paragraph-curator
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseParagraph Curator (select -> evaluate -> subset -> fuse)
Paragraph Curator(选择→评估→筛选→融合)
Purpose: turn “keep rewriting and getting longer” into a controlled convergence step.
This skill adds a decision layer between “draft paragraphs” and “polish voice”:
- keep the best paragraphs
- merge redundant ones
- rewrite for clearer argument moves
- expand only when coverage is missing (using existing evidence cards)
This is a content-structure pass (not a style pass). Run and after curation.
style-harmonizeropener-variator目的:将“反复重写且内容不断变长”转变为可控的内容收敛步骤。
该Skill在“草稿段落”与“润色措辞”之间新增了一个决策层:
- 保留最优段落
- 合并冗余段落
- 重写以让论证逻辑更清晰
- 仅在内容覆盖不足时进行扩充(使用现有证据卡片)
这是一次内容结构优化(非风格优化)。整理完成后,请运行和。
style-harmonizeropener-variatorInputs
输入项
Required:
- (especially H3 bodies:
sections/)sections/S<sub_id>.md - (what each H3 must cover + allowed citations)
outline/writer_context_packs.jsonl - (single source of truth for terminology + premises)
output/ARGUMENT_SKELETON.md
Recommended:
- (paragraph moves + outputs)
output/SECTION_ARGUMENT_SUMMARIES.jsonl - (paragraph linkage risks)
output/SECTION_LOGIC_REPORT.md - (style smells / scope/citation warnings)
output/WRITER_SELFLOOP_TODO.md
必填:
- (尤其H3级内容:
sections/)sections/S<sub_id>.md - (每个H3需覆盖的内容范围 + 允许使用的引用)
outline/writer_context_packs.jsonl - (术语与前提的唯一权威来源)
output/ARGUMENT_SKELETON.md
推荐:
- (段落功能+输出要点)
output/SECTION_ARGUMENT_SUMMARIES.jsonl - (段落关联风险)
output/SECTION_LOGIC_REPORT.md - (风格问题 / 范围/引用警告)
output/WRITER_SELFLOOP_TODO.md
Outputs
输出项
- Updated (same filenames; body-only; no headings)
sections/*.md - (short; PASS/FAIL + what changed)
output/PARAGRAPH_CURATION_REPORT.md - Create when done (empty file; pipeline contract signal)
sections/paragraphs_curated.refined.ok
- 更新后的(文件名不变;仅包含内容主体;无标题)
sections/*.md - (简短;包含PASS/FAIL状态 + 变更内容)
output/PARAGRAPH_CURATION_REPORT.md - 完成后创建(空文件;用于标识流程契约完成)
sections/paragraphs_curated.refined.ok
What this skill optimizes (rubric)
该Skill的优化方向(评分标准)
You are not trying to “shorten”. You are trying to increase information density while keeping the section verifiable.
Score each paragraph on a simple 0-2 rubric:
| Criterion | 0 (bad) | 1 (ok) | 2 (good) |
|---|---|---|---|
| Coverage | does not match any required axis/card | matches one axis, thin | directly executes a must-use card/comparison |
| Novelty | repeats nearby content | partially redundant | adds a distinct comparison/insight |
| Move clarity | unclear what it does | move exists, weak output | clear move + reusable output |
| Consistency | premise/term drift vs skeleton | minor mismatch | fully aligned with Consistency Contract |
| Citation hygiene | uncited when it should be; cite-dump vibe | acceptable | citations are local and anchored (not just tail) |
| Fusion readiness | cannot merge; tangled | mergeable with edits | clean unit that can be fused or kept |
Decision labels:
- : keep mostly as-is
KEEP - : keep content, rewrite for clearer move/output
REWRITE - : merge with neighbor(s) and rewrite into one stronger paragraph
FUSE - : keep the slot, but rewrite using existing evidence cards (when coverage is missing)
REPLACE
目标并非“缩短内容”,而是在保证章节可验证性的同时提升信息密度。
采用0-2分的简单评分标准对每个段落进行评估:
| 评估维度 | 0分(差) | 1分(合格) | 2分(优) |
|---|---|---|---|
| 内容覆盖 | 未匹配任何要求的维度/卡片 | 匹配一个维度,但内容单薄 | 直接对应必须使用的卡片/对比内容 |
| 内容新颖性 | 重复附近内容 | 部分冗余 | 新增独特的对比/见解 |
| 逻辑清晰度 | 功能定位模糊 | 有明确功能,但输出效果弱 | 功能清晰 + 输出内容可复用 |
| 一致性 | 与骨架文档的前提/术语存在偏差 | 存在微小不匹配 | 完全符合一致性契约 |
| 引用规范 | 应引用未引用;存在堆砌引用的情况 | 符合要求 | 引用本地化且锚定到对应语句(并非仅放在段落末尾) |
| 可融合性 | 无法合并;内容混乱 | 经编辑后可合并 | 内容规整,可直接融合或保留 |
决策标签:
- :基本保持原样
KEEP - :保留内容,重写以让功能/输出更清晰
REWRITE - :与相邻段落合并并重写为更有力的单一段落
FUSE - :保留段落位置,但使用现有证据卡片重写(当内容覆盖不足时)
REPLACE
Paragraph budget (profile-aware)
段落数量预算(基于文档类型)
Default per-H3 target:
- : 10-12 paragraphs
draft_profile=survey - : 11-13 paragraphs
draft_profile=deep
If you exceed the budget, do not delete content blindly. Prefer (merge redundancy) and make the fused paragraph denser.
