mit einem Klick
FuFan-VibeCodingCourse
FuFan-VibeCodingCourse enthält 67 gesammelte Skills von fufankeji, mit Repository-Berufsabdeckung und Skill-Detailseiten auf SkillsMP.
Skills in diesem Repository
单后端闭环测试执行器(stack-agnostic / project-agnostic)。当一个后端 feature 收尾、需要补 superpowers/spec-kit 开发期 TDD 与契约测试覆盖不了的后端结构性测试缺口时使用:先读项目栈 (package.json / pyproject.toml / go.mod / pom.xml 等)按栈实例化对应工具,再针对实际命中的缺口 RED→GREEN 补测并固化为回归。覆盖四类后端结构性缺口——真库数据层/迁移/事务/约束、鉴权与越权 (BOLA/BFLA)、并发/竞态/限频原子性、韧性/故障注入(重试/超时/降级)。触发词:单后端测试 / 后端缺口补测 / backend testing / 真库测试 / 越权测试 / 并发测试 / 韧性测试 / 后端集成测试 / 后端回归补测。遵循 testing-system-blueprint 蓝本(风险分级 / 可追溯 / 发布门 / 三层节奏),受自愈护栏约束(只写 tests/、断言 不可弱化、禁伪造修复、有界重试、产 PR 人审)。被 test-routing-advisor 在判定"单后端"时调用,也可直接触发。
单前端闭环测试执行器(stack-agnostic / project-agnostic)。当一个前端 feature 收尾、需要补 superpowers/spec-kit 开发期覆盖不了的前端结构性测试缺口时使用:先读项目栈(package.json 或 等价清单)按栈实例化对应成熟工具,再针对实际命中的层 RED→GREEN 补测并固化为回归。与 backend-testing 的根本区别——后端缺口无现成工具需自建("施工队"),前端成熟工具齐全,本 skill 不重新发明工具,只做 三件事:装 + 配 + 把项目的视觉/交互契约翻译成这些工具能跑的断言/规则("装配工 + 监理")。覆盖五层 前端结构性缺口——编译期 lint 门、L0/L1 单测(含 getComputedStyle 断言 token/深色/涨跌色)、a11y + 跨浏览器/响应式、前后端契约 mock、视觉回归。触发词:单前端测试 / 前端缺口补测 / frontend testing / 视觉回归 / a11y 测试 / 跨浏览器测试 / 响应式测试 / token 断言 / 前后端契约 mock / 前端回归补测。 常见地基为零:很多前端 [FE] 任务出参验证只写"手测"、测试运行器/RTL/Playwright 可能完全没装—— 本 skill 第一动作是"识栈→若无测试运行器先立地基",不假设 L0/L1 已就绪。遵循 testing-system-blueprint 蓝本(风险分级 / 可追溯 / 发布门 / 三层节奏),受自愈护栏约束(只写测试不改产品码、断言不可弱化、禁伪造 修复、有界重试、产 PR 人审)。被 test-routing-advisor 在判定"单前端"时调用,也可直接触发。
完整功能链路测试执行器(stack-agnostic / project-agnostic)。当多个 feature 已陆续收尾、 某条**跨多 feature 的端到端旅程(A→B→C)首次贯通**、需要为这条关键旅程织一张端到端安全网时使用: 与前三格根本不同——本格的被测对象**不是给定的,要先从系统结构里【挖掘】出来**。先从三源 (静态代码图 + 运行时 trace + spec 契约)挖出一份**跨多 feature 的通路清单(path inventory)**, 挑 P0 关键旅程,再端到端验证、固化成安全网。覆盖跨多 feature 且含非 UI 跳步(定时任务 / 异步 / 跨通道)的 完整链路;单 feature 内单切片归"局部前后端"第三格,不在这里。两个最独特点:① 被测对象要先被挖掘出来; ② 它是**安全网层**——少而精只盖 P0;**若 bug 首次在这层被发现,说明下层(单后端/单前端/局部前后端)漏测了, 应回补下层**。六步闭环:挖通路+选 P0 → 全系统编排+只 stub 外部边界 → 条件命中 → 两层落地 → RED→GREEN(禁 sleep, fake clock/手动触发定时任务/poll-retry 等最终态)→ 归档+拆栈+PR 人审。触发词:完整功能链路 / 端到端旅程 / 跨 feature 测试 / 通路挖掘 / path inventory / 关键旅程 / 安全网 / journey 测试 / full chain testing / e2e journey / 链路贯通测试。天然要最多代码才能跑、是最后才能执行的一格;代码未落地时只能"挖候选通路 + 写 RED E2E"不能跑。遵循 testing-system-blueprint 蓝本(风险分级 / 可追溯 / 发布门 / 三层节奏),受自愈护栏约束 (只写测试不改产品码、断言不可弱化、禁伪造修复、有界重试、产 PR 人审,发现真 bug 回交 superpowers TDD)。 被 test-routing-advisor 判定"完整功能链路"时调用,也可直接触发。兄弟:backend-testing(单后端)/ frontend-testing(单前端)/ fullstack-slice-testing(局部前后端)。
