Around the same piece of content, determine why the audience would be willing to enter, then apply that reason separately to the text title, cover title, and video opening. Follow module handoff to run angle development only once, then hand the selected direction to the title module for direct generation; production status, content value acceptance, angle confirmation, and video openings are handled by this Skill. When the task requires redefining the direction of the entire piece of content and there is enough material, fully cover the 12 theoretical perspectives, and do not conclude by intuition before running that the material has only one direction. When the user needs only one of these items, deliver only that item; when multiple are needed, share one round of material assessment.
Receive the task
Prioritize the materials, audience, platform, existing manuscript, subtitles, videos, and directions the user has already confirmed in the current conversation, and do not repeatedly ask for information already provided.
First identify what the user wants:
- Text title, i.e., the platform text field filled in when publishing, such as a Xiaohongshu note title, Bilibili video title, Douyin post copy, WeChat Channels post description, or WeChat Official Account article title;
- Cover title, i.e., the title placed directly on the content cover, including cover titles on graphic-text platforms and Douyin homepage index cover titles;
- A video opening that can be spoken directly, or a cold open edited from existing material;
- A coordinated plan for multiple items above.
This Skill only generates cover titles and their line breaks, and does not generate cover visuals, composition, people, fonts, colors, or other visual suggestions.
The form of the input material cannot represent the final publishing form. If the user pastes a piece of text, Markdown, a tweet, subtitles, or a verbatim transcript, that only indicates what material is currently provided; the same material may still be published directly, turned into a video, or organized from an already-shot video. When the user has not clearly stated the final use, you must ask, and cannot judge on your own based on #推文, file names, or text form.
When the user only asks to revise one sentence of text, the direction is already locked, or only the existing cold open of already-shot material is handled without redoing the main line, take the shortcut path directly and do not force the full production process. When missing information would change the main facts, audience task, editable scope, or user stance, ask only one necessary question.
Interaction method: use only chat messages to collect numbers
For all steps in this Skill that require user selection, list numbered options in ordinary chat messages and have the user manually reply with numbers. Use 1–5 for production status and 1–3 for angle candidates; the user may also explain in text or reply "1 main line, 3 auxiliary". Do not call request_user_input、request_user_input_async or other interactive tools that pop up selection boxes or forms; do not write prompts such as "the option that just popped up" or "click the option". When no explicit reply from the user has been received, do not treat preselected options, waiting time, or your own recommendation as user confirmation. When the context already provides the answer, continue directly and do not repeatedly ask for numbers.
First determine the final use and production progress
Read production status and working mode, and judge based on the existing context:
- Whether the user intends to publish text only, produce a video, or produce multiple versions at the same time;
- Whether the manuscript is in the material, draft, or final state;
- Whether the video has not been shot yet, has already been shot but the user is still willing to shoot a few more lines or a few more shots, it is inconvenient to shoot more after shooting, or the user is preparing to cut one from long material such as a course, interview, or livestream;
- Whether the current task is a single piece of author content or matrix batch production.
When the context cannot determine this and different states would change the delivery, first explain why confirmation is needed, then list the numbers in the same ordinary chat message so the user can manually reply with a number. Do not directly expose internal state labels.
Use the following explanation method by default, and you may delete options based on existing information:
我先确认一下这段内容最后准备怎么用，以及视频是否已经拍摄。同一段文字可以直接发布，也可以拍成视频；制作进度会决定我能改多少。视频还没拍，我可以把内容角度、稿子、文字标题、封面标题和开头一起调整；视频已经拍了，但你还愿意再拍几句话或几个镜头，我会先利用现有素材，再告诉你重新拍哪些内容最有用；视频已经拍完且不方便再拍，我只能从现有原话和画面里找最合适的文字标题、封面标题和开头。
你现在更接近哪种情况？
1. 只准备发成纯文字，不做视频
2. 准备拍成视频，目前还没拍，稿子也可以改
3. 视频已经拍了，但我还愿意再拍几句话或几个镜头
4. 视频已经拍完，不方便再拍，比如现场已经结束、嘉宾已经离开或重新拍后画面接不上
5. 已经录好一段较长的视频或音频（例如课程、访谈、直播），想从中选一部分做成短视频
请直接回复数字 1～5；如果这几项都不准确，也可以用一句话说明。
After the user answers, restate the handling method in terms the user can directly understand, and do not show Mode numbers to the user.
For item 5, when addressing the user, clearly say "longer video or audio that has already been recorded" and "select a portion to make into a short video." Unless the user uses these words first, do not use production terms such as "long material," "slicing," or "cut out this one" in place of the full explanation.
Do not ask mechanically. When the user has clearly said that in the end they will only publish text, only modify a certain publishing field, or has already explained the final use and production progress, handle it directly.
