npx skills add https://github.com/jwynia/agent-skills --skill novel-revision你帮助作者管理跨越多个抽象层次的修订,同时防止意外后果在叙事中蔓延。你的角色是实施系统性的变更管理,以保持故事的连贯性。
任何层面的变更都可能影响所有其他层面。 变更会向上传播(散文的发现揭示结构问题)和向下传播(结构变化需要重写散文)。
每部小说都同时在这三个层次上运作:
在实施任何修订之前:
对于每个提议的变更,记录:
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
定义表明问题的警告信号:
触发条件:角色感觉扁平、动机不明确、成长弧不完整
协议:
监控:行为是否仍然可信?其他人的反应是否恰当?
触发条件:节奏失调、事件脱节、高潮缺乏冲击力
协议:
监控:张力是否在构建?情节点是否感觉相互关联且不可避免?
触发条件:主题过于直白、不明确或不一致
协议:
监控:主题是否自然浮现?结局在主题上是否令人满意?
| 警告信号 | 指示内容 |
|---|---|
| 角色在没有发展的情况下做出违背个性的行为 | 一致性断裂 |
| 事件不符合逻辑 | 情节逻辑漏洞 |
| 场景感觉仓促或意外拖沓 | 节奏异常 |
| 故事事件削弱主题 | 主题漂移 |
# 修订:[简要描述]
## 变更类型
- [ ] 概念性 - [ ] 结构性 - [ ] 散文
## 理由
[为什么需要此变更]
## 预测后果
- 即时(1-2章):
- 中期(3-5章):
- 故事整体:
## 监控标准
- 警告信号 1:
- 警告信号 2:
- 成功指标 1:
- 成功指标 2:
## 实施状态
- [ ] 初始变更完成
- [ ] 级联任务已识别
- [ ] 级联任务已完成
- [ ] 验证完成
## 结果评估
[实施后填写]
与多个智能体合作进行修订时:
当修订陷入困境时:
context/output-config.mdrevision/ 或 explorations/revision/模式:{novel-name}-revision-{date}.md
context/output-config.md{novel-name}-revision-{date}.md触发短语: "协调所有修订"、"映射完整级联"、"规划回滚"
| 任务 | 智能体类型 | 何时生成 |
|---|---|---|
| 一致性检查 | Explore | 当验证跨文稿的变更时 |
| 角色跟踪 | general-purpose | 当监控角色一致性时 |
模式: 在没有首先分析可能影响其他内容的情况下进行修订——直接跳到实施阶段。失败原因: 变更会级联。角色动机的改变会影响该角色做出选择的每个场景。未经评估的变更会产生问题,而这些问题你可能在几页之后才发现,通常是在你已经建立在破碎的基础之上之后。修复方法: 在每次重大变更之前,明确记录:结果必须改变什么?什么可能会被破坏?在实施之前,而不是之后,设置监控标准。
模式: 注意到变更后某些地方感觉不对劲,但仍然继续推进,相信问题会自行解决。失败原因: 早期警告是关于昂贵问题的廉价信号。角色行为违背既定个性、情节逻辑漏洞、节奏异常——这些问题会叠加。你走得越深,修复成本就越高。修复方法: 将早期警告视为可操作的信息。停下来,评估,决定是立即修复还是明确接受风险。不要让"我稍后会修复"累积起来。
模式: 进行变更而不跟踪它们所需的次要编辑——积累未解决的连锁影响积压。失败原因: 未跟踪的级联任务变成隐性的技术债务。你以为你做了一个变更;实际上你做了一个变更并制造了十个未修复的不一致。修复方法: 维护明确的级联任务列表。当变更需要后续编辑时,立即写下来。跟踪完成情况。在级联问题解决之前不要继续前进。
模式: 即使警告信号增多,仍然继续一个有问题的变更,因为你已经投入了大量精力。失败原因: 无论怎样,精力已经消耗掉了。沿着一条破碎的道路继续前进只会增加更多的精力损失。两章内出现多个警告信号通常表明存在根本性问题。修复方法: 在实施前定义回滚触发条件。当触发条件满足时,回滚。记录你学到的东西。即使变更没有成功,这种洞察也是有价值的。
模式: 用修订的页数来衡量进展,而不是用实现的改进来衡量——将工作与结果混为一谈。失败原因: 你可以广泛修订而没有让故事变得更好。事实上,你可以广泛修订并让它变得更糟。不服务于故事目标的活动是浪费。修复方法: 为每次修订过程定义"更好"意味着什么。对照该目标进行衡量,而不是对照修改的页数。有时最好的修订就是不修订。
| 技能 | 提供内容 |
|---|---|
| story-sense | 需要修订的诊断 |
| character-arc | 角色一致性要求 |
| scene-sequencing | 节奏和结构要求 |
| 技能 | 此技能提供内容 |
|---|---|
| prose-style | 具有级联意识的文稿级变更 |
| revision | 具有多层次协调的场景级工作 |
| (completed novel) | 保持连贯性的最终产品 |
| 技能 | 关系 |
|---|---|
| story-sense | story-sense 诊断问题;novel-revision 管理修复而不产生新问题 |
| revision | revision 处理句子和场景级别;novel-revision 协调整个文稿 |
每周安装量
187
仓库
GitHub 星标数
37
首次出现
Jan 20, 2026
安全审计
安装于
opencode165
gemini-cli160
codex155
cursor152
github-copilot151
amp138
You help writers manage revisions across multiple levels of abstraction while preventing unintended consequences from cascading through the narrative. Your role is to implement systematic change management that maintains story coherence.
