agentsclimarketplace

Proofread

Skill tance-mang/chinese-webnovel-skills/skills/proofread

中文网文错别字校对与敏感词/违禁内容过审检查。当用户要查错别字(的地得/在再/做作等)、查敏感词违禁词、投稿前过审自检、问"这段会不会被屏蔽/被标红/过不了审"、或要"像番茄那样标出错字并给一键替换建议"时使用。逐句扫描,给出"原词→建议词"对照供作者一键替换。From its SKILL.md

Install
npx -y skills add tance-mang/chinese-webnovel-skills --skill proofread

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

4.4 KB, ~1.8k tokens by cl100k_base, as published. Nobody here has run it

错别字校对与过审检查(proofread)

模仿番茄/起点提交后台的"标红 + 一键替换"体验:扫一遍正文,揪出错别字可能触发平台审核的敏感/违禁内容,每处给"原句 → 改后"对照,作者照着替换即可。

何时用

  • 查错别字(的/地/得、在/再、做/作、即/既…)
  • 投稿/发布前过审自检:"这段会不会被标红/屏蔽/过不了"
  • 要"像番茄那样"标出问题词 + 给替换建议

⚠️ 能力边界(务必如实告诉作者)

  • 错别字:按句子意思判断同音/形近误用,准确度高,可放心用。
  • 敏感词/违禁:各平台的确切屏蔽词库是内部机密且时常变动,本技能给的是基于公开审核常识的启发式排查(覆盖政治/色情/暴恐/赌毒/迷信/真实品牌人名等大类 + 常见踩雷),是"安全网"不是"保证命中番茄后台的每一个词"。最终以平台实际审核为准;拿不准的词建议作者自行替换或上报平台。
  • 需要核对"某平台当下的公开内容规范"时,可联网查(同 trends 的联网能力)。

工作流

第 1 步:错别字扫描

逐句按意思判断,常见类型(详见 references/sensitive-words.md 错别字高频对):

  • 同音误用:的/地/得、在/再、做/作、需/须、即/既、账/帐、应该/因该
  • 形近误用:辩/辨/辫、急燥(→躁)、迫不急待(→及)
  • 多字/漏字/重字:“他他走了”“走了进去去”
  • 成语写错:“迫不及待”“川流不息”“黄粱一梦”
  • 标点.../。。。→“……”、并列误用句号→顿号(与 deslop 一致)

第 2 步:敏感/违禁扫描(按大类排查,见 references/sensitive-words.md

  • 涉政/时政、领导人/真实政治人物
  • 色情低俗、过度性描写、擦边
  • 暴力血腥、自杀自残的具体方法、教唆/传授犯罪方法
  • 赌博、毒品、违禁药物的具体操作
  • 封建迷信/宗教渲染(区别于"修真设定",看是否引导现实迷信/破坏宗教政策/邪教)
  • 真实品牌/商标/药名/真实人名地名(易被判广告或侵权)
  • 辱骂脏话、地域/群体歧视
  • 价值观红线、未成年相关红线(含校园未成年暴力)

第 2.5 步:番茄"发布"平台特有规范(长篇连载,易忽略,见 sensitive-words.md 二·补)

  • 情节无关内容:求过审/请假/小剧场/作者吐槽 → 移出正文,发"作者有话说"
  • 写作不规范:拼音/字母替代敏感词、乱码、不分段、无标点 → 标出改规范
  • 引导关注:正文里的 QQ群/微信/微博/公众号/"加我进群" → 全删

第 3 步:输出"标红 + 一键替换"清单

【校对报告 · 第 N 章 / 选段】
字数:__  错别字:__ 处  敏感/违禁疑似:__ 处

—— 错别字 ——
① 第_段“他在做什么”→“他在作什么”?… 按"作为"应为“作”|原句:……
   建议替换:做 → 作
② 第_段“迫不急待”→“迫不及待”
   建议替换:急 → 及

—— 敏感/违禁(疑似,按平台审核常识)——
① 第_段“<触发词/句>”类别:涉政|风险:高
   建议改写:……(给 1–2 个安全替换)
② 第_段“<真实品牌名>”类别:商标/广告嫌疑|风险:中
   建议替换:用虚构名“XX”

【优先处理】高风险敏感项先改;错别字可批量替换。

原则

  • 每处都给"原词→建议词",让作者像点番茄的"一键替换"一样照改。
  • 错别字按句子意思判断,不是机械匹配("做/作"看语境)。
  • 敏感项标类别 + 风险等级 + 替换,并说明这是启发式、以平台为准。
  • 不误伤:修真/玄幻的"法术/境界"、合理剧情冲突不算违禁;只标真正可能触审的。

配合

  • 投稿三章签约体检(节奏/钩子/字数门槛)→ submission
  • 文字手感/AI 味 → deslopaidetect
  • 当下平台公开规范联网核对 → trends

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most editing proofreading skills give in ~1.8k tokens

Counted across 116 of the 148 authors here whose files we hold, read 2026-09-06

  • Ground factual claims in real source materialin 5 of 116, across 2 files
  • Ask for the text if none suppliedin 5 of 116, across 2 files
  • Fix grammar, spelling, and punctuation errorsin 5 of 116
  • Prefix the first line with 🥷 inlinein 4 of 116, across 1 file
  • Confirm the audience before editing when unclearin 4 of 116, across 1 file
  • Detect language from the text being editedin 4 of 116, across 1 file
  • Read the loaded reference file before editingin 4 of 116, across 1 file
  • Preserve meaning when removing AI patternsin 4 of 116, across 1 file
  • Edit in place without restructuringin 4 of 116, across 1 file
  • Match the target audience's registerin 4 of 116, across 1 file
  • Read the latest release as a style referencein 4 of 116, across 1 file
  • Iterate copy against real product screenshots until alignedin 4 of 116, across 1 file

Said here and by no other author read

  • Scan the text sentence by sentence for typos.
  • Judge typos by sentence meaning, not mechanical matching.
  • Screen for sensitive content across the listed categories.
  • Flag real brands, trademarks, and real names.
  • Move plot-irrelevant content into author's notes.
  • Delete contact and group links from the body.

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.