Back to Details

conference-writing-adapter

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Conference Writing Adapter

会议写作适配工具

Adapt an existing research paper to the writing conventions, reviewer expectations, and paper archetypes of a target ML/AI conference. The goal is to make the paper easier for the right reviewers to understand, trust, and champion without inventing unsupported claims or new experimental results.
Use this skill for:
  • paper structure diagnosis before submission
  • conference-specific rewrite plans
  • abstract, introduction, method, experiment, limitation, and related-work rewrites
  • venue-aware figure/table story, visual style, and result presentation decisions
  • paragraph-level outlines for an existing draft
  • learning from accepted, oral, spotlight, best-paper, or highly discussed papers at the target venue
  • accumulating reusable knowledge about venue taste and successful paper patterns
Do not use this skill as a substitute for running experiments, proving claims, or checking final submission compliance. Pair it with 
submit-paper
for final readiness checks and with experiment skills when the paper has evidence gaps. Use 
figure-results-review
when the main issue is whether a figure/table's claim support, style, caption, or visual encoding is paper-ready.
Pair this skill with 
research-project-memory
when rewriting changes paper claims, section structure, figure/table roles, writing risks, or experiment actions.
将现有研究论文适配到目标ML/AI会议的写作惯例、审稿人预期以及论文范式。目标是让合适的审稿人更容易理解、信任并支持你的论文，同时不编造无依据的主张或新的实验结果。
本技能适用于以下场景：
  • 提交前的论文结构诊断
  • 针对特定会议的重写方案
  • 摘要、引言、方法、实验、局限性及相关工作部分的重写
  • 符合会议风格的图表叙事、视觉样式及结果呈现决策
  • 现有草稿的段落级大纲梳理
  • 学习目标会议中已录用、口头报告、 spotlight、最佳论文或高关注度论文的写作方式
  • 积累关于会议偏好和成功论文模式的可复用知识
请勿将本技能替代实验开展、主张验证或最终提交合规性检查。若需最终就绪检查，请搭配
submit-paper
技能使用；若论文存在证据缺口，请搭配实验类技能使用；若核心问题是图表的主张支撑、样式、标题或视觉编码是否符合论文要求，请使用
figure-results-review
技能。
当重写涉及论文主张、章节结构、图表作用、写作风险或实验操作变更时，请搭配
research-project-memory
技能使用。

Skill Directory Layout

技能目录结构

text
<installed-skill-dir>/
├── SKILL.md
└── references/
    ├── conference-profiles.md
    ├── exemplar-analysis.md
    ├── memory-model.md
    ├── paper-archetypes.md
    ├── quality-gates.md
    └── paragraph-protocol.md
text
<installed-skill-dir>/
├── SKILL.md
└── references/
    ├── conference-profiles.md
    ├── exemplar-analysis.md
    ├── memory-model.md
    ├── paper-archetypes.md
    ├── quality-gates.md
    └── paragraph-protocol.md

Progressive Loading

渐进式加载

  • Always read 
    references/paper-archetypes.md
    and 
    references/paragraph-protocol.md
    .
  • Read 
    references/conference-profiles.md
    when the user names a target conference or asks to compare venues.
  • Read 
    references/exemplar-analysis.md
    when the task involves learning from accepted, oral, spotlight, best-paper, or otherwise strong target-venue papers.
  • Read 
    references/quality-gates.md
    before finalizing a rewrite plan or rewritten section.
  • Read 
    references/memory-model.md
    whenever learning from accepted papers, saving venue knowledge, or reusing prior knowledge.
  • When target venue rules may affect writing or formatting, verify current official instructions from the venue website. Do not rely only on memory for page limits, checklist requirements, anonymity, supplementary rules, or ethics/impact requirements.
  • 务必阅读
    references/paper-archetypes.md
    和
    references/paragraph-protocol.md
    。
  • 当用户指定目标会议或要求对比不同会议时，阅读
    references/conference-profiles.md
    。
  • 当任务涉及学习目标会议中已录用、口头报告、spotlight、最佳论文或其他优质论文时，阅读
    references/exemplar-analysis.md
    。
  • 在确定重写方案或完成重写章节前，阅读
    references/quality-gates.md
    。
  • 每当学习已录用论文、保存会议知识或复用已有知识时，阅读
    references/memory-model.md
    。
  • 当目标会议规则可能影响写作或格式时，请从会议官网核实当前官方指南。请勿仅依赖记忆判断页数限制、 checklist要求、匿名规则、补充材料规则或伦理/影响要求。

