1. 收到审稿意见后先别急着写回复把“情绪”与“证据”分开投稿英文期刊被审稿人质疑是几乎每个研究者都会遇到的场景。你打开邮件看到 Referee A 说“XRD of these calcined materials should be provided”Referee B 又说“This point should be clarified in the main text”还有一条直接质疑你的统计方法。第一反应往往是委屈我明明做了实验为什么审稿人看不懂但真正决定稿件命运的不是你的委屈而是回复信里有没有逐条、可核查、礼貌且逻辑自洽的应答。我试过把审稿意见直接丢给通用对话模型让它“帮我写回复”结果生成的文字客气得发虚却没有落到我的实验数据上。后来我把流程拆成三步先分类意见再逐条构造“审稿人原话—我的回应—稿件改动位置”的三元组最后用 TaoToken 统一调用模型做润色与逻辑检查。这样做的核心检索词就是“英文期刊审稿意见逐条回复模板”它能帮你在 48 小时内交出一封让编辑看得懂、让审稿人挑不出硬伤的回复信。适合谁第一次投 SCI 的硕博生、需要返修但英文写作吃力的研究者、以及被“major revision”压得不知道从哪下笔的人。你不需要精通学术英语只需要一套可复制的结构加上一个能稳定调用多个模型的 API 入口。下面我把这套流程完整拆开包括回复信骨架、意见分类表、礼貌措辞清单以及如何用 TaoToken 把润色和逻辑检查串起来。先明确一个判断审稿人质疑分两类。一类是“事实性缺失”比如没给 XRD、没写样品尺寸、没做统计检验另一类是“表达性分歧”比如他觉得你的措辞不准确、讨论不够深入。前者必须补数据或补说明后者可以据理力争但要用证据说话。把这两类混在一起回复就会出现“该补的没补、该争的没争”的尴尬。所以第一步不是写英文而是把每条意见贴进一张表标上类型、优先级、是否需要新实验、对应稿件行号。这张表就是你后续所有工作的总控台。2. TaoToken 前置一个 API 入口统一调用模型做润色与逻辑检查回复审稿意见时你需要的模型能力其实有三种第一把中文思路转成学术英文第二检查回复与稿件改动是否一致第三模拟审稿人视角挑逻辑漏洞。如果每种能力都去不同平台开账号、配 Key光是环境切换就够烦。TaoToken 的价值在于用一个 API 入口统一调用模型你只需要维护一套 Base URL 和 Key就能在脚本里切换模型完成不同任务。TaoToken 官网地址是 https://taotoken.net/?utm_sourcetaotoken_aicg_blog_endutm_mediumcsdnutm_campaignrewriteutm_content API 请求地址是 https://taotoken.net/api 。注意 API 地址不带 UTM 参数直接用于代码里的 base_url。你需要先到控制台创建 API Key控制台入口在 https://taotoken.net/console?utm_sourcetaotoken_aicg_blog_endutm_contentconsoleutm_campaignrewrite Key 管理页面在 https://taotoken.net/api-keys?utm_sourcetaotoken_aicg_blog_endutm_contentapi-keysutm_campaignrewrite 。如果你只是想先试试模型对话效果可以打开 https://taotoken.net/models?utm_sourcetaotoken_aicg_blog_endutm_contentmodelsutm_campaignrewrite 直接对话确认模型对学术英文的润色风格符合预期。这里要强调一个原则TaoToken 是模型调用入口不是替你写论文的工具。你的实验数据、稿件行号、审稿人原话必须由你自己提供。模型负责的是语言组织和逻辑一致性检查。比如你把“Comment 1: The titania material formed after calcining at 450 oC is not characterized”和你的补充说明“XRD of the calcined TiO2 film was given in Supporting Information (Figure S1)”一起发给模型让它生成一段礼貌回应并检查你是否在正文里真的引用了 Figure S1。这种“生成核查”的组合才是回复审稿意见时最稳的用法。如果你长期要做编码类或 Agent 类任务比如批量处理多条审稿意见、自动生成对照表可以了解 Coding Planhttps://taotoken.net/coding-plan?utm_sourcetaotoken_aicg_blog_endutm_contentcoding-planutm_campaignrewrite 。接入文档在 https://taotoken.net/doc?utm_sourcetaotoken_aicg_blog_endutm_contentdocutm_campaignrewrite 里面写了不同语言调用示例。对于 Claude Code 用户Anthropic 兼容接入说明在 https://taotoken.net/ClaudeCodeAnthropic?utm_sourcetaotoken_aicg_blog_endutm_contentClaudeCodeAnthropicutm_campaignrewrite 。这些入口的共同点是你只配一次 Base URL 和 Key后续换模型只改 Model ID。3. 可复制配置把回复信骨架、分类表和 API 调用写成文件这一节给你三份可直接复制的配置。第一份是回复信 Markdown 骨架第二份是审稿意见分类表 JSON第三份是调用 TaoToken 做润色的 Python 配置片段。路径和原文保持一致你按自己的项目目录调整即可。先建一个工作目录比如reviewer-reply/里面放三个文件response-letter.md、comments.json、polish.py。