局部前后端接缝测试执行器(stack-agnostic / project-agnostic)。当一个 feature 的前后端两侧都收尾、 需要把"单前端阶段拿来 mock 的那份假设"与"单后端的真实行为"做一次对账时使用:在**单 feature 内**, 让消费者侧(前端/调用方)↔ 提供者侧(后端/被调方)的**单条切片**以真实形态对接、停止 mock,验证接缝。 它是 backend-testing / frontend-testing 的对账——单前端 mock 撒的谎(字段/类型/状态码/错误体/鉴权/时序) 在这里穿帮。**仅限单 feature 内单切片,不跨 feature**(跨多 feature 旅程属"完整功能链路"第四格)。覆盖四个 能力缺口——环境编排(让两侧+依赖以真实形态可复现一键起来)、契约真实性(消费者 mock↔提供者规格比对)、 接缝粘合(凭证透传/序列化往返/错误映射/头部跨域)、真实时序(条件命中:仅切片含流式/实时/异步才测)。 与前两格的根本区别——新难点是"起真栈(环境编排)"而非写断言:前两格能把另一侧 mock 掉,这格不能, 必须真把两侧拉活。先读两侧栈清单按栈实例化编排/契约工具,再 RED→GREEN 补接缝断言并固化为回归。 触发词:局部前后端测试 / 前后端接缝测试 / 切片对账 / 真前后端对接 / 起真栈集成 / 契约漂移验证 / 消费者契约对账 / fullstack slice testing / seam testing / contract drift / 前后端集成补测。 遵循 testing-system-blueprint 蓝本(风险分级 / 可追溯 / 发布门 / 三层节奏),受自愈护栏约束(只写测试不改 产品码、断言不可弱化、禁伪造修复、有界重试、产 PR 人审)。被 test-routing-advisor 判定"局部前后端"时调用, 也可直接触发。兄弟:backend-testing(单后端)/ frontend-testing(单前端)。
测试路由顾问:在某个 feature 的所有 task 完成(全绿)、进入收尾阶段时,先通篇扫描该 feature 的所有 task/文件,把它们归入一个「开放可增减」的场景类别集合(单后端 / 单前端 / 局部前后端 / 完整功能链路 / 可增类…),判定本 feature 整体属于哪一类,再把每个测试缺口 「路由到对应类别的执行器 skill」去闭环补测。它是 stack-agnostic 的检测器 / 路由器: 本 skill 只做「判类 + 标缺口 + 路由」,不写死任何单栈工具为答案——具体工具由被路由到的 类别 skill 按栈(读 package.json / pyproject / go.mod 等判栈后)实例化。当前路由去向: 单后端 → `backend-testing` skill;单前端 → `frontend-testing` skill;局部前后端 → `fullstack-slice-testing` skill;完整功能链路 → `full-chain-testing` skill(gstack `/qa` 折叠进它的 UI 可走段,非依赖)。判类、缺口标注、 三层节奏、可追溯、发布门等方法标准统一遵循 `testing-system-blueprint` skill。判类信号来自 任务范围标签、依赖图、跨模块契约声明、验收标准 AC,因此与具体项目无关。它的杀手锏:从依赖图 推导出「完成本 feature 后某条完整功能链路(A→B→C)首次贯通」并主动提示「这条链路现在可以 端到端测了」——这是人最容易忘的事。 当某个 feature 正在收尾、用户说出类似「这个 feature 该测什么 / 收尾测试 / 测试路由 / feature 完成后测什么 / 该用什么测试 / run test routing / what should I test now / which testing tool should I use / what tests does this feature need」时务必使用。 它只做建议与路由——绝不编写、运行或强制任何测试;测试是否真的写了、过了、断言有没有被弱化, 靠确定性的 CI 闸和护栏化自愈 agent 来保证。
测试体系蓝本(testing blueprint / test strategy)——一份 stack-agnostic、project-agnostic 的"测试作为体系"方法骨架。当需要确定测试策略、做风险分级(P0–P3)、建立需求↔测试可追溯、 规划三层测试节奏、设计闭环补测、定义发布 go/no-go 门(release gate)时遵循本蓝本。 触发词:测试体系蓝本、测试策略、测试策略蓝本、风险分级、可追溯、需求追溯、发布门、 闭环补测、testing blueprint、test strategy、test pyramid、release gate。 它被 test-routing-advisor(路由器)和各类别测试 skill(如 backend-testing)共同遵循。 本蓝本只给"方法与标准",不写死任何语言/框架/工具,不执行测试也不强制门禁—— 能力是通用原语,工具由调用方读项目栈后实例化;门禁的强制由 CI/hook/pre-commit 承担。