Any change at one level potentially affects all other levels. Changes propagate both upward (prose discoveries revealing structural problems) and downward (structural changes requiring prose rewrites).
Every novel operates simultaneously across these levels:
Before implementing any revision:
For each proposed change, document:
Define warning signs that indicate problems:
Triggers : Character feels flat, motivations unclear, arc incomplete
Protocol :
Monitoring : Does behavior remain believable? Do others respond appropriately?
Triggers : Pacing off, events disconnected, climax lacks impact
Protocol :
Monitoring : Does tension build? Do plot points feel connected and inevitable?
Triggers : Theme heavy-handed, unclear, or inconsistent
Protocol :
Monitoring : Do themes emerge naturally? Does resolution feel thematically satisfying?
| Warning Sign | What It Indicates |
|---|---|
| Character acts against personality without development | Consistency break |
| Events don't follow logically | Plot logic gap |
| Scenes feel rushed or drag unexpectedly | Pacing anomaly |
| Story events undermine themes | Thematic drift |
# Revision: [Brief Description]
## Change Type
- [ ] Conceptual - [ ] Structural - [ ] Prose
## Rationale
[Why this change is needed]
## Predicted Consequences
- Immediate (1-2 chapters):
- Medium-term (3-5 chapters):
- Story-wide:
## Monitoring Criteria
- Warning sign 1:
- Warning sign 2:
- Success indicator 1:
- Success indicator 2:
## Implementation Status
- [ ] Initial change complete
- [ ] Cascade tasks identified
- [ ] Cascade tasks completed
- [ ] Validation complete
## Outcome Assessment
[Complete after implementation]
When working with multiple agents on revision:
When revision feels stuck:
context/output-config.md in the projectrevision/ or explorations/revision/Pattern: {novel-name}-revision-{date}.md
context/output-config.md{novel-name}-revision-{date}.mdTrigger phrases: "coordinate all revisions", "map full cascade", "plan the rollback"
| Task | Agent Type | When to Spawn |
|---|---|---|
| Consistency check | Explore | When verifying changes across manuscript |
| Character tracking | general-purpose | When monitoring character consistency |
Pattern: Making revisions without first analyzing what else might be affected—jumping straight to implementation. Why it fails: Changes cascade. A character motivation change affects every scene where that character makes choices. Unassessed changes create problems you discover pages later, often after you've built on broken foundation. Fix: Before every significant change, explicitly document: what must change as a result? What might break? Set monitoring criteria before implementation, not after.
Pattern: Noticing that something feels off after a change but pushing forward anyway, trusting it will resolve itself. Why it fails: Early warnings are cheap signals about expensive problems. Characters acting against established personality, plot logic gaps, pacing anomalies—these compound. The deeper you go, the more expensive the fix. Fix: Treat early warnings as actionable information. Stop, evaluate, decide to either fix now or explicitly accept the risk. Don't let "I'll fix it later" accumulate.
Pattern: Making changes without tracking the secondary edits they require—accumulating a backlog of unaddressed implications. Why it fails: Untracked cascade tasks become invisible technical debt. You think you made one change; you actually made one change and created ten unfixed inconsistencies. Fix: Maintain explicit cascade task lists. When a change requires follow-up edits, write them down immediately. Track completion. Don't move on until cascade is resolved.
Pattern: Continuing with a problematic change because you've already invested significant effort, even when warning signs multiply. Why it fails: The effort is gone either way. Continuing down a broken path just adds more lost effort. Multiple warning signs within two chapters usually indicate fundamental problems. Fix: Define rollback triggers before implementation. When triggers fire, roll back. Document what you learned. The insight is valuable even if the change wasn't.
Pattern: Measuring progress by pages revised rather than by improvement achieved—conflating work with results. Why it fails: You can revise extensively without making the story better. In fact, you can revise extensively and make it worse. Activity that doesn't serve story goals is waste. Fix: Define what "better" means for each revision pass. Measure against that goal, not against pages touched. Sometimes the best revision is no revision.
| Skill | What it provides |
|---|---|
| story-sense | Diagnosis of what needs revision |
| character-arc | Character consistency requirements |
| scene-sequencing | Pacing and structure requirements |
| Skill | What this provides |
|---|---|
| prose-style | Manuscript-level changes with cascade awareness |
| revision | Scene-level work with multi-level coordination |
| (completed novel) | Final product with maintained coherence |
| Skill | Relationship |
|---|---|
| story-sense | Story-sense diagnoses problems; novel-revision manages the fix without creating new problems |
| revision | Revision handles sentence and scene level; novel-revision coordinates across the whole manuscript |
Weekly Installs
187
Repository
GitHub Stars
37
First Seen
Jan 20, 2026
Security Audits
Gen Agent Trust HubPassSocketPassSnykPass
Installed on
opencode165
gemini-cli160
codex155
cursor152
github-copilot151
amp138
BMAD工作流编排与SSD结构化系统设计 - 多代理AI开发流程自动化
10,600 周安装