回复信骨架如下注意每一条都保留“审稿人原话”和“作者回应”两个块编辑一眼就能对照# Response to Reviewers Manuscript number: BXXXXXK Title: XXXX Correspondence Author: XXX Dear Dr. Fay Riordan: Thank you very much for your attention and the referees evaluation and comments on our paper BXXXXK. We have revised the manuscript according to your kind advice and the referees detailed suggestions. Enclosed please find the responses to the referees. ## Response to Referee A ### Comment 1 The titania material formed after calcining at 450 oC is not characterized. XRD of these calcined materials should be provided to understand the crystallinity and phase. **Response:** Thanks for the referees kind suggestion. According to his/her advice, X-ray diffractometry (XRD) of the calcined TiO2 film was given in Supporting Information (Figure S1) in this revised version. It illustrates that the hydrothermally synthesized TiO2 materials after calcining at 450 oC are entirely anatase, confirmed by the X-ray diffractometer with Cu Kα radiation (Rigaku D/max-2500). **Change in manuscript:** Supporting Information Figure S1 added; main text Line 165-168, page 2. ### Comment 2 The authors must state the mechanical strength of these materials after the removal of PS by calcination. **Response:** Thanks for the referees suggestion. By a scotch tape peel test, the TiO2 film cannot be stripped from the conducting glass substrate, which indicates that the mechanical strength of the as-prepared composite film is strong enough for the fabrication of solar cell devices. **Change in manuscript:** Line 165-168, page 2. ## Response to Referee B ### Comment 1 The microtube structure with the size of 500-800 nm cannot only scatter the visible light in the region of 500-800 nm, but also can scatter more efficiently the visible light in the region below 500 nm, and can scatter near infrared light. This point should be clarified in the main text. **Response:** Thanks for the referees kind advice. Just like what the referee said, the microtube network structure can scatter not only visible light but also near infrared light. We added this point in the revised manuscript. **Change in manuscript:** Line 194-205, page 2-3.第二份是意见分类表用 JSON 存方便脚本读取。字段包括id、referee、type、need_experiment、manuscript_location、status。类型我建议只分四类data_missing缺数据、clarification需澄清、wording措辞、disagree需据理力争。