Core Principles

核心原则

  • Optimize for reviewer cognition, not ornamental prose. A good paper makes the contribution, evidence, and limitations easy to evaluate.
  • Match the paper to an archetype before rewriting. A method paper, benchmark paper, theory paper, dataset paper, system paper, and empirical study need different narrative skeletons.
  • Learn from venue exemplars. For a target conference, inspect accepted/oral/spotlight/best-paper examples when available, then extract reusable writing patterns rather than copying text.
  • Keep every paragraph accountable. Each paragraph should have one reviewer-facing job: motivate, identify a gap, state an insight, define a method component, support a claim, preempt a concern, interpret evidence, or delimit scope.
  • Do not fabricate evidence. If the paper needs an experiment, ablation, theorem, user study, or analysis that does not exist, mark it as a writing blocker and suggest how to phrase the current limitation.
  • Preserve the user's scientific intent. Improve positioning, structure, and clarity without changing the underlying claim beyond available evidence.
  • 以审稿人的认知为优化目标，而非追求华丽辞藻。优质论文应让贡献、证据和局限性易于评估。
  • 重写前先匹配论文范式。方法类论文、基准类论文、理论类论文、数据集类论文、系统类论文和实证研究需要不同的叙事框架。
  • 向会议范例学习。针对目标会议，若有可用的已录用/口头报告/spotlight/最佳论文示例，先研究它们，再提炼可复用的写作模式，而非直接复制文本。
  • 每个段落都要有明确作用。每个段落都应承担一项面向审稿人的任务：激发兴趣、指出缺口、阐述见解、定义方法组件、支撑主张、预先回应疑虑、解读证据或界定范围。
  • 不得编造证据。若论文需要现有实验、消融实验、定理、用户研究或分析以外的内容，需将其标记为写作障碍，并建议如何表述当前局限性。
  • 保留用户的科学意图。在现有证据范围内优化定位、结构和清晰度，不改变核心主张。

Step 1 - Define the Adaptation Task

步骤1 - 定义适配任务

Identify:
  • target conference and year, if known
  • submission track, if relevant
  • current paper format: LaTeX source, PDF, Markdown draft, outline, or notes
  • desired output: diagnosis, rewrite plan, paragraph blueprint, rewritten section, full-paper pass, or venue knowledge note
  • paper stage: idea, early draft, full draft, rebuttal revision, camera-ready
  • whether internet access or user-provided exemplar papers are available
If the user only says a venue name and provides a draft, default to:
  1. diagnose the paper archetype
  2. compare it with target venue expectations
  3. produce a section-level and paragraph-level rewrite plan
  4. rewrite the abstract and introduction if the draft contains enough evidence
明确以下信息：
  • 目标会议及年份（若已知）
  • 提交赛道（若相关）
  • 当前论文格式：LaTeX源码、PDF、Markdown草稿、大纲或笔记
  • 期望输出：诊断报告、重写方案、段落蓝图、重写章节、全文润色或会议知识笔记
  • 论文阶段：想法阶段、早期草稿、完整草稿、 rebuttal修订版、终稿
  • 是否可访问互联网或用户是否提供了范例论文
若用户仅提供会议名称和草稿，默认执行以下操作：
  1. 诊断论文范式
  2. 对比目标会议的预期要求
  3. 生成章节级和段落级重写方案
  4. 若草稿包含足够证据，重写摘要和引言

Step 2 - Gather Paper Evidence

步骤2 - 收集论文证据

Prefer primary draft files and notes over memory.
Look for:
  • main.tex
    , 
    paper.tex
    , 
    sections/*.tex
    , 
    *.bib
    , 
    README.md
    , 
    docs/
    , 
    notes/
  • title, abstract, introduction, contribution list
  • figures and tables, especially the main result figure
  • existing visual style choices: palette, markers, symbols, font sizes, line widths, figure sizing, and table conventions
  • method claim, assumptions, theorem statements, or system design
  • experiment setup, baselines, ablations, datasets, metrics, qualitative examples
  • limitations, broader impact, ethics, checklist, and appendix
Extract the paper into this working summary:
markdown
undefined
优先使用原始草稿文件和笔记，而非记忆。
查找以下内容：
  • main.tex
    、
    paper.tex
    、
    sections/*.tex
    、
    *.bib
    、
    README.md
    、
    docs/
    、
    notes/
  • 标题、摘要、引言、贡献列表
  • 图表，尤其是核心结果图
  • 现有视觉风格选择：调色板、标记、符号、字体大小、线条宽度、图表尺寸及表格惯例
  • 方法主张、假设、定理陈述或系统设计
  • 实验设置、基线模型、消融实验、数据集、指标、定性示例
  • 局限性、更广泛影响、伦理、checklist及附录
将论文提炼为以下工作摘要：
markdown
undefined

