fiction-workshop by rhavekost/author-toolkit
npx skills add https://github.com/rhavekost/author-toolkit --skill fiction-workshop协作小说创作的三阶段编辑工作流:故事设定构建、章节开发、读者测试。
此技能适用于:
此技能不适用于:
对于叙事性非虚构作品(回忆录、带有故事元素的自助类书籍),请改用 narrative-nonfiction 技能。
在章节开发阶段,通过请求特定视角在这些角色间切换:
| 角色 | 调用方式 | 关注点 |
|---|---|---|
| 发展性编辑 | "作为发展性编辑..." | 情节、节奏、结构、利害关系、主题 |
| 文字编辑 | "作为文字编辑..." | 散文节奏、措辞、"展示而非讲述" |
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
| "作为角色顾问..." |
| 声音一致性、动机、角色弧光、关系 |
| 连续性追踪员 | "作为连续性追踪员..." | 时间线、世界事实、内部一致性 |
| 头脑风暴伙伴 | "头脑风暴模式..." | "如果...会怎样"探索、问题解决、摆脱困境 |
有关每个角色的详细指导,请参阅 references/。
目标: 在起草或编辑之前,建立共享的故事基础。
情节: 前提、三幕结构/节拍表、重大转折、结局(即使是粗略的)
角色: 主角(欲望/需求/创伤/弧光)、反派(动机/威胁)、配角(功能/关系)、POV 声音笔记
世界: 设定(时间/地点/规则)、科技/魔法系统、社会结构、感官调色板
主题: 核心问题、道德论点、反复出现的主题
如果已有故事设定文档,请审阅它。如果没有,可提议使用 assets/story-bible-template.md 创建一个。
故事设定条目示例(角色):
ALEX CHEN - 主角
欲望:揭露阴谋并洗清自己的罪名
需求:学会相信自己的直觉而非机构权威
创伤:导师在前机构背叛了她,导致事业受挫
弧光:独行侠 → 意识到她需要盟友 → 与团队建立信任
声音笔记:分析性强,压力大时有冷幽默,避免使用情感化语言
关键关系:与处理人之间的紧张关系(想要信任,但无法完全信任)
退出条件: 对故事基础有充分把握。能够讨论角色动机、预测情节影响并识别主题线索,而无需询问基本问题。
目标: 通过头脑风暴 → 筛选 → 起草 → 精炼的循环来起草或完善章节。
起草新内容? → 创作工作流 | 编辑现有内容? → 编辑工作流
示例(惊悚场景): 同一辆车连续三天停在咖啡馆外 | 凌晨 3 点手机响起,来电号码被屏蔽 | 门缝下发现监控照片 | 同事提到只有监控才会知道的细节 | 窗户上反射出相机镜头 | 情报投放点已被清理 | 安全屋钥匙不匹配 | 联系人错过第一次报到
然后筛选:"哪些能制造即时紧张感?可以组合哪些?"
str_replace 进行修订,切勿重新打印。str_replace。修改后链接到文件。当调用特定编辑角色时,加载相应的参考文件:
references/developmental-editing.mdreferences/line-editing.mdreferences/character-work.mdreferences/continuity-tracking.mdreferences/brainstorming.mdreferences/thriller-craft.mdreferences/scifi-worldbuilding.md目标: 验证手稿在没有作者背景信息的情况下是否有效。
使用全新的子代理(无故事设定):
常见问题: 动机不明确 | 节奏拖沓 | 情节发展未铺垫 | 困惑
如果遇到困难: 识别差距 → 返回阶段 2 → 重新测试
退出条件: 读者无需作者解释即可理解并投入其中。
使用以下检查点来验证你是否正确遵循了工作流。
故事设定构建之后:
调用角色之后:
进行编辑之后:
str_replace 进行精准修改,而不是重新打印整个部分?头脑风暴之后:
宣布"完成"之前:
如果你对任何检查点的回答是"否",请在继续之前返回该阶段。
| 错误 | 原因 | 解决方法 |
|---|---|---|
| 跳过故事设定 | "我足够了解我的故事" | 故事设定不是为你准备的——它是为 Claude 准备的。没有共享的上下文,反馈会错失关键的故事元素。构建它。 |
| 无特定角色的通用反馈 | 匆忙,忘记调用特定角色 | 在提示中明确说"作为发展性编辑..."或"作为文字编辑..."。不同的视角能发现不同的问题。 |
| 重新打印整个章节 | 其他编辑场景的习惯 | 仅使用 str_replace 进行精准编辑。重新打印会消耗上下文并使更改难以追踪。编辑后链接到文件。 |
| 在结构问题解决前跳转到文字编辑 | 想立即"修复"散文 | 如果存在情节/节奏/角色问题,文字编辑就是白费力气。总是先进行发展性编辑审阅。参见下面的示例。 |
| 跳过读者测试 | "我已经读过太多遍了" | 你拥有作者背景。读者测试使用没有故事设定的全新子代理来发现读者会遇到的问题。 |
| 一次使用太多角色 | 试图一次性修复所有问题 | 每次审阅调用一个角色。发展性编辑 → 角色顾问 → 文字编辑 → 连续性追踪员。聚焦的反馈才是可操作的反馈。 |
| 头脑风暴后不筛选 | 采用第一个听起来不错的想法 | 生成 5-15 个选项,然后筛选。第一个想法很少是最好的想法。数量带来质量。 |
同一段落,不同视角:
Sarah walked into the office. Her boss looked angry. "We need to talk," he said. She sat down nervously.
