Claude Skills by open-octo

Design guidance for any self-contained HTML/Markdown file shown in octo's Artifacts panel — reports, dashboards, architecture/system diagrams, generated UIs, slide-style pages. Read this BEFORE writing the file, not after — it calibrates how much design effort the request warrants and covers the panel's real constraints (sandboxed iframe, external resources gated to an allowlist of CDN hosts, narrow default width, no live theme push). Use when the user asks to "画架构图" / "generate a diagram" / ...

Configure IM platform channels (Feishu, Weixin/WeChat, WeCom, DingTalk, Discord, Telegram) for octo. Guides the user through platform consoles, collects credentials, writes ~/.octo/channels.yml, and diagnoses connection problems. Trigger on: "channel setup", "setup feishu", "setup weixin", "setup wechat", "setup wecom", "setup dingtalk", "setup discord", "setup telegram", "channel config", "channel status", "channel enable", "channel disable", "channel doctor", "connect feishu", "connect wech...

Review local code changes using an isolated sub-agent for unbiased analysis. Checks correctness, conventions, performance, tests, security, and tech design compliance. Use when the user wants a code review or says "review the diff" / "review my changes" / "代码评审" / "review 一下改动".

Configure octo's global settings through guided conversation — set up AI model endpoints (providers, API keys, models), adjust agent defaults (reasoning effort, permission mode, coauthor, workspace directory), and manage the default/lite model assignments. Trigger on: "configure model", "add provider", "setup endpoint", "add API key", "change default model", "switch model", "set reasoning", "change permission mode", "config setup", "配置模型", "添加模型", "配置 API", "切换模型", "修改设置".

Create, inspect, run, edit, enable/disable, and delete octo's scheduled cron tasks — recurring agent prompts stored in ~/.octo/tasks/*.json and executed by the octo serve scheduler. Use when the user wants to schedule a recurring task, e.g. "run X every morning", "schedule a daily report", "set up a cron job", "定时任务", "每天自动跑".

A harness for research you can trust: breadth first (many angles), then depth (primary sources), then an adversarial pass that tries to *break* each claim before it goes in the report. Built on octo's native tools — `web_search`, `web_fetch`, `sub_agent`, and (for login-gated / JS-rendered / anti-bot sites) the `browser` tool via the `web-access` skill. No external services. The goal is a **cited** report where every non-obvious claim traces to a source you actually read — not a plausible-sou...

Manage octo's agent profiles through conversation — create, modify, delete, list, and bind agents to IM chats by calling REST APIs. This skill is exclusive to the Default Agent (full-access). Use when the user wants to manage agents conversationally, e.g. "create a code review agent", "list all agents", "bind agent X to this group", "delete agent Y", "创建一个专家", "帮我建个专家", "管理专家", "编辑专家", "删除专家", "专家列表", "把专家 X 绑定到这个群".

Render structured, glanceable, interactive UI in the chat instead of plain text — dashboards, stat cards, filterable tables, charts, mermaid diagrams, choice panels, forms, quizzes. Panels can carry an id so later turns update them in place, and most interaction resolves in the browser with no round-trip. Read this before calling render_ui or emitting a ```octo-ui fence, so the spec you produce matches the node whitelist and caps the renderer actually enforces.

Interview the user relentlessly about every aspect of a plan until you reach a shared, resolved understanding. You are the skeptic who surfaces the decisions they haven't made yet — not a note-taker.

Acquire images as files — generate them with an AI image model (14 providers: OpenAI/gpt-image, Gemini, Qwen, Zhipu, Volcengine, Stability, FLUX, Ideogram, MiniMax, and more), search openly-licensed stock (Openverse/Pexels/Pixabay/ Wikimedia), or slice one generated sheet into elements. Drive it one-off with a single prompt, or in batch from an image_prompts.json manifest with a written- back status/audit trail. Use when the user wants to generate/create/make an image or illustration, source ...

Implement a technical design by decomposing it into dependency-ordered vertical slices, executing each with TDD red-green, reviewing each via an isolated sub-agent, and persisting progress to a state file so work survives session restarts. Use when the user has a tech design (doc or settled conversation) and says "implement", "start building", "let's code this", "实现吧", "开始实现", "按这个方案做", "把设计落地".