Paper Snapshot

论文快照

  • Current title:
  • Target venue:
  • Claimed contribution:
  • Core technical idea:
  • Primary evidence:
  • Strongest result:
  • Most likely reviewer concern:
  • Missing or weak evidence:
  • Current structure:

If the draft is too incomplete to rewrite, produce a structural plan and a list of missing evidence instead of pretending the paper is ready.
  • 当前标题：
  • 目标会议：
  • 宣称的贡献：
  • 核心技术思路：
  • 主要证据：
  • 最强结果：
  • 最可能的审稿人疑虑：
  • 缺失或薄弱的证据：
  • 当前结构：

若草稿过于残缺无法重写，请生成结构方案和缺失证据清单，而非假装论文已就绪。

Step 3 - Learn the Target Venue Taste

步骤3 - 了解目标会议偏好

Read 
references/conference-profiles.md
, then refine it with current evidence. When learning from multiple exemplar papers, read 
references/exemplar-analysis.md
and produce a compact style matrix before recommending a rewrite strategy.
Sources to prefer:
  • official venue call for papers and author instructions
  • accepted/oral/spotlight/best-paper lists
  • OpenReview pages, using API/exported data/user-provided PDFs when the visible page is client-rendered
  • proceedings pages and official PDFs
  • reviewer guidelines, checklist requirements, and area descriptions
When using OpenReview:
  • The group page may load notes client-side and expose only "Loading" in static HTML.
  • If direct page content is unavailable, use OpenReview API endpoints, the venue's official accepted-paper export, search results, or user-provided links/PDFs.
  • Prioritize papers with status such as oral, spotlight, notable, award, or high-review visibility, but include archetype diversity rather than only the most famous papers.
Extract patterns from exemplars:
  • paper archetype
  • title style
  • abstract shape
  • introduction move sequence
  • where the core claim appears
  • how contributions are enumerated
  • method exposition density
  • experiment narrative order
  • figure/table role
  • visual style: palette, typography, symbol conventions, figure density, caption style, and table compactness
  • limitation handling
  • reviewer concern preemption
Do not copy phrasing from exemplar papers. Distill patterns.
阅读
references/conference-profiles.md
，然后结合当前证据细化内容。 当学习多篇范例论文时，请先阅读
references/exemplar-analysis.md
，并在推荐重写策略前生成简洁的风格矩阵。
优先参考以下来源：
  • 官方会议征稿启事和作者指南
  • 已录用/口头报告/spotlight/最佳论文列表
  • OpenReview页面，当可见页面为客户端渲染时，使用API/导出数据/用户提供的PDF
  • 论文集页面和官方PDF
  • 审稿人指南、checklist要求及领域说明
使用OpenReview时：
  • 群组页面可能通过客户端加载笔记，静态HTML中仅显示"Loading"。
  • 若无法直接获取页面内容，请使用OpenReview API端点、会议官方已录用论文导出数据、搜索结果或用户提供的链接/PDF。
  • 优先选择带有口头报告、spotlight、 notable、获奖或高审稿可见性状态的论文，但需涵盖多样的范式，而非仅选择最知名的论文。
从范例中提炼模式：
  • 论文范式
  • 标题风格
  • 摘要结构
  • 引言逻辑顺序
  • 核心主张的位置
  • 贡献的列举方式
  • 方法阐述的密度
  • 实验叙事顺序
  • 图表作用
  • 视觉风格：调色板、排版、符号惯例、图表密度、标题风格及表格紧凑度
  • 局限性处理方式
  • 审稿人疑虑的预先回应
请勿复制范例论文的措辞，仅提炼模式。