Generate or refresh a project root CLAUDE.md by READING the project's existing documentation (PRD / architecture decisions / constitution / DESIGN.md / package.json) and producing a 200-line index that follows 2026 Anthropic best practices (WHAT/WHY/HOW framework, @path progressive disclosure, sparse use of YOU MUST/IMPORTANT) AND embeds Karpathy's 4 coding behavior principles (Think Before Coding / Simplicity First / Surgical Changes / Goal-Driven Execution). Use this whenever the user says "帮我写 CLAUDE.md", "项目需要 CLAUDE.md", "生成项目 CLAUDE.md", "refresh CLAUDE.md", "bootstrap CLAUDE.md", "把 PRD 整理进 CLAUDE.md", or asks how to make Claude Code understand their project's tech stack, architecture, conventions, and workflows. ALSO use this skill even when the user does not explicitly mention CLAUDE.md — as long as they want Claude Code to consistently understand their project's structure and rules, this skill applies. Do NOT use for a brand new empty project that has no documentation yet (in that case the user need
对采用 spec-kit 式工作流的项目,实现/推进**单个**已就绪 feature(默认 specs/<id>-<name>/,含 spec+plan+tasks)的端到端标准流程。当用户说"实现 001 / 开发某个 feature / 跑 feature / 按这个流程做 X / run-feature / 继续下一个 feature / 把 <feature> 做了"等,或要对某个已就绪 feature 开工编码时,必须调用本 skill。它封装一套 opinionated 流程:worktree 隔离 → 读项目自身的 constitution/约定 → 按 tasks.md 走 TDD(RED→GREEN→REFACTOR) → 按任务标签([BE]/[FE]/[INT] 等)分流 → 多类缺陷 code review → 收尾 merge/tag;一次只跑一个 feature、跑完停下等审。区别于通用的 speckit-implement / superpowers:executing-plans——后两者是泛用执行器,本 skill 多了 worktree 隔离、读项目宪法合规、结构化缺陷审查清单、串行节奏控制。**项目特定的规则(技术栈、宪法条款、设计系统、里程碑命名)一律不写死,由项目自身的约定文件提供,skill 运行时去读、没有就跳过。** 不要用于:项目脚手架/基础设施初始化本身、撰写 spec/plan/tasks、跨多个 feature 并发。
把"多候选技术架构选型"打包成"法庭式 5 角色对抗调研"的标准化流程,用 Claude Code Agent Teams 让代言人/红队/集成评估师互相质疑,主 Claude 充当法官综合判决,最终产出架构基线决策文档。适用:多个开源项目 fork 选型、多个 SaaS/云厂商选型、多个技术栈对决(React vs Vue vs Svelte)、多个开源库选型(LangChain vs LlamaIndex)、多个架构方案选型(Monolith vs Microservices vs Serverless)、PRD 写到一半发现技术决策有争议需要 deep dive。即使用户没明确说"用 Skill",只要任务沾边"多候选架构选型需要更严谨的对抗评估"都要调用。触发关键词:架构选型 / 技术选型 / fork 选型 / 候选对抗 / 多方案对比 / 选型纠结 / 架构决策 / 技术栈对决 / 帮我决定用 X 还是 Y / SaaS 选型 / 库选型 / 对抗调研 / architecture decision / tech selection / framework comparison / vendor evaluation / adversarial review。不用于:单一候选无对比(不需要对抗)、纯产品决策不涉技术(用 brainstorming)、已有强烈倾向只想确认(用 devil's advocate 单 agent Skill 即可,本 Skill 过重)。