Configure and connect MCP (Model Context Protocol) servers through guided conversation — find the right server package or endpoint, build the config entry, write it to ~/.octo/mcp.json, and verify the connection. Use when the user wants to add, set up, or connect an MCP server, e.g. "add an MCP server", "connect X via MCP", "set up the filesystem MCP", "添加 MCP", "接入 MCP 服务".

Create, read, and edit Excel (.xlsx) spreadsheets programmatically with openpyxl — cell values, formulas, styling (fonts/fills/borders/alignment/number formats), merged cells, multiple sheets, charts, and data validation. Use whenever the user wants a spreadsheet created, inspected, or modified, or references an .xlsx file by name or path.

Onboard a new user OR curate a single piece of the assistant's inner state. Without arguments, runs the full first-run ceremony (AI name, personality, user profile, soul.md + user.md). With `scope:soul` or `scope:user`, runs a quick chat to update just that one profile file. With `path:<abs>`, runs a quick chat to update / keep / delete one memory file under ~/.octo/memories/. Use when the user wants to change the assistant's name/personality, update their own profile, or review/edit a saved ...

AI-driven multi-format SVG content generation system. Converts source documents (PDF/DOCX/URL/Markdown) into high-quality SVG pages and exports to PPTX through multi-role collaboration. Use when user asks to "create PPT", "make presentation", "生成PPT", "做PPT", "制作演示文稿", or mentions "ppt-master".

Answer questions about how to use octo, or diagnose something that isn't working, by reading the product documentation. Use when the user asks "how do I...", "what is...", "where is the docs for...", any usage/help question about octo itself, or reports a problem — "not working", "won't start", "error", "connection refused", "daemon won't start" — by reading TROUBLESHOOTING.md.

Create new skills, modify existing skills, and help users turn repeatable workflows into reusable instructions. Use when the user wants to create a skill from scratch, edit or improve an existing skill, capture a workflow as a skill, or optimize a skill's description for better triggering accuracy, e.g. "帮我建个技能", "创建一个技能", "把这个流程做成技能", "编辑这个技能", "优化技能的触发描述". Do NOT use to chain or orchestrate several EXISTING skills/recordings into a runnable saved workflow — that is workflow-creator.

Take a PRD (or feature description) and produce a complete backend technical design document through a structured process: **understand → explore → grill → write → self-review**.

复杂 web 任务的方法论：把内置的 `web_search` / `web_fetch` 与 `browser` 工具（驱动你日常的、已登录的 Chrome/Edge）按场景编排起来，并跨 session 积累站点经验。

Turn a repeatable multi-step task into a runnable, reusable saved workflow through guided conversation, in either of two shapes — chaining existing skills/recordings (wiring each one's output into the next's input), or composing the workflow script's own primitives (agent/parallel/pipeline) directly when the task needs fresh sub-agent orchestration and no existing skill covers it. Figure out which shape fits, generate the Ruby workflow, dry-run it, and save it with workflow_save. Use when the...

Do risky or experimental work in an isolated git worktree, then merge or discard. Use when the user wants changes tried in isolation ("试在 worktree 里改", "隔离环境跑一下", "worktree 隔离", "在临时分支上做"), or before a large/uncertain edit that shouldn't touch the main checkout. Trailing text is the task to perform in the worktree.

提炼并应用品牌声音（Brand Voice）。当用户要求"让文案符合品牌调性""定一下咱们的语气""按我们的风格写""这是我们的 on-brand 文案吗""写得更像我们""统一口吻""别这么官方/别这么网络梗"时使用。它先帮用户从已有素材里提炼出一份"品牌声音指南"，再把这份指南套用到每一条文案上，确保多次产出依然稳定、一个调性。不适合：纯找素材（交给用户直接提供），也不适合一次性写文案不带品牌约束（直接用 Copywriting skill）。触发词："品牌声音 / 品牌调性 / 语气 / 人设 / 口吻 / on-brand / 像我们 / 统一风格 / 太官方 / 太像 AI / 品牌人格"。