Step 4 - Diagnose the Paper Archetype

步骤4 - 诊断论文范式

Read 
references/paper-archetypes.md
and classify the user's draft.
Use one primary archetype and optional secondary archetypes:
  • method
    : new algorithm, architecture, objective, training recipe, inference procedure, or theoretical method
  • empirical-study
    : systematic finding about models, data, scaling, evaluation, or failure modes
  • benchmark-dataset
    : new dataset, task, benchmark, protocol, or evaluation suite
  • theory
    : theorem-led paper with formal assumptions, guarantees, or impossibility results
  • systems
    : infrastructure, serving, training, tool, compiler, data pipeline, or large-scale engineering contribution
  • analysis
    : interpretability, diagnostic, causal, or mechanistic analysis
  • application
    : strong domain result where novelty comes from ML adaptation plus evidence in a demanding setting
Then decide whether the current venue favors this archetype and what style variant fits best.
阅读
references/paper-archetypes.md
并对用户草稿进行分类。
使用一种主范式和可选的次要范式：
  • method
    ：新算法、架构、目标函数、训练流程、推理过程或理论方法
  • empirical-study
    ：关于模型、数据、缩放、评估或失效模式的系统性发现
  • benchmark-dataset
    ：新数据集、任务、基准、协议或评估套件
  • theory
    ：以定理为主导的论文，包含正式假设、保证或不可能性结论
  • systems
    ：基础设施、服务、训练、工具、编译器、数据管道或大规模工程贡献
  • analysis
    ：可解释性、诊断性、因果性或机制性分析
  • application
    ：在严苛场景中通过ML适配获得的强领域结果，创新性源于此
然后判断当前会议是否偏好该范式，以及哪种风格变体最适合。

Step 5 - Build the Reviewer-Facing Narrative

步骤5 - 构建面向审稿人的叙事

Write a narrative diagnosis before rewriting:
markdown
undefined
重写前先撰写叙事诊断：
markdown
undefined

Venue Fit Diagnosis

会议适配诊断

  • Target venue:
  • Paper archetype:
  • Best-fit style:
  • Reviewer promise:
  • Main tension:
  • What must be obvious by the end of page 1:
  • What must be proven by experiments/theory:
  • Main figure/table role and visual style:
  • What should move to appendix:
  • Writing risks:

The "reviewer promise" is the sentence a reviewer should be able to say after reading the introduction:

> This paper matters because [problem], and it deserves acceptance because [specific contribution] is supported by [specific evidence].
  • 目标会议：
  • 论文范式：
  • 最适配风格：
  • 审稿人预期：
  • 核心矛盾：
  • 第1页末尾必须明确的内容：
  • 必须通过实验/理论证明的内容：
  • 核心图表的作用及视觉风格：
  • 应移至附录的内容：
  • 写作风险：

“审稿人预期”是审稿人读完引言后应能说出的句子：

> 这篇论文很重要，因为[问题]，它值得录用，因为[具体贡献]得到了[具体证据]的支持。

Step 6 - Produce a Section-Level Rewrite Plan

步骤6 - 生成章节级重写方案

For a standard ML conference paper, plan:
  • Title
  • Abstract
  • Introduction
  • Background or Problem Setup
  • Method / Approach / Theory / Benchmark Design
  • Experiments / Evaluation
  • Analysis / Ablations / Discussion
  • Related Work
  • Limitations / Ethics / Broader Impact
  • Appendix / Supplementary
For each section, specify:
  • section job
  • target length or page budget
  • required claims
  • required evidence
  • paragraphs to include
  • material to cut or move
  • venue-specific cautions
  • figure/table role and visual style constraints when the section depends on visual evidence
If official page limits are relevant, verify current rules and state the source.
针对标准ML会议论文，规划以下章节：
  • 标题
  • 摘要
  • 引言
  • 背景或问题设定
  • 方法/方案/理论/基准设计
  • 实验/评估
  • 分析/消融实验/讨论
  • 相关工作
  • 局限性/伦理/更广泛影响
  • 附录/补充材料
针对每个章节，明确：
  • 章节目标
  • 目标长度或页数预算
  • 必需的主张
  • 必需的证据
  • 需包含的段落
  • 需删减或移动的内容
  • 特定会议的注意事项
  • 当章节依赖视觉证据时，图表的作用及视觉风格约束
若官方页数限制相关，请核实当前规则并注明来源。