FUSE默认每个H3的目标段落数:
- :10-12段
draft_profile=survey - :11-13段
draft_profile=deep
若超出预算,请勿盲目删除内容。优先选择(合并冗余内容),并让融合后的段落更紧凑。
FUSEMust-have coverage checklist (per H3)
必选内容覆盖清单(每个H3)
Each H3 must contain at least:
- 1x (only if this H3 introduces a new term/protocol field)
Definition/Setup - 2x concrete paragraphs (A-vs-B comparisons; not just “many papers do...”)
Contrast - 1x paragraph (task + metric + constraint/budget/tool access; cite-backed)
Evaluation anchor - 1x cross-paper paragraph (what generalizes, what does not; cite-backed)
Synthesis - 1x paragraph (limitations; threats to validity; cite-backed when possible)
Boundary/Failure - 1x (a reusable takeaway used downstream)
Local conclusion
If any item is missing, use to write that paragraph from the writer context pack (do not invent new facts).
REPLACE每个H3至少需包含:
- 1个段落(仅当该H3引入新术语/协议字段时需要)
Definition/Setup - 2个具体的段落(A与B的对比;而非仅“多篇论文提及...”)
Contrast - 1个段落(任务+指标+约束/预算/工具权限;有引用支撑)
Evaluation anchor - 1个跨论文的段落(总结通用结论与例外情况;有引用支撑)
Synthesis - 1个段落(局限性;有效性威胁;尽可能有引用支撑)
Boundary/Failure - 1个段落(可复用的结论,供后续环节使用)
Local conclusion
若有任何项缺失,请使用,基于writer context pack撰写对应段落(不得编造新内容)。
REPLACEWorkflow (minimal)
最简工作流程
- Pick the target set
- Start with the H3 bodies listed in , plus any H3 flagged in
output/SECTION_LOGIC_REPORT.mdas repetitive/template-y, plus any H3 that keeps growing across edits.output/WRITER_SELFLOOP_TODO.md - Work file-by-file: each target is a concrete .
sections/S<sub_id>.md
- Build a paragraph inventory (scratch only; do not paste into the paper)
- If exists, use its per-paragraph
output/SECTION_ARGUMENT_SUMMARIES.jsonl/movesas the first draft of your inventory, then reconcile with the actual text. For each paragraph, write one line:output P<i> :: move(s) -> output (1 sentence) :: citations (keys)
- Apply the rubric and label each paragraph
- Mark .
KEEP/REWRITE/FUSE/REPLACE - If two adjacent paragraphs repeat the same axis, .
FUSE - For any paragraph you plan to change (/
REWRITE/REPLACE), draft 2-3 candidate rewrites in parallel (different angles: contrast-first / protocol-first / synthesis-first).FUSE- Score candidates quickly with the rubric; keep one winner (or fuse two if they cover complementary axes).
- Keep citation keys unchanged while sampling; you are choosing surface form + structure, not changing the evidence set.
- Construct the curated set
- Use to enforce must-have coverage (paragraph_plan/must_use/comparison_cards/limitation_hooks) without inventing new content.
outline/writer_context_packs.jsonl - Enforce the must-have coverage checklist.
- Enforce the paragraph budget by fusing redundancy rather than deleting substance.
- Fuse + rewrite (keep citation keys fixed) Rules that keep the pipeline stable:
- Do not add/remove citation keys; when fusing, carry citations forward and re-anchor them to the right sentence.
- Do not move citations across subsections.
- Avoid adjacent citation blocks (e.g., ) and duplicate keys in one block (e.g.,
[@a] [@b]).[@a; @a] - When fusing, it is often faster to write two fused candidates (one contrast-heavy, one synthesis-heavy) and pick the better one.