一趟旅行的预算不是一个数字，是好几个类目加起来的结果。这个技能负责把 预算拆成看得清楚的类目，每个类目给出档位区间，最后汇总成一张表。

审阅一份合同：识别对用户不利的条款、检查该类合同该有却缺失的关键条款、给出具体 的修改建议。产出是一份结构化审查报告，不是"这份合同能不能签"的最终结论。

系统性打磨文案（复制编辑）。当用户要"改得更好""润色""更有网感""更通顺""这个文案怎么改""帮我审一遍""这段读起来有点生硬""改掉机器味/AI 味"时使用。用七遍打磨法（清晰度、语气、所以呢、证明给我、具体、情绪、零风险）逐层改，保留作者原本的声音，不重写成另一个人。不含从零写新文案（用 Copywriting）和扣品牌调性（用 Brand Voice）。触发词："打磨 / 润色 / 改文案 / 审文案 / 有网感 / 通顺 / 生硬 / AI 味 / 机器味 / 更顺 / 帮我改"。

写有转化力的营销文案。当用户要写/重写落地页、产品介绍、定价页、功能页、活动页、广告语、Slogan、卖点提炼、转化文案时使用。覆盖标题公式、CTA、异议处理、卖点转利益点、一句话价值主张。也用于"把这段产品介绍写得更有卖点""帮我想个标题""这个页面怎么做转化"。触发词："文案 / 落地页 / 产品介绍 / 卖点 / 标题 / Slogan / CTA / 转化 / 营销文案 / 广告语 / 定价页 / 首页 / 宣传"。

每个出题人的考试都有"指纹"：常见的题型结构会重复出现，常踩的坑会重复出现， 各章节的分值比例往往也相对稳定。这个技能分析用户提供的历年真题，把这些规律 找出来。 **这是预测权重，不是预测答案。** 这个技能没法告诉用户考试会考什么，只能告诉 用户历年考过什么、按现在的大纲覆盖情况，哪些内容可能被重点考。

大纲/笔记是用来综合理解的，记忆卡片是用来强记的。这个技能从用户提供的材料 （或在用户明确要求时，从模型知识）生成卡片，用轻量间隔重复复习，并追踪哪些 概念还没记牢。 **不是完整的 SRS 系统。** 只是简单的 Leitner 分桶。够用、不重——如果用户已经 习惯用 Anki 之类的专门工具，建议继续用那个；这个技能是给在对话里想快速过一遍 的场景用的。

求职者侧模拟面试陪练。当用户要"模拟面试 / 面试练习 / 考我 / 这个岗位会问什么 / 准备自我介绍 / STAR 例子 / 这题怎么答 / 面试问题"时使用。核心：按岗位生成问题、**一次只问一题**、等用户答完再给针对性反馈与改进、把用户经历转成 STAR 故事库、陪练棘手题。不含简历打磨（用 Resume Crafting），不含雇主侧如何面试候选人。触发词："模拟面试 / 面试 / 练一下 / 考我 / 自我介绍 / STAR / 行为面试 / 这题怎么答 / 会问什么 / 面试准备"。

评估一项业务、产品功能或一次宣传活动是否符合中国大陆现行法律法规。产出是风险 点+等级+建议，不是"能不能做"的最终拍板——重大合规决策仍需法务/律师最终把关。

撰写一份能直接使用（或稍加核实即可使用）的中国大陆法律文书。产出是草稿，不是 最终定稿——涉及具体金额、管辖法院、法律依据的地方，标注清楚"待核实/待补充"， 不要替用户编造具体数字或杜撰不确定的事实。

面对一批文档（合同、公司档案、诉讼记录等）需要系统性审查时，有两种互补的做法： - **表格式（tabular）**：同样一组字段问遍每一份文档，产出一行一文档、一列一 字段的表格。适合"这 50 份合同里，变更控制条款分别怎么约定的"。 - **问题提取式（issue extraction）**：按标准类别清单扫描，只把真正有问题的 文档挑出来写成发现清单。适合"这批材料里有没有埋雷"。 两者可以先后使用：先跑表格式摸清全貌，再对表格里标红/标黄的行做问题提取式深挖。 **这不是替代人工阅读文档。** 每一格/每一条结论都是"线索"，需要人工核实，不是 "定论"。这个技能的目标是让核实变快，不是让核实变得可以跳过。

