worldwonderer/oh-story-claudecode
14 skills121k installs65.6k starsGitHub
Install
npx skills add https://github.com/worldwonderer/oh-story-claudecodeSkills in this repo
1Story Long Writename story-long-write version 1 0 0 description 长篇网文写作 从大纲到正文 辅助长篇网络小说的创作 包括世界观 人物 情节线管理 触发方式 story-long-write 写长篇 帮我开书 写大纲 日更 续写 继续写 修改第X章 回炉 重写第X章 metadata openclaw source https github com worldwonderer oh-story-claudecode story-long-write 长篇网文写作 你是网络小说创作教练 你的任务是帮用户从零开始写一本长篇网络小说 从选题确认到大纲搭建再到正文输出 核心方法 我们写网文先抓情绪 再用验证过的方法可靠地交付这个情绪 灵感只做素材来源 1 先定情绪 再定故事 每个场景都必须服务于一个明确的情绪目标 说不清交付什么情绪的场景不该存在 2 从验证过的模式出发 先问 什么被验证过有效 我如何重新交付 少从 我想写什么 直接起步 扫榜找方向 拆文找模块 对标找节奏 3 用模块组装 不要重新发明 每个题材都有验证过的剧情模式 反转怎么铺 爽点怎么爆 感情怎么拉扯 找到对的模块 把对标书的具体角色看成功能位 对手 盟友 催化剂 再映射到你的角色 用你自己的素材填充这些功能位 4 只加载必需信息 写每章时只加载 不知道就会写错 的信息 涉及角色的状态 待回收的伏笔 相关设定 其余留在文件系统里 题材 核心情绪 重点参考 打脸 逆袭 爽感释放 genre-writing-formulas md 身份反转 震撼 痛快 reversal-toolkit md 感情拉扯 意难平 emotional-methods md 悬疑 惊悚 紧张 好奇 hooks-suspense md 日常装逼 期待感 hooks-chapter md 情绪反查题材 如果用户先说了情绪感觉但没提题材 从上表反向匹配 例如 爽感释放 指向打脸 逆袭 再从 genre-catalog md 找该题材下的细分方向 写作流程 根据用户意图和项目状态选择场景 场景 触发条件 执行流程 开书 帮我开书 项目目录为空 完整 Phase 1 2 3 4 5 下方全部流程 日更续写 关键词 日更 续写 继续写 且 项目已有正文 追踪 加载 references workflow-daily md 大修 修改第X章 回炉 重写第X章 加载 references workflow-revision md 开新卷 如果新卷引入新角色 势力 设定 先回 Phase 2 增量补充 再进 Phase 3 补充新卷细纲 最后 Phase 4 写作9.8kinstalls2Story Deslopname story-deslop version 1 0 0 description 网文去AI味 检测并清除文本中的AI写作痕迹 让文字回归自然 非模板化 触发方式 story-deslop 去AI味 去AI味 这篇太AI了 网文去AI味 metadata openclaw source https github com worldwonderer oh-story-claudecode story-deslop 网文去AI味 你是网文润色专家 你的任务是把 AI 味浓重的网文文本改写自然 降低模板化 书面腔和过度工整感 核心信念 AI 味的主要问题不是语法 而是过度圆滑 工整 解释充分 改写目标是保留剧情功能 同时增加口语 停顿 跳跃和具体动作 核心哲学 原则 1 不是改错 是改味 AI味不是语法错误 不需要 修正 AI味是一种风格问题 过于书面化 过于对仗工整 过于面面俱到 去AI味的本质是把文字从过度工整拉回具体 自然 可读 原则 2 改最少 效果最大 去AI味不是重写 目标是改最少的字 让整段文字的 味 变过来 能改一个词就不改一句 能删一句就不重写一段 没有问题的句子尽量保留原句 人名 地名 数字 章节名 专有名词优先保留 过度去AI味保护 不得整段删除正文内容 如果某段被标记为多处AI味 应逐句修改而非删除整段 删除前必须确认 被删除的内容是否包含伏笔 钩子 角色特征 情节推进等关键信息 如果删除会破坏情节连贯性 改为 降AI重写 而非删除 删除比例上限按 AI 味等级分级 轻度 15 中度 25 重度 35 重度文本可通过 合并重复描写 重写降AI 产生更大字符差 但仍不得整段删除或删掉剧情功能 超过对应比例应在报告中标记超限风险 并输出分段处理方案 如果逐句修改后某段仍不满意 在去AI味报告中标注 需复核 而非删除 不计入当前等级的删除比例上限 对于 疑似AI味但不确定 的内容 在去AI味报告中标注 需复核 而非插入正文 原则 3 保留创作意图 去AI味只改 怎么说 不改 说什么 剧情 人设 情节走向一概不动 不新增原文没有的情节 设定 关系或时间线 如果原文有逻辑问题 那不是去AI味的活 原则 4 保留有功能的语气 不保留长停顿符号 去AI味不是把文字全部磨成句号 质问里的 爆发峰值的少量 可以保留 犹豫 未尽 打断或拖长用动作 短句 换行 逗号或句号重排 正文产物不保留 也要清理无功能的 和随机标点堆砌 自然文本基准 去AI味需要知道自然网文文本的特征 以下是从热门网文中提炼的非模板化写作特征 作为对比基准 自然文本特征 与AI味对比 维度 自然文本 AI味文本 段落长度 1-3句为主 偶尔1句独占1行 每段4-6句 整齐均匀 对话标签 60 无标签 用动作替代 说 几乎每句都有 说道 问道 情绪表达9.7kinstalls3Story Long AnalyzeThe story-long-analyze skill is a long-form web fiction analysis pipeline for agents acting as novel structure analysts. It confirms the target book and source text, backs up originals to a local library folder, and runs Stages 0 through 6 with checkpoint recovery via _progress.md. Stage 1 produces golden-three-chapter deep breakdowns and a quick preview report before user confirmation to continue. Stage 2 spawns chapter-extractor agents for per-chapter summaries with plot points and character filtering. Stages 3 through 6 aggregate plot lines, rhythm and emotion modules, world-building settings, faction files, a full analysis report, and style synthesis with anchor examples. Outputs land under a structured directory with chapters, characters, plot, settings, and progress tracking. The skill treats analysis as transformative literary criticism on legally held fiction, continuing through sensitive fictional elements without refusing whole chapters. Use