生成符合行业标准的产品需求文档(PRD / Product Requirements Document)。当用户提到"写 PRD / 产品需求文档 / 需求文档 / 功能规约 / 产品规约 / spec / 写需求 / 整理需求 / 把想法变成 PRD / 把头脑风暴结果落成需求文档 / 写 product spec / 把脑暴结论沉淀成文档"等场景时务必调用。输出 14 章结构化 PRD,支持 Brainstorming → PRD → TRD 的 Vibe Coding 工作流。中文优先,AI Agent 友好。即使用户没明确说"用 Skill",只要任务沾边产品需求文档撰写都要调用。
把"任何 0→1 新产品的立项前期调研"打包成"4 路并行 + 双引擎验证 + 1 路收敛"的标准化引导流程。通过苏格拉底式提问从用户抽取项目骨架,自动生成可直接喂给 Claude Code Task 工具的 4-agent 并行调研提示词。用于个人/小团队启动新项目时的产品立项调研、技术选型前的事实摸底、PRD 写作前的输入材料准备。触发关键词:新项目立项 / 产品调研 / 立项调研 / 项目可行性调研 / 帮我调研一个新想法 / 我想做个 X 怎么开始调研 / 把想法变成调研计划 / 调研同类竞品 / 摸底数据来源 / 摸底开源生态 / kickoff research / product research / 0 to 1 research。即使用户没明确说"用 Skill",只要任务沾边"新产品立项前的事实摸底/竞品扫描/资源盘点/技术方案选型调研"都要调用。不用于:已经决定技术栈后的具体实现调研(那是 spec/plan 阶段)、纯学术研究(不是产品立项)、已有产品的迭代调研(用 brainstorming)。
Inject a design system (Stitch DESIGN.md / Figma export / Claude Design output / hand-written DESIGN.md) into an existing GitHub Spec-Kit project WITHOUT re-running the whole spec/plan/tasks pipeline. Use this whenever the user has already completed Spec-Kit's specify/clarify/plan/tasks for one or more features AND now wants to add UI design guidance (DESIGN.md, Stitch zip, Figma exports, Claude Design output, HTML prototypes, design tokens) so that the frontend implementation matches a specific visual standard. Triggers include phrases like "我已经跑完 spec-kit 又拿到了设计", "DESIGN.md 怎么融入已有 spec", "Stitch 设计接入 spec-kit", "Figma 设计融入已有的 spec/plan", "Claude Design 怎么用到我已有的项目里", "想把视觉规范注入到 constitution", "不想推倒重跑 spec-kit 但要加入设计", "前端 UI 怎么和我已有的 spec 文档对接". Use this skill even if the user does not explicitly mention "Spec-Kit" — as long as they have spec.md / plan.md / tasks.md files and a separate design artifact, this skill applies. Do NOT use this skill for fresh projects that haven't run Spec-Kit yet (use Spec-Kit d
Use when writing content that cites sources (URLs, numbers, quotes) across social platforms — each platform has different rules for whether bare URLs should appear at all. Prevents inline URL overflow on platforms where links are unclickable or unwanted (小红书, Instagram-like). Applies whenever a Writer receives research with URLs and must decide per-platform how to surface citations.
Use when writing a WeChat Official Account article (公众号推文) — long-form structured Chinese content with H2 sections, scene-setting openings, cited sources, and data-driven analysis. Triggers on 公众号, WeChat article, subscription account post, or 推文 requests where depth and credibility matter.
