ra-video-title
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
Chinese视频标题 Skill
Video Title Skill
核心原则
Core Principle
先定主题,再起标题。视频标题必须服务这条视频的核心命题,不能只套“AI 工具”“翻车”“原因找到了”这类泛钩子。
默认替代 处理视频标题。 仍可用于通用小红书图文标题或纯公式浏览;视频成片、视频洗稿、对标标题、两段式抖音标题一律走本 Skill。
dbs-xhs-titledbs-xhs-titleDetermine the topic first, then create the title. A video title must serve the video’s core proposition; it must not rely solely on generic hooks such as “AI Tools,” “It Went Wrong,” or “Here’s the Reason.”
By default, this Skill replaces for video titles. can still be used for general Xiaohongshu image-and-text titles or formula-based title browsing; final videos, rewritten videos, benchmark titles, and two-part Douyin titles should all go through this Skill.
dbs-xhs-titledbs-xhs-title两段式流程中的位置
Position in the Two-Stage Workflow
在「洗稿(Claude)→ 待制作队列 → 制作(Codex)」两段式流程里,本 skill 在洗稿阶段运行,输入是刚洗好的最终脚本(主题最清楚的时刻):
- 只产出候选:8-12 个两段式候选 + Top 3 推荐理由,写入交接稿「标题候选」一节
- 不在洗稿阶段定稿:最终标题由用户发布时人工从候选里选
- 制作端与发布标题无关:不起标题、不应用标题、不生成
视频标题.md
In the two-stage workflow of Rewriting (Claude) → Production Queue → Production (Codex), this Skill runs during the rewriting stage, with the freshly rewritten final script as its input, which is when the topic is clearest:
- Produce candidates only: 8-12 two-part candidates + reasons for the Top 3, written into the section of the handoff draft
标题候选 - Do not finalize a title during rewriting: the user manually selects the final title from the candidates when publishing
- The production stage is unrelated to publishing titles: do not create titles, apply titles, or generate
视频标题.md
工作流
Workflow
-
锁定视频主题
- 优先读用户给出的最终脚本、视频制作简报、、项目说明或成片目录上下文。
narration_segments.json - 输出一句主题判断,格式为:
这条视频讲的是:... - 标题候选必须能回扣这句主题。
- 优先读用户给出的最终脚本、视频制作简报、
-
拆对标标题
- 如果用户给对标标题,提炼共性:结构、前半句钩子、后半句补充、情绪、专业词、结果承诺。
- 不复制对标的具体表达,不把别人的账号人设写进标题。
- 默认重点检查“两段式”:前半句抛场景/冲突/问题,后半句给结果/解释/悬念。
-
生成候选
- 生成 8-12 个候选。
- 默认用两段式,常见连接符:、
,、?、!。: - 每条候选都要同时满足:
- 命中视频主题,不是泛 AI 标题;
- 前半句有钩子;
- 后半句补充方向或悬念;
- 口语化,适合短视频平台;
- 不把视频答案全说完;
- 资讯类主题(新论文/新产品/新事件):前半句直接给事实主体(谁+发了什么),不用比喻/拟人当首要钩子,比喻最多出现在后半句补充里。
-
筛选与推荐
- 给出 Top 3,并说明一句理由。
- 如果用户要求“你选定”“直接写入”“替我选”,选 1 个最终标题。
- 如果用户明确说“我来选”,只给候选并等待编号。
-
写入文件
- 两段式洗稿流程中:写入目标只有交接稿「标题候选」一节(8-12 候选 + Top 3 推荐),不写 frontmatter、不生成 ,最终由用户人工选定。
视频标题.md - 其他场景当用户要求写入时,在最终 MP4 所在的 目录新建或更新
成片/。视频标题.md - 文件第一行只写最终标题。
- 只有在用户要求保留或候选对后续有价值时,才追加 小节。
## 候选标题 - 使用 编辑
apply_patch;不要改动 MP4 或其他成片文件。视频标题.md
- 两段式洗稿流程中:写入目标只有交接稿「标题候选」一节(8-12 候选 + Top 3 推荐),不写 frontmatter、不生成
-
Lock in the video topic
- Prioritize reading the user-provided final script, video production brief, , project documentation, or context from the final-video directory.
narration_segments.json - Output a one-sentence topic assessment in this format:
This video is about: ... - Every title candidate must connect back to this topic.
- Prioritize reading the user-provided final script, video production brief,
-
Deconstruct benchmark titles
- If the user provides benchmark titles, extract their common patterns: structure, first-half hook, second-half supporting detail, emotional tone, professional terminology, and promised result.
