.md
├── ui-prompt-generator/ # UI 提示词设计师技能包
│ ├── SKILL.md
│ └── templates/
│ └── ui-prompt-template.md
└── dev-builder/ # 全栈开发工程师技能包
└── SKILL.md
具体内容需要付费RMB:800/年,见: video: OpenCode 实测|从零开发一个 AI 应用,能替代 Claude Code 吗? 2026-01-14 AI 编程 | 产品经理开发技能 3.0 by 废才 (bak)
code-simplifier #
ref by AI全文总结 插件:专治AI生成的屎山代码与过度封装 2026-01-10:简化代码逻辑、注释等等。
Install in ClaudeCode : plugin install code-simplifier 评论区建议手动调用即可。
also: https://github.com/Zhen-Bo/pragmatic-clean-code-reviewer
also: 人类的3本书: Clean Code, Clean Architecture, The Pragmatic Programmer.
SUGGESTED SKILLS #
# ? npx skills add https://github.com/vercel-labs/skills --skill find-skills