Use when generating a Weibo post (微博) — ultra-short punchy Chinese microblog updates of ≤280 characters with 1–2 hashtags and a direct hook. Triggers on 微博, Weibo post, Chinese microblog, or short social update requests where brevity and immediacy are required.
Use when generating content for Xiaohongshu (Little Red Book / 小红书) — short lifestyle notes with emoji, hook-driven openings, pain-point or contrast titles, and interactive CTAs. Triggers on requests for 小红书 posts, rednote content, or any short-form Chinese social note styled for discovery feeds.
Use when wiring any messaging channel to a NanoClaw agent group, or diagnosing adapter failures. Covers channel adapter install, routing wire (messaging_groups / messaging_group_agents / agent_destinations), push credentials bootstrap, and delivery verification. Common pitfalls: SDK peer dep split, withRetry timeout misread, credentials missing before adapter factory runs.
Add Atomic Chat MCP server so the container agent can call local models served by the Atomic Chat desktop app via its OpenAI-compatible API.
Use Codex (CLI + AppServer) as the full agent provider — planning, tool orchestration, native compaction, MCP tools, session resume — in place of the Claude Agent SDK. ChatGPT subscription or OPENAI_API_KEY. Per-group via agent_provider. Distinct from using OpenAI as an MCP tool (where Claude remains the planner).
Add a monitoring dashboard to NanoClaw. Installs @nanoco/nanoclaw-dashboard and a pusher that sends periodic JSON snapshots.
Add DeltaChat channel integration via @deltachat/stdio-rpc-server. Native adapter — no Chat SDK bridge. Email-based messaging with end-to-end encryption.
Add Discord bot channel integration via Chat SDK.
Add Emacs as a channel. Opens an interactive chat buffer and org-mode integration so you can talk to NanoClaw from within Emacs (Doom, Spacemacs, or vanilla). Local HTTP bridge — no bot token or external service needed.
Add Google Calendar as an MCP tool (list calendars, list/search/create events, free/busy queries) using OneCLI-managed OAuth. Multi-calendar and multi-account supported. Mirrors /add-gmail-tool's stub pattern — no raw credentials ever reach the container; OneCLI injects real tokens at request time.
Add Google Chat channel integration via Chat SDK.
Add GitHub channel integration via Chat SDK. PR and issue comment threads as conversations.
Add Gmail as an MCP tool (read, search, send, label, draft) using OneCLI-managed OAuth. The agent gets Gmail tools in every enabled group; OneCLI injects real tokens at request time so no raw credentials are ever in the container or on disk in usable form.
Add iMessage channel integration via Chat SDK. Local (macOS) or remote (Photon API) mode.
Add a persistent wiki knowledge base to a NanoClaw group. Based on Karpathy's LLM Wiki pattern. Triggers on "add wiki", "wiki", "knowledge base", "llm wiki", "karpathy wiki".
Add Linear channel integration via Chat SDK. Issue comment threads as conversations.
Add a macOS menu bar status indicator for NanoClaw. Shows a bolt icon with a green/red dot indicating whether NanoClaw is running, with Start, Stop, and Restart controls. macOS only.
Add Matrix channel integration via Chat SDK. Works with any Matrix homeserver.
Route a NanoClaw agent group to a local Ollama model instead of the Anthropic API. Ollama speaks the Anthropic API natively (v1/messages), so no provider code changes are needed — just env var overrides and a model setting. Use when the user wants to run their agent locally, cut API costs, or experiment with open-weight models. See docs/ollama.md for background.
Add Ollama MCP server so the container agent can call local models and optionally manage the Ollama model library.
Use OpenCode as an agent provider (AGENT_PROVIDER=opencode). OpenRouter, OpenAI, Google, DeepSeek, etc. via OpenCode config — not the Anthropic Agent SDK. Per-session and per-group via agent_provider; host passes OPENCODE_* and XDG mount when spawning containers.
Add Resend (email) channel integration via Chat SDK.
Add Signal channel integration via signal-cli TCP daemon. Native adapter — no Chat SDK bridge.
Add Slack channel integration via Chat SDK.
Add Microsoft Teams channel integration via Chat SDK.
Add Telegram channel integration via Chat SDK.