这样你在回复时不会把“补实验”和“改措辞”混为一谈{ manuscript: BXXXXXK, comments: [ { id: A1, referee: A, type: data_missing, need_experiment: false, manuscript_location: SI Figure S1; Line 165-168, p2, status: done }, { id: A2, referee: A, type: clarification, need_experiment: false, manuscript_location: Line 165-168, p2, status: done }, { id: B1, referee: B, type: clarification, need_experiment: false, manuscript_location: Line 194-205, p2-3, status: done }, { id: B2, referee: B, type: wording, need_experiment: false, manuscript_location: Line 217, p3, status: done } ] }第三份是调用 TaoToken 做润色的 Python 片段。这里用 OpenAI 兼容方式base_url填https://taotoken.net/apiapi_key从环境变量读Model ID 按你控制台里可用的填。注意三件套必须齐全Base URL、Key、Model ID缺一个都会报 401 或 model not foundimport os from openai import OpenAI client OpenAI( base_urlhttps://taotoken.net/api, api_keyos.environ[TAOTOKEN_API_KEY], ) def polish_response(comment, draft_response, manuscript_location): prompt fYou are helping a researcher reply to a journal reviewer. Reviewer comment: {comment} Draft response: {draft_response} Manuscript location: {manuscript_location} Please: 1. Rewrite the response in polite academic English. 2. Check whether the response clearly states what was changed and where. 3. Point out any logical gap between the comment and the response. Return the polished response and a short checklist. resp client.chat.completions.create( modelyour-model-id, messages[{role: user, content: prompt}], temperature0.3, ) return resp.choices[0].message.content if __name__ __main__: print(polish_response( commentThe titania material formed after calcining at 450 oC is not characterized., draft_responseWe added XRD in Supporting Information Figure S1., manuscript_locationSI Figure S1; Line 165-168, page 2, ))运行前先export TAOTOKEN_API_KEY你的Key。如果你在 Windows PowerShell 里用$env:TAOTOKEN_API_KEY你的Key。这段脚本的输出会给你润色后的英文和一份检查清单你把它贴回response-letter.md对应位置即可。4. 验证请求与成功结果用一条真实审稿意见跑通全流程配置写好后不要直接批量处理所有意见。先用一条真实意见做端到端验证确认 API 能通、模型输出可用、你的分类表字段能对上。我拿 Referee B 的第 2 条意见做例子原话是“They described the simulated sunlight as one-simulating-sunlight condition (AM1.5, 100 mW cm-2). But in my opinion, it would be better to use the phrase like AM 1.5 simulated solar light (100 mW cm-2).”这是一条典型的措辞类意见不需要补实验只需要改正文并说明行号。把这条意见和你的草稿回应一起传给模型。草稿可以写得很朴素“Thanks for the suggestion. We changed the phrase to AM 1.5 simulated solar light (100 mW cm-2) in Line 217, Page 3.”模型返回的润色版本通常会加上“We sincerely appreciate the reviewers careful reading”之类的礼貌开头并提醒你确认 Line 217 是否真的是修改后的行号。这一步的验证点是模型有没有编造行号如果它擅自写了“Line 220”而你的稿件实际是 217就说明你需要把“行号以我提供的为准”写进 prompt。成功结果长这样终端打印出润色后的英文段落末尾附一个 checklist比如“Response states the exact phrase change”“Manuscript location provided”“No new experiment claimed”。你对照 checklist 逐项打勾再把英文贴进回复信。接着用同样的方式跑 Referee A 的第 1 条这条是数据缺失类模型应该提醒你“Supporting Information Figure S1 must actually exist in the submission package”。