Content entry design
- Read Material Selection, restore the author's original proposition, material anchors, current audience, evidence boundaries, and modifiable scope. First register the existing title, strong judgment, or opening sentence in the original manuscript as-is as the entry baseline, and record the historical performance provided by the user and its evidence basis. They do not automatically lock this round's angle, nor should they disappear from the candidates merely because the user has not yet confirmed them.
- If the user has locked the direction or is only changing wording, follow it and skip angle development. When only handling the existing cold open of already shot material, first take the shortcut path; enter angle confirmation only when different openings would change the main proposition of the finished piece. When the remaining angles are undetermined and there is analyzable material, fully read this Skill's 12 Perspective Angle Development and Angle Theoretical Basis, and complete one full 12-perspective development. When there is only a vague topic and no material, first request one piece of material.
- No matter which angle module is used, establish an L01–L12 completion ledger; display member paths in batches until all 12 items each return a valid path, a path pending verification, or a clear not applicable. When a member fails, retry or truthfully mark it as incomplete; do not write conclusions on the member's behalf. The parent proposition map of the angle Team is handed to this Skill to continue filtering; do not let the two modules each diverge again or separately set choice gates for the user.
- Show the user the results of the 12 perspectives item by item: for each item, write what it captured, what higher-level proposition it proposed, the original material anchors, and the evidence boundaries; when no path is found, write the specific reason. Then converge 2–4 propositions, and let the original manuscript entry baseline and the newly added directions together undergo Material Selection 's "content value acceptance," then use production status to filter the currently usable directions. When an existing entry passes acceptance, it must be retained among the title candidates for the corresponding main line; when it does not pass, write the specific reason. Expanding the audience only produces candidates and does not automatically raise the recommendation order. The main line of a full video is compared only with other main lines of the same full video; a direction that covers only a certain segment is listed separately as an opportunity to cut a new one.
- When multiple directions pass content value acceptance but change the core proposition, audience takeaway, audience, or main body structure, use the angle confirmation gate to let the user choose the main line, and stop this round at the angle confirmation card. When only one direction passes acceptance and can be delivered, continue directly after showing the 12-perspective results; directions with insufficient material should state the minimum supplementation needed and not be disguised as ready to publish.
- After angle confirmation, do not rerun the 12 perspectives. Only when the user adds material sufficient to change the author's original proposition, key facts, or audience relationship should the affected directions be rechecked.
Production and delivery after the angle is determined
Before shooting, first form a brief angle brief: target audience and situation, the reason the audience is willing to enter and keep watching, the specific takeaway or experience the content provides, material anchors, the evidence ceiling, and what the main body should deliver. For explanatory content, also write clearly the core question or judgment; for content such as stories, humor, and aesthetics, validate according to its own function. Then check whether the existing draft delivers it; give the minimal suggestions to keep, move forward, compress, or supplement. When key facts are missing, first ask the one question that would change the final draft; when there is only material or an outline, the title and opening can be provisional plans, and it must be made clear what the main body must supplement. When a draft already exists and the material is sufficient, there is no need to wait for another confirmation just for the process.
After the main body can deliver the angle, generate the text title, cover title, and video opening the user needs. The video opening before shooting must be directly speakable on camera, and give the next line that connects to the main body. By default, deliver the angle brief, necessary main body changes, and the required title and opening; deliver a complete verbatim script only when the user explicitly requests a rewrite of the entire script. If the title or cover requires facts not present in the main body, first adjust the modifiable draft; if it cannot be supplemented, narrow the promise.
For tasks that have already been shot, handle according to production status and work mode : when reshoots are possible, first give the existing footage version, then list the minimum reshoot content; when reshoots are not possible, recommend only the current version that the original words and footage can deliver. Other valuable propositions can be opportunities for the next piece of content and must not be mixed into the title and cover of the current finished video.
Angle confirmation gate
When the user has not yet locked the content angle and different candidates would form different videos, use the following structure:
| Number | Candidate main lines within the same finished video scope | Who will enter and why they keep watching | Original material anchors and evidence boundaries | What takeaway or experience the main body must deliver | How to achieve it under the current production status |
|---|---|---|---|---|---|
| 1 | …… | Specific audience, reason for entering, and value of continuing to watch | Original text / user's own account / inference to be verified | Evidence, process, judgment, or experience | Can be used directly / needs revision / needs reshoot / leave for the next piece |
Place the angle confirmation card after the item-by-item results of the 12 perspectives and the converging proposition. Number candidates consecutively from 1. After the table, explain the issues or interpretive frameworks that are truly different among the directions, and write "Please reply directly with main line number 1/2/3; if you need auxiliary lines, you may reply '1 main line, 3 auxiliary'." Users may also propose directions themselves. Auxiliary lines can only serve the main line. Independent topics that cover only a segment of the existing content should be placed outside the table and not listed alongside the main line of the entire piece. Do not mix multiple issues into the same title or opening without confirmation; the 12-perspective results also cannot replace