when users invoke long-form novel breakdown, golden three chapter analysis, or provide fiction text file paths.9.7kinstalls4Story Short WriteThe story-short-write skill is a short web fiction writing executor that completes stories from concept to finished draft with emotion as the primary target. Execution rules require choosing an emotion goal before plotting, using one reversal to anchor the piece, deleting sentences that do not advance plot or emotion, and defaulting to first-person narration for platform-style short fiction. Phase 1 confirms the reader feeling and genre direction with an emotion selection table covering regret, twist shock, catharsis, warmth, unease, and resonance. Phase 2 builds a core framework, optionally loading reference analysis from analyze skill outputs in library or benchmark folders. Format rules mandate single-newline paragraph spacing in body.md, unified dialogue quote style, and section markers like ###1. The skill can spawn story-architect agents for framework design and narrative-writer sub-agents for section drafting. Use when users invoke short fiction writing, salt-style stories, or platform short-form web novel creation.9.6kinstalls5Story Long ScanThe story-long-scan skill analyzes long-form web fiction market rankings on platforms like Qidian, Fanqie, and Jinjiang to surface repeatable genre trends rather than single-book spikes. Core philosophy treats one ranking as a clue while cross-sample repeated patterns become signals. Traffic platforms like Fanqie emphasize completion rates while subscription platforms like Qidian focus on paid read metrics and Jinjiang on favorites and points. Phase 1 confirms target platform and genre direction for focused or overview scans. Data sourcing prioritizes script scraping with platform-specific scrapers, falling back to user-provided screenshots or built-in knowledge trends. Qidian collection uses qidian-rank-scraper.js with mobile SSR pageContext by default. Output follows scan-output-format references with structured Markdown files per rank list. Each candidate direction receives feasibility checks on material reserves, genre boundaries, length capacity, and platform sample sufficiency. Use when users invoke long-form rank scanning, ask what genres are hot, or request Qidian ranking analysis.9.6kinstalls6Story Short AnalyzeThe story-short-analyze skill deconstructs popular short web fiction across platforms like Fanqie, story magazines, Zhihu salt-style stories, and revenge or rebirth genres. It runs a single full pipeline through Stages 2-6 producing complete analysis reports saved under a library folder per title. Downstream story-short-write reads the report, plot nodes, writing techniques, source text, and _meta.json when drafting the next piece. Phase 1 confirms the title and source, routes by word count under 15000 as short pipeline, handles a 15000-20000 gray zone with user choice, and redirects above 20000 to story-long-analyze unless the user insists on short mode. Analysis treats legally held fiction as transformative literary criticism continuing through sensitive fictional elements without refusing whole works. Outputs include breakdown reports with story core, structure, emotional lines, twist design, resonance layers, and writing craft notes. Use when users invoke short fiction breakdown for salt-style or platform short stories.9.6kinstalls7Browser CdpThis skill controls Google Chrome through the Chrome DevTools Protocol using the agent-browser CLI and a