Step 7 - Produce a Paragraph-Level Blueprint

步骤7 - 生成段落级蓝图

Read 
references/paragraph-protocol.md
.
For every important paragraph, output:
markdown
undefined
阅读
references/paragraph-protocol.md
。
针对每个重要段落，输出：
markdown
undefined

[Section] P[N]

[章节] 第[N]段

  • Function:
  • Reader question answered:
  • Claim:
  • Evidence or support:
  • Opening move:
  • Closing move:
  • Keep:
  • Cut or move:
  • Risk if weak:

Use this level of detail especially for the abstract, introduction, method overview, main experiment setup, and result interpretation.
  • 功能：
  • 解答读者的问题：
  • 主张：
  • 证据或支撑：
  • 开篇方式：
  • 收尾方式：
  • 保留内容：
  • 删减或移动内容：
  • 薄弱环节的风险：

尤其针对摘要、引言、方法概述、核心实验设置及结果解读部分，需使用此层级的细节。

Step 8 - Rewrite Text

步骤8 - 重写文本

When rewriting, preserve traceability:
  • Rewrite only sections that have enough evidence.
  • Keep claims no stronger than the paper's evidence.
  • Use explicit placeholders for missing facts, such as 
    [INSERT MAIN RESULT: metric, baseline, dataset]
    .
  • Prefer concrete nouns and verbs over hype.
  • State contributions as claim plus evidence, not as vague novelty.
  • Add reviewer-facing transitions that explain why the next section exists.
For LaTeX projects, edit the smallest relevant file. Do not reorganize the entire source tree unless the user asks.
重写时需保持可追溯性：
  • 仅重写证据充足的章节。
  • 主张强度不得超过论文现有证据。
  • 对缺失的事实使用明确占位符，例如
    [INSERT MAIN RESULT: metric, baseline, dataset]
    。
  • 优先使用具体名词和动词，而非夸大表述。
  • 将贡献表述为“主张+证据”，而非模糊的创新性描述。
  • 添加面向审稿人的过渡语句，解释下一章节存在的原因。
对于LaTeX项目，编辑最小的相关文件。除非用户要求，否则不要重组整个源码树。

Step 9 - Run Quality Gates

步骤9 - 运行质量检查

Read 
references/quality-gates.md
before finalizing a plan or rewritten section.
At minimum, check:
  • venue fit: the structure and tone match the target venue rather than generic ML writing
  • archetype fit: the chosen paper shape matches the actual contribution
  • claim-evidence alignment: every major claim has evidence or is marked as missing
  • paragraph function: each important paragraph has one reviewer-facing job
  • rebuttal readiness: obvious objections are preempted in the main text or flagged
  • reproducibility and assumptions: rules, assumptions, and appendix promises are explicit
If a gate fails, revise once before returning the result. If it still fails because evidence is missing, report the blocker instead of smoothing it over.
在确定方案或完成重写章节前，阅读
references/quality-gates.md
。
至少检查以下内容：
  • 会议适配性：结构和语气符合目标会议，而非通用ML写作风格
  • 范式适配性：所选论文结构匹配实际贡献
  • 主张-证据一致性：每个主要主张都有证据支持，或标记为缺失
  • 段落功能：每个重要段落都有明确的面向审稿人的任务
  • Rebuttal就绪性：明显的异议已在正文中预先回应或标记
  • 可复现性与假设：规则、假设及附录承诺明确
若某一检查未通过，请先修订一次再返回结果。若因证据缺失仍未通过，请报告障碍，而非掩盖问题。