- Do not copy the benchmark’s exact wording or incorporate another account’s persona into the title.
- By default, focus on the two-part structure: the first half presents a scenario, conflict, or question, while the second half delivers a result, explanation, or suspense.
-
Generate candidates
- Generate 8-12 candidates.
- Use a two-part structure by default, with common connectors such as ,
,,?, and!.: - Every candidate must meet all of the following requirements:
- Match the video topic rather than using a generic AI title;
- Include a hook in the first half;
- Add an angle or suspense in the second half;
- Use a conversational style suitable for short-video platforms;
- Avoid revealing the video’s entire answer;
- For news-oriented topics such as a new paper, product, or event, directly identify the factual subject in the first half using who + what was released, rather than using a metaphor or personification as the primary hook. A metaphor may appear at most in the second-half supplement.
-
Filter and recommend
- Present the Top 3 and explain each with one sentence.
- If the user asks you to “select it,” “write it directly,” or “choose it for me,” select one final title.
- If the user explicitly says “I’ll choose,” provide only the candidates and wait for a number.
-
Write to the file
- In the two-stage rewriting workflow, the only target is the section of the handoff draft, containing 8-12 candidates + Top 3 recommendations. Do not add frontmatter or generate
标题候选; the user makes the final selection manually.视频标题.md - In other scenarios, when the user requests it, create or update in the
视频标题.mddirectory containing the final MP4.成片/ - The first line of the file must contain only the final title.
- Only append a section if the user requests it or the candidates have value for later use.
## 候选标题 - Use to edit
apply_patch; do not modify the MP4 or any other final-video files.视频标题.md
- In the two-stage rewriting workflow, the only target is the
标题评分
Title Scoring
每个候选用下面 5 项心里过一遍:
- 主题贴合:离视频主题越近越好;能套到任何 AI 视频上的标题扣分。
- 两段式张力:前后两段要有递进,不是简单重复。
- 点击理由:用户为什么要点开?是怕落后、怕踩坑、想省事、想变强,还是想看争议?
- 信息差:标题要留缺口,不要把结论说满。
- 口语感:像抖音标题,不像论文标题或课程标题。
Mentally assess every candidate against the following five criteria:
- Topic Fit: The closer it is to the video topic, the better. Deduct points if the title could apply to any AI video.
- Two-Part Tension: The two parts should build on each other rather than simply repeat the same idea.
- Click Motivation: Why should the user click? Are they afraid of falling behind, concerned about potential pitfalls, looking for convenience or improvement, or interested in a controversy?
- Information Gap: Leave a gap in the information instead of revealing the full conclusion.
- Conversational Feel: It should sound like a Douyin title, not a paper title or course title.
禁用项
Prohibited Practices
- 不要在没读主题的情况下直接起标题。
- 资讯类视频不要用比喻/拟人起手的标题(如「跟AI说别想金门大桥,它满脑子都是金门大桥」);事实主体(如「Anthropic 新论文」「Anthropic 造出读心工具」)必须出现在前半句。
- 不要只围绕工具名起标题,例如只写 ,除非视频主题就是该工具故障。
Claude Code 翻车原因 - 不要让标题偏离视频,比如主题是 ,标题却变成普通
Model × Harness。AI 工作流提效 - 不要为了追热点滥用“接管”“破甲”“零报错”;这些词只有贴合主题时才用。
- 不要默认把候选标题全部写入文件;文件优先保持干净。
- Do not create a title before understanding the topic.
- Do not open a news-style video title with a metaphor or personification, such as “Tell AI not to think about the Golden Gate Bridge, and it can’t stop thinking about the Golden Gate Bridge.” The factual subject, such as “Anthropic’s New Paper” or “Anthropic Built a Mind-Reading Tool,” must appear in the first half.
- Do not build the title solely around a tool name, such as , unless a failure involving that tool is specifically the video’s topic.
Why Claude Code Failed - Do not let the title drift away from the video. If the topic is , the title must not become a generic
Model × Harnesstitle.AI Workflow Efficiency - Do not overuse trend-driven hooks such as “Take Over,” “Breakthrough,” or “Zero Errors”; use them only when they fit the topic.
- Do not write every candidate into the file by default; keep the file clean whenever possible.
参考
References
两段式标题模式和本工作区沉淀的样例见 。需要生成候选或解释对标共性时读取它。
references/two-part-video-title.mdThe two-part title patterns and examples accumulated in this workspace are available in . Read it when generating candidates or explaining common benchmark-title patterns.
references/two-part-video-title.md