setup script on port 9222. It prioritizes a detect-only probe that reports CDP_STATUS ready or needs-setup before mutating the user environment. When Chrome is already running, the agent must ask for consent because setup can kill regular Chrome and lose unsaved tabs. Ready sessions connect with agent-browser --cdp 9222 for open, wait, eval, snapshot, click, and type operations. Complex JavaScript uses base64 eval or stdin heredoc to avoid shell escaping errors. Auth tokens can be read from localStorage or cookies. The workflow supports resetting the debug profile when login state expires and documents exit codes for consent, errors, and success. Use it when browser automation must reuse an existing logged-in Chrome session rather than starting cold.9.5kinstalls8Story Short ScanThe story-short-scan skill analyzes short web fiction market rankings on platforms like Zhihu salt-style stories -Fanqie short, Qimao, and Heiyan to surface actionable emotion directions and genre candidates. Core philosophy treats short fiction as an emotion market where readers complete a feeling in one sitting, emphasizing completion rate and shareability over long-form chase metrics. Short trend signals expire quickly, so reports must label sample dates, confidence, saturation risk, and next rescan time. Phase 1 confirms platform and genre direction for focused or overview scans. Data sourcing prioritizes browser-cdp page scraping, then user-provided screenshots, then built-in knowledge trends. Outputs include emotion frequency, trigger scenes, burst points, and viral hooks rather than genre names alone. Cross-platform comparison mode contrasts platform-specific success patterns. Use when users invoke short rank scanning, ask what short genres are hot, or request Zhihu story ranking analysis.9.5kinstalls9Story CoverThe story-cover skill generates professional Chinese web novel covers from a book title and author name. Agents infer genre from title keywords using cover-styles references, then build English prompts with text, style, and scene layers including platform-specific typography rules. Required environment variables include GPT_IMAGE_API_KEY and BOOK_DIR; optional settings cover base URL, model, size, and reference images for image-to-image edits. The workflow collects title, pen name, target platform, and output directory, produces two or three composition variants, calls gpt-image-2 via curl with jq-safe JSON bodies, decodes base64 PNG output with versioned filenames, and saves prompt copies for iteration. Quality checks cover text rendering, genre match, composition, and platform fit. Image-to-image mode downloads URL references and posts multipart form data to the edits endpoint. Triggers include /story-cover, cover design requests, and Chinese phrases for novel cover generation. Use when creating web fiction covers with embedded title and author text.9.2kinstalls10StoryA router skill that dispatches user requests to specialized web novel (网文) production tools. When developers or writers invoke /story, $story, or natural language triggers like "帮我写书", it analyzes intent and routes to skills for long-form writing, short-form writing, novel analysis (拆文), trending topic scanning (扫榜), AI style removal (去AI味), cover generation, or project setup. Supports multi-book project management, version update checks via GitHub releases API, and fallback lookup when dedicated agents (story-explorer, story-researcher) are unavailable. Works across Claude Code, Codex CLI, and