文字编辑反馈(散文层面):
发展性编辑反馈(结构/利害关系):
区别: 文字编辑打磨句子。发展性编辑确保场景在故事中占有一席之地。总是先进行发展性编辑。
| 需求 | 命令 |
|---|---|
| 开始新项目 | "让我们为 [项目] 构建一个故事设定" |
| 发展性编辑审阅 | "作为发展性编辑,分析 [章节/部分]" |
| 文字编辑 | "作为文字编辑,润色 [场景/段落]" |
| 角色检查 | "作为角色顾问,[角色] 的 [行为] 是否符合其性格?" |
| 连续性审计 | "作为连续性追踪员,检查 [第 X-Y 章] 是否存在不一致之处" |
| 摆脱困境 | "头脑风暴模式——我需要 [解决问题]" |
| 测试可读性 | "对 [章节/部分] 进行一次全新阅读" |
references/developmental-editing.md - 情节、结构、节奏分析references/line-editing.md - 散文层面的润色references/character-work.md - 声音、动机、角色弧光追踪references/continuity-tracking.md - 时间线和事实一致性references/brainstorming.md - 创意生成技巧references/thriller-craft.md - 悬疑题材的特定指导references/scifi-worldbuilding.md - 技术准确性、推测规则assets/story-bible-template.md - 空白故事设定结构assets/scene-worksheet.md - 场景级分析模板每周安装量
283
仓库
GitHub 星标
2
首次出现
2026年1月27日
安全审计
安装于
opencode267
gemini-cli257
codex257
cursor254
github-copilot253
kimi-cli244
Editorial workflow for collaborative fiction writing in three stages: Story Bible Building, Chapter Development, and Reader Testing.
This skill is for:
This skill is NOT for:
For narrative nonfiction (memoir, self-help with story elements), use the narrative-nonfiction skill instead.
Switch between these roles during Chapter Development by requesting a specific lens:
| Role | Invocation | Focus |
|---|---|---|
| Developmental Editor | "As developmental editor..." | Plot, pacing, structure, stakes, theme |
| Line Editor | "As line editor..." | Prose rhythm, word choice, "show don't tell" |
| Character Consultant | "As character consultant..." | Voice consistency, motivation, arc, relationships |
| Continuity Tracker | "As continuity tracker..." | Timeline, world facts, internal consistency |
| Brainstorm Partner | "Brainstorm mode..." | "What if" exploration, problem-solving, unsticking |
See references/ for detailed guidance on each role.
Goal: Establish shared story foundation before drafting or editing.
Plot: Premise, three-act structure/beat sheet, major turns, ending (even if rough)
Characters: Protagonist (want/need/wound/arc), antagonist (motivation/threat), supporting cast (function/relationships), POV voice notes
World: Setting (time/place/rules), tech/magic systems, social structures, sensory palette
Theme: Central question, moral argument, recurring motifs
If a Story Bible document exists, review it. If not, offer to create one using assets/story-bible-template.md.
Example Story Bible entry (character):
ALEX CHEN - Protagonist
Want: Expose the conspiracy and clear her name
Need: Learn to trust her instincts over institutional authority
Wound: Mentor betrayed her at previous agency, causing career setback
Arc: Lone wolf → realizes she needs allies → builds trust with team
Voice notes: Analytical, dry humor when stressed, avoids emotional language
Key relationship: Tension with Handler (wants to trust, can't fully)
Exit condition: Confident grasp of story fundamentals. Can discuss character motivations, predict plot implications, and identify thematic threads without asking basic questions.
Goal: Draft or refine chapters through brainstorm → curate → draft → refine cycles.