Step 10 - Update Writing Memory

步骤10 - 更新写作记忆

Read 
references/memory-model.md
.
When the task involved studying target-venue exemplars or the user explicitly wants knowledge to persist:
  1. create or update project-local memory under 
    .agent/conference-writing/
  2. add venue observations to 
    .agent/conference-writing/venues/<venue>.md
  3. add exemplar notes to 
    .agent/conference-writing/exemplars/<venue>-<year>.md
  4. add reusable decisions to 
    .agent/conference-writing/project-style.md
Memory entries must separate:
  • observed patterns from specific papers
  • inferences about venue taste
  • decisions for the current paper
  • unresolved questions
Do not store copyrighted paper text beyond short titles, bibliographic metadata, and brief paraphrased notes.
If the project uses 
research-project-memory
, also update:
  • memory/claim-board.md
    : revised claim wording, paper locations, and unsupported claims to cut or weaken
  • memory/evidence-board.md
    : figure/table roles and evidence items cited by rewritten sections
  • memory/risk-board.md
    : writing, positioning, venue-fit, and evidence-gap risks
  • memory/action-board.md
    : missing experiments, citation checks, section rewrites, or figure updates exposed by writing
  • paper/.agent/paper-status.md
    : section map, paragraph-level decisions, and stale figures/tables
  • paper/.agent/visual-style.md
    or 
    .agent/conference-writing/project-style.md
    : venue-facing figure style decisions, including palette, symbol, typography, and table conventions
Do not strengthen claims in memory beyond the evidence available in the paper.
阅读
references/memory-model.md
。
当任务涉及研究目标会议范例或用户明确希望保留知识时：
  1. 在
    .agent/conference-writing/
    下创建或更新项目本地记忆
  2. 将会议观察结果添加至
    .agent/conference-writing/venues/<venue>.md
  3. 将范例笔记添加至
    .agent/conference-writing/exemplars/<venue>-<year>.md
  4. 将可复用决策添加至
    .agent/conference-writing/project-style.md
记忆条目必须区分：
  • 从特定论文中观察到的模式
  • 关于会议偏好的推断
  • 当前论文的决策
  • 未解决的问题
除短标题、文献元数据和简短转述笔记外，请勿存储受版权保护的论文文本。
若项目使用
research-project-memory
，还需更新：
  • memory/claim-board.md
    ：修订后的主张措辞、论文位置及需删减或弱化的无依据主张
  • memory/evidence-board.md
    ：图表作用及重写章节引用的证据项
  • memory/risk-board.md
    ：写作、定位、会议适配及证据缺口风险
  • memory/action-board.md
    ：写作过程中发现的缺失实验、引用检查、章节重写或图表更新需求
  • paper/.agent/paper-status.md
    ：章节地图、段落级决策及过时图表
  • paper/.agent/visual-style.md
    或
    .agent/conference-writing/project-style.md
    ：面向会议的图表风格决策，包括调色板、符号、排版及表格惯例
记忆中的主张强度不得超过论文现有证据。

Output Formats

输出格式

Diagnosis Only

仅诊断报告

markdown
undefined
markdown
undefined

Conference Writing Diagnosis

会议写作诊断

Paper Snapshot

论文快照

Venue Fit Diagnosis

会议适配诊断

Archetype Classification

范式分类

Main Reviewer Risks

主要审稿人风险

Recommended Rewrite Strategy

推荐重写策略

undefined
undefined

Rewrite Plan

重写方案

markdown
undefined
markdown
undefined

[Venue] Rewrite Plan

[会议名称]重写方案

Paper Snapshot

论文快照

Venue Pattern Summary

会议模式总结

Narrative Strategy

叙事策略

Section-Level Plan

章节级方案

Paragraph-Level Blueprint

段落级蓝图

Evidence Gaps

证据缺口

Appendix / Page-Budget Plan

附录/页数预算方案

Memory Updates

记忆更新

undefined
undefined

Rewritten Section

重写章节

markdown
undefined
markdown
undefined

Rewritten [Section Name]

重写后的[章节名称]

Before/After Intent

改写前后意图

Draft Text

草稿文本

Remaining Placeholders

剩余占位符

Why This Fits [Venue]

适配[会议名称]的原因

undefined
undefined

Final Sanity Check

最终 sanity检查

Before finalizing:
  • the target conference and year are explicit
  • current venue rules were verified if page limits or mandatory sections matter
  • exemplar-derived observations are paraphrased and attributed at a high level
  • the user's paper archetype is stated
  • every paragraph in the blueprint has a clear reviewer-facing function
  • the quality gates have been run and any failing gate is reported
  • unsupported claims are marked as gaps
  • memory updates, if any, are written to project-local files
完成前请确认：
  • 目标会议及年份明确
  • 若页数限制或强制章节相关，已核实当前会议规则
  • 从范例中提炼的观察结果已转述并进行了高层级归因
  • 用户论文的范式已明确
  • 蓝图中的每个段落都有清晰的面向审稿人的功能
  • 已完成质量检查，且所有未通过的检查已报告
  • 无依据主张已标记为缺口
  • 若有记忆更新，已写入项目本地文件