OpenClaw runtimes with equivalent slash/dollar/natural language invocations.8.9kinstalls11Story Setupstory-setup deploys a complete agent-based writing infrastructure into a user's Claude Code project. It installs 7 custom agents (story-architect, narrative-writer, consistency-checker, etc.), git hooks for story validation, path-scoped rules, and reference materials. The skill merges with existing CLAUDE.md and settings.local.json without overwriting user config. Developers writing Chinese web novels use this to bootstrap a multi-agent workflow where each agent handles specific narrative tasks. After deployment, a new Claude Code session is required for agents to register as subagent_type, enabling full multi-agent collaboration in story-review and story-long-write workflows.8.9kinstalls12Story Reviewname story-review version 1 1 0 description 多视角对抗式审查 full lean 模式在已部署 reviewer agents 时并行 spawn 缺失 异常 agents 或 spawn 失败时自动降级 参考文件不可读时使用内置 rubric fallback 触发方式 story-review 审查 审查一下 帮我审一下 metadata openclaw source https github com worldwonderer oh-story-claudecode story-review 多视角对抗式审查 你是审查协调器 你的职责是找出小说文本中的结构 角色 文字 设定问题 并给出可执行修改建议 执行铁律 审查是找问题 不是验证正确性 Review Mode 选择 story-review 或 story-review full 优先 spawn 全部 4 个 Agent 如果当前已经在子代理内 核心 Agent 未部署 异常 或 spawn 失败 自动降级为 story-review lean 优先 spawn story-architect consistency-checker 如果当前已经在子代理内 任一所需 Agent 未部署 异常 或 spawn 失败 自动降级为 story-review 不 spawn Agent 由当前会话执行基础审查 未指定 默认 full 并在报告里写明最终实际执行模式 Phase 0 预检与降级 必须先执行 1 确定请求模式 解析用户输入中的 full lean 未指定时目标模式为 full 2 确认是否允许 spawn 如果当前已经在子代理 Agent 内执行 不再递归 spawn 直接降级为 3 检查核心 Agent 部署状态 检查项目内 agents 同时兼容 Claude Code 和 OpenCode 优先检查 claude agents 其次检查 opencode agents 两个目录任一存在即视为已部署 full 必需 story-architect md character-designer md narrative-writer md consistency-checker md lean 必需 story-architect md consistency-checker md 对每个必需 Agent 文件 Claude Code agent claude agents 读取 frontmatter 确认 name 与 subagent_type 完全一致 frontmatter 缺失 不可解析或 name 不匹配时视为 malformed agent8.8kinstalls13Story Import| --- name: story-import version: 1.0.0 description: | 逆向导入已有小说。将已写好的小说（半成品或完本）反向解析为标准项目目录结构， 兼容 story-long-write / story-short-write 后续写作流程。内部复用 story-long-analyze / story-short-analyze 的拆解管道，按篇幅自动分流。 触发方式：/story-import、「导入小说」「反向解析」「导入」「把我的书导进来」 metadata: openclaw: source: https://github.com/worldwonderer/oh-story-claudecode --- # story-import：逆向导入已有小说 你是小说项目逆向工程师。将用户已有的小说文本（半成品或完本）解析为标准项目目录结构，使其可以无缝接入 story-long-write / story-short-write 的后续写作流程。导入流程按篇幅分流：长篇走长篇路径，短篇走短篇路径。 **核心信念：好的工具不是从零开始，而是从你已有的东西开始。** **交付物是写作工程**：把作者已有的书重建为可续写的**写作工程**（项目结构 + 拆文库分析资产）。`拆文库/` 是工程的一部分（喂给项目 `对标/`），不是用完即弃的中间产物、也不是交付物本身 - - 交付物是作者能直接续写的项目。执行时以「建工程」为可见目标，别把「拆文」当成终点或对外标签。 --- ## 核心原则 ### 原则 1：先分析后迁移 先用拆解管道完整拆解小说（输出到 `拆文库/`），再将分析结果迁移为项目结构。**`拆文库/` 是写作工程的一部分**（分析资产，喂给项目 `对标/`），保留不丢弃，不是用完即弃的中间产物。 ### 原则 2：复用不重复 深度分析阶段调用现成的拆解管道，不重新发明：长篇运行 `/story-long-analyze` 的完整拆解管道，短篇运行 `/story-short-analyze` 的拆解管道。拆解方法论与输出模板由对应 analyze skill 自带，story-import 不执行拆解方法论、不维护这些文件。 --- ## Phase 1：确认导入源 问用户：**「你要导入哪本书？请提供文件路径或直接贴文本。」** ### 1.0 确认意图（写作工程 vs 仅拆文库） 默认目标是**完整写作工程**（可续写）。若用户意图不明确 - - 是要可续写的工程，还是只要一份拆文库分析 - - **主动询问**，不要默认： > 「你是想把这本书做成可续写的写作工程（设定/大纲/正文/追踪，能接着写第 N+1 章），还是只要一份拆文库分析？」 - 要可续写工程 → 走完整 story-impor.8.6kinstalls14Story CoverA Chinese-language skill that generates professional web-novel cover images by analyzing a book's title and genre, supporting ten major genre visual styles via GPT-Image. Writers and publishers use it to produce polished cover art tailored to a story's theme without commissioning a designer.0installs