Drafting new? → Creation workflow | Editing existing? → Editing workflow
Scene Planning
Brainstorm Beats (5-15 options): Opening hooks, key moments, dialogue, sensory details, closing
Example (thriller scene): Same car outside coffee shop three days running | Phone buzzing at 3am with blocked caller | Surveillance photo under door | Colleague mentions detail only surveillance would know | Camera lens reflection in window | Dead drop cleaned out | Safe house key doesn't fit | Contact misses first check-in
Then curate: "Which create immediate tension? Combine any?"
Curate: Ask which to keep, combine, or discard. Reasons help calibrate.
Draft: Write chapter. Use str_replace for revisions, never reprint.
Refine: Iterate on feedback. After 3 passes with minimal changes, ask: "What could be cut?"
Read and Diagnose: What chapter tries to do, where it succeeds, where it loses energy/clarity
Invoke Persona: Structure/pacing → Developmental | Prose → Line | Voice → Character | Facts → Continuity
Propose Changes: Specific, surgical edits with brief "why"
Implement: Use str_replace. Link to file after changes.
Iterate: Until chapter achieves purpose
When a specific editorial persona is invoked, load the corresponding reference file:
references/developmental-editing.mdreferences/line-editing.mdreferences/character-work.mdreferences/continuity-tracking.mdreferences/brainstorming.mdreferences/thriller-craft.mdreferences/scifi-worldbuilding.mdGoal: Verify manuscript works without author context.
Using fresh sub-agent (no story bible):
Common issues: Unclear motivation | Pacing lags | Unearned moments | Confusion
If struggles: Identify gap → Return to Stage 2 → Re-test
Exit condition: Reader understands and engages without author explanations.
Use these checkpoints to verify you're following the workflow correctly.
After Story Bible building:
After invoking a persona:
After making edits:
str_replace for surgical changes, not reprinting entire sections?After brainstorming:
Before claiming "done":
If you answered "no" to any checkpoint, return to that stage before proceeding.
| Mistake | Why It Happens | Fix |
|---|---|---|
| Skipping Story Bible | "I know my story well enough" | Story Bible isn't for you—it's for Claude. Without shared context, feedback will miss key story elements. Build it. |
| Generic feedback without persona | Rushing, forgetting to invoke specific role | Explicitly say "As developmental editor..." or "As line editor..." in your prompt. Different lenses catch different issues. |
| Reprinting entire chapters | Habit from other editing contexts | Use str_replace for surgical edits only. Reprinting burns context and makes changes hard to track. Link to file after edits. |
| Jumping to line edits before structure | Wanting to "fix" prose immediately | If plot/pacing/character issues exist, line edits are wasted effort. Always developmental pass first. See example below. |
| Skipping Reader Testing | "I've read it so many times already" | You have author context. Reader Testing uses fresh sub-agent without story bible to catch gaps readers will hit. |
Same passage, different lenses:
Sarah walked into the office. Her boss looked angry. "We need to talk," he said. She sat down nervously.
Line Editor feedback (prose-level):
Developmental Editor feedback (structure/stakes):
The difference: Line edits polish sentences. Developmental edits ensure the scene earns its place in the story. Always developmental first.
| Need | Command |
|---|---|
| Start new project | "Let's build a story bible for [project]" |
| Developmental pass | "As developmental editor, analyze [chapter/section]" |
| Line edit | "As line editor, polish [scene/passage]" |
| Character check | "As character consultant, is [character]'s [action] in character?" |
| Continuity audit | "As continuity tracker, check [chapters X-Y] for inconsistencies" |
| Get unstuck | "Brainstorm mode—I need to [solve problem]" |
| Test readability | "Run a fresh read on [chapter/section]" |
references/developmental-editing.md - Plot, structure, pacing analysisreferences/line-editing.md - Prose-level refinementreferences/character-work.md - Voice, motivation, arc trackingreferences/continuity-tracking.md - Timeline and fact consistencyreferences/brainstorming.md - Idea generation techniquesreferences/thriller-craft.md - Genre-specific guidance for suspensereferences/scifi-worldbuilding.md - Technical accuracy, speculation rulesassets/story-bible-template.md - Blank story bible structureassets/scene-worksheet.md - Scene-level analysis templateWeekly Installs
283
Repository
GitHub Stars
2
First Seen
Jan 27, 2026
Security Audits
Gen Agent Trust HubPassSocketPassSnykPass
Installed on
opencode267
gemini-cli257
codex257
cursor254
github-copilot253
kimi-cli244
超能力技能使用指南:AI助手技能调用优先级与工作流程详解
41,800 周安装
| Too many personas at once | Trying to fix everything in one pass | Invoke one persona per pass. Developmental → Character → Line → Continuity. Focused feedback is actionable feedback. |
| Brainstorming without curation | Taking first idea that sounds good | Generate 5-15 options, then curate. First idea is rarely best idea. Quantity enables quality. |