如果它没提醒你就在 prompt 里加一句“If the response mentions a figure or table, verify that it is listed in the manuscript location field.”验证通过后你可以写一个循环把comments.json里的每条意见依次传给模型生成一份完整的回复信草稿。但注意模型输出必须逐条人工核对尤其是涉及实验数据、统计方法、文献引用的部分。审稿人推荐的文献一定要引用并讨论透彻这一点模型帮不了你只能你自己去读。我踩过的坑是有一次模型把审稿人推荐的文献编号写错了幸好我在提交前逐条对了参考文献列表。所以批量生成后务必做一次“意见—回应—稿件位置”三列对照检查。5. 本篇常见错排查401、local proxy failed、reading choices、OAuth 报错怎么处理接入 TaoToken 做回复信润色时最常见的报错有四类。第一类是 401 Unauthorized通常是你 Key 没设对或环境变量没生效。检查echo $TAOTOKEN_API_KEY是否有输出以及 Key 是否复制完整。注意不要把 Key 写进代码里提交到 Git用环境变量或.env文件。第二类是local proxy failed这通常出现在你本机有网络代理设置但代理未启动或端口不对。处理方式是检查你的 HTTP_PROXY/HTTPS_PROXY 环境变量如果不需要代理就清空它们让请求直连https://taotoken.net/api。第三类是reading choices相关报错比如KeyError: choices或list index out of range。这往往是因为模型返回了错误信息而不是正常 completion你的代码却直接读resp.choices[0]。正确做法是先判断resp结构打印完整返回体。常见原因是 Model ID 写错或者请求参数里messages格式不对。第四类是 OAuth 相关报错如果你用 Claude Code 或某些 CLI 工具接入可能会遇到 OAuth token 过期或 scope 不足。这时不要反复重试直接到控制台重新生成 Key并确认你用的是 API Key 而不是 OAuth token。Claude Code 的接入方式参考 https://taotoken.net/ClaudeCodeAnthropic?utm_sourcetaotoken_aicg_blog_endutm_contentClaudeCodeAnthropicutm_campaignrewrite 里面区分了不同认证方式。除了 API 报错回复信本身也有常见错误。比如审稿人问“What is the sample size for all tests performed?”你回复“We have added the sample size”却没写具体数值这等于没回答。再比如审稿人说“Figure 7. There is no scientific evidence presented in the TEM figure to convince this reviewer of sub-jets.”你如果只回“Necessary change has been made”而不说明改了什么、删了哪句、补了哪张图审稿人会觉得你在敷衍。正确做法是引用修改后的句子并给出行号。还有一条高频错误把“percent of migration”写成“percent inhibition of migration”的反例。审稿人指出“Migration of CB CD34 was reduced to 73.3%”应改为“reduced by 73.3%”这种数学表述错误必须逐字改并在回复里明确写出修改前后。如果你在回复里需要引用审稿人推荐的文献记得在回复信中写“We have cited Reference [X] and discussed it in the revised manuscript (Line XXX, Page XX)”。不要只在正文里加引用而不在回复信里说明编辑和审稿人不会去逐页找。最后所有行号和页码必须以修改后的稿件为准。如果你在回复信里写“Line 165-168, page 2”提交前一定打开修改稿确认这个位置确实是新增内容。模型可以帮你检查一致性但最终确认必须由你完成。6. 语义一致 CTA把润色、逻辑检查和逐条验证串成一条流水线回复审稿意见的完整流水线是分类意见 → 写中文草稿 → 用 TaoToken 润色成学术英文 → 检查回复与稿件位置一致 → 逐条对照审稿意见验证。这条流水线里模型对话适合快速试润色风格API Keys 适合把流程脚本化接入文档适合查不同语言的调用细节。如果你只是偶尔回一封审稿信直接用模型对话就够了https://taotoken.net/models?utm_sourcetaotoken_aicg_blog_endutm_contentmodelsutm_campaignrewrite 。如果你要批量处理多条意见、生成对照表或者把回复信生成做成可复用脚本就去 API Keys 页面创建 Keyhttps://taotoken.net/api-keys?utm_sourcetaotoken_aicg_blog_endutm_contentapi-keysutm_campaignrewrite 然后对照接入文档写代码https://taotoken.net/doc?utm_sourcetaotoken_aicg_blog_endutm_contentdocutm_campaignrewrite 。长期做研究写作或 Agent 工作流的人可以看 Coding Planhttps://taotoken.net/coding-plan?utm_sourcetaotoken_aicg_blog_endutm_contentcoding-planutm_campaignrewrite 。它的意义不是替代你的学术判断而是让你在返修季不用反复折腾环境。最后给你一个实用技巧把最终回复信导出为 PDF 前用“查找”功能搜索每个Comment编号确认没有漏答再搜索每个Line行号确认没有写错。这两步做完再提交。