- Write the report + marker
- should be short and actionable:
output/PARAGRAPH_CURATION_REPORT.md- Status: PASS|FAIL- per H3: paragraph count before/after; what was fused; any remaining gaps
- (minimal) how many candidates you tried for the main rewrites (e.g., 2-3), so future passes can see whether this was a real selection step
- Create .
sections/paragraphs_curated.refined.ok
- 选择目标范围
- 从中列出的H3内容开始,加上
output/SECTION_LOGIC_REPORT.md中标记为重复/模板化的H3,以及在多次编辑中持续变长的H3。output/WRITER_SELFLOOP_TODO.md - 逐文件处理:每个目标对应具体的文件。
sections/S<sub_id>.md
- 建立段落清单(仅用于草稿;请勿粘贴到正式文档)
- 若存在,使用其中的段落
output/SECTION_ARGUMENT_SUMMARIES.jsonl/moves作为清单初稿,再与实际文本核对。output - 为每个段落编写一行信息:
P<i> :: move(s) -> output(一句话总结):: citations(引用标识)
- 应用评分标准并为每个段落打标签
- 标记。
KEEP/REWRITE/FUSE/REPLACE - 若两个相邻段落重复同一维度的内容,标记为。
FUSE - 对于任何计划修改的段落(/
REWRITE/REPLACE),并行撰写2-3个重写候选版本(不同角度:对比优先 / 协议优先 / 综合优先)。FUSE- 用评分标准快速评估候选版本;保留最优版本(若两个版本覆盖互补维度,可融合)。
- 评估过程中保持引用标识不变;仅调整表述形式与结构,不改变证据集合。
- 构建整理后的内容集合
- 使用确保必选内容覆盖(paragraph_plan/must_use/comparison_cards/limitation_hooks),且不编造新内容。
outline/writer_context_packs.jsonl - 执行必选内容覆盖清单要求。
- 通过融合冗余内容而非删除核心信息来控制段落数量预算。
- 融合 + 重写(保持引用标识固定) 维持流程稳定的规则:
- 不得添加/移除引用标识;融合时,将引用带入新段落并重新锚定到对应语句。
- 不得跨小节移动引用。
- 避免相邻的引用块(如)和同一引用块中的重复标识(如
[@a] [@b])。[@a; @a] - 融合时,通常可以快速撰写两个融合候选版本(一个侧重对比,一个侧重综合),再选择更优的版本。
- 撰写报告 + 标记文件
- 应简短且具备可操作性:
output/PARAGRAPH_CURATION_REPORT.md- 状态: PASS|FAIL- 每个H3:整理前后的段落数量;融合的内容;剩余的内容缺口
- (简要说明)主要重写环节尝试了多少个候选版本(如2-3个),以便后续环节了解是否为真正的筛选步骤
- 创建文件。
sections/paragraphs_curated.refined.ok
Routing rules
路由规则
- If you cannot fill a missing must-have paragraph without new evidence: stop and route upstream (/ C3-C4). Do not pad.
evidence-selfloop - If you feel forced to change a definition or evaluation premise: update first, then rerun
output/ARGUMENT_SKELETON.md# Consistency Contract.argument-selfloop - If the only issue is surface cadence/openers: do not overwork curation; run /
style-harmonizer.opener-variator
- 若无法在不新增证据的情况下填补必选段落的缺口:停止操作并路由到上游环节(/ C3-C4)。请勿凑数。
evidence-selfloop - 若必须修改定义或评估前提:先更新,再重新运行
output/ARGUMENT_SKELETON.md# Consistency Contract。argument-selfloop - 若仅存在措辞节奏/开头问题:无需过度投入整理工作;直接运行/
style-harmonizer。opener-variator
Done checklist
完成检查清单
- Each targeted H3 stays within its paragraph budget (survey 10-12; deep 11-13) without losing required moves.
- Redundant paragraphs are fused into denser, clearer ones (not just deleted).
- No citation keys were added/removed; citation shape is reader-facing (no adjacent blocks, no dup keys).
- exists and is understandable.
output/PARAGRAPH_CURATION_REPORT.md - exists.
sections/paragraphs_curated.refined.ok
- 每个目标H3在不丢失必要功能的前提下,段落数量符合预算(survey类型10-12段;deep类型11-13段)。
- 冗余段落已融合为更紧凑、清晰的段落(而非直接删除)。
- 未添加/移除引用标识;引用格式便于阅读(无相邻引用块、无重复标识)。
- 已存在且内容易懂。
output/PARAGRAPH_CURATION_REPORT.md - 已存在。
sections/paragraphs_curated.refined.ok