yrzsh/paper-workbench/paper-staged-gen
Claude Code and agent skill by yrzsh/paper-workbench
Trust tier 2
- community
Source check pending
Checking that the SKILL.md file is still in the repository.
Use when generating or regenerating a full manuscript with the DSH Paper Workbench staged pipeline — contract-first, per-section generation with citation-pool gates, programmatic tables, post-hoc abstract. Routes the `wb.py generate` subcommands; never one-shot a long manuscript.
Install
npx vskill@latest install yrzsh/paper-workbench/paper-staged-genSKILL.md
Paper Staged Generation (paper-staged-gen)
分批生成编排协议。配套引擎:~/.dsh/papers/workbench/staged_gen.py + wb.py generate。 设计见工作区《分批生成方案》。核心信条:长文/多图表稿件绝不一次性整篇生成。
何时触发
- 用户要"写一篇综述/长文"或"重写 main.md",且目标稿 > 3000 词或含 ≥2 张图表
- 一次性生成的稿件出现表格损坏、图文分离、幻觉引用、摘要漂移,需要重写
- 已有
draft/contract.md,要继续/恢复分段生成
不可违反的纪律
- 契约先行,锁定才写:
draft/contract.md未"已锁定"前,禁止生成任何正文段落。锁定是人工检查点(S0),你只能提示用户去锁定,不能替用户锁定。 - 引文池硬约束:每段只能引用契约「引文分配」中分给该段的文献编号。段级门禁会自动拦截池外引用——被拦时不要硬塞,回头检查分配是否合理。
- 表格只产数据不产排版:表格一律
generate tables(JSON→管道表)。你只负责产出行数据(或--extract迁移存量),绝不手写表格排版。 - 摘要后置:摘要在全部正文段落定稿后用
generate abstract生成,基于各段摘要而非全文;对齐校验失败就修,不要绕过。 - 失败重试有上限:单段门禁失败自动重试 1 次,仍失败则停下报告用户,不要无限循环。
标准编排序列(S0→S8)
每段生成后必须跑 generate status 确认 done 再进下一段;出现 failed 先看 issues 再决定重试或人工。
段级门禁三件套(自动,勿绕过)
- 占位符/双标点/超长句(复用
toolbox.mechanical_check的 P0/P1) - 引文范围 ⊆ 本节分配池(幻觉引用段级拦截)
- 字数预算 ±20%
决策快捷
- "这段引用了不在分配池的文献" → 查契约引文分配,缺则补分配并提醒用户重新锁定,不要删引用硬凑
- "表格又乱了" → 一定是有人手写了表格;改走
generate tablesJSON 渲染 - "摘要说了正文没有的" → 对齐校验会拦;让摘要回退到各段摘要重写,不要从全文"发挥"
- "能不能一次生成完" → 不能。除非稿件 < 2500 词且无图表,否则坚持分段
与其它技能的关系
- 写作风格仍从
quality_rules.md+build_stage_prompt取;本技能只管生成编排与门禁,不改写作风格源 - 审查仍走
review-auto门禁,本技能不替代审查 - 期刊格式以
journal/chosen.md为准,跨刊不得套用其它期刊预算
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/paper-staged-gen/SKILL.md
Popularity
- Repository stars
- 88
- Repository forks
- 7
- 7-day trend score
- 22
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/yrzsh/paper-workbench/paper-staged-gen)