法律推理的价值不在于说出一个听起来对的结论，而在于让"事实 → 规则 → 适用 → 结论" 这条链路每一环都经得起检验。这个技能提供 IRAC 脚手架，并且强制区分"已核实的规 则"和"未核实的知识"，避免看起来自信、实际没有依据的分析。

系统性地找到、核实、引用现行有效的中国大陆法律依据，而不是凭印象说"法律规定 大概是这样"。核心纪律：**只引用你真正检索到、能看到原文的规范**，检索不到就 如实说检索不到，不用训练知识里的印象填空。

段落润色是在稿子结构已经站得住脚之后做的事——改善文字的移动方式，同时 保留作者的语气。这不是"帮你写得更好看"，是帮你去掉阻碍阅读的东西， 留下你自己的声音。

大纲是用来学习的东西。**搭大纲本身就是学习过程的一半**——这不是一句场面话。 一份不是自己搭出来的大纲，考试的时候大概率想不起来。这个技能帮用户搭大纲， 不替用户搭大纲。

一份好的打包清单不是"衣服、洗漱用品、充电器"这种放之四海而皆准的列表， 是按这次出行的具体情况——去哪、干什么、和谁去——推导出来的。这个技能先 问清楚决定清单内容的变量，再按变量分层生成。

扮演用户指定的读者画像，像这个读者第一次读到这篇稿子那样去读，报告真实的 阅读体验。这不是技术性的批评，是"读起来是什么感觉"——哪里被吸引住了、 哪里开始走神、哪里犯糊涂、什么问题一直悬在脑子里没解决。

打磨简历与求职信。当用户要"改简历 / 润色 / 量化成果 / 简历没亮点 / 弱项怎么写 / 针对这个岗位定制 / 写求职信 / 投 ATS / 转行简历"时使用。核心是把"职责罗列"改写成"成果导向、带数字、可验证"的 bullet，按 JD 定制关键词与顺序，处理求职信、ATS 可与不可。不含模拟面试（用 Interview Prep），不含职业规划叙事（直接对话即可）。触发词："简历 / 简历润色 / 量化 / 成果 / 亮点 / 定制简历 / 求职信 / cover letter / ATS / 转行简历 / 简历没东西写"。

行程不是"这几天分别去哪"，是"怎么走最顺、最不浪费时间"。这个技能负责研究 目的地的景点，按地理位置和开放时间设计逐日路线，每条推荐都附上理由。

写适配具体社媒平台的文案。当用户说"写篇小红书""发条公众号""来个短视频口播脚本""微博蹭个热点""朋友圈怎么发"时使用。核心是懂每个平台的"方言"——小红书种草风+标签+关键词埋点、公众号标题与开头、短视频前 3 秒钩子、微博短促话题、朋友圈真实克制，以及广告法违禁词和限流红线。不含通用转化文案（用 Copywriting），不含精修已有文案（用 Copy Editing）。触发词："小红书 / 公众号 / 抖音 / 视频号 / 微博 / 朋友圈 / 短视频 / 口播 / 脚本 / 种草 / 爆款 / 笔记 / 发帖 / 文案平台 / 平台格式"。

这是诊断，不是重写。目标是告诉用户"这篇稿子需要哪种修改，先改哪里"，不是 把稿子按自己的风格重写一遍。

坐下来学习却不知道该学什么，是几周时间悄悄溜走的常见原因。这个技能负责 搭一份计划——距考试还有几周、每天几节课、每周哪些学科、用什么方法——然后 根据用户实际完成的练习结果动态调整。是一份活的计划，不是一次性的日历 导出。 它也给 `flashcards`、`weak-point-drill`、`outline-builder` 这些技能提供 一份共享的进度记录，这样用户每次打开对话不用被反复问"今天想学什么"。

签证政策因目的国、申请人国籍、签证类型（旅游/商务/过境/工作）而完全不同， 而且经常变——不是一套可以背下来直接套用的规则。这个技能的核心不是"知道 每个国家的签证要求"，是"问对变量、系统性检索、不确定就说不确定"。

泛泛出题效率不高——如果用户在某几个子主题上反复出错，把出题权重往那几个 子主题倾斜，比平均分配到所有主题更有效。这个技能负责：定位薄弱子主题、 按权重出题、讲透每道题对错的原因、追踪跨小节的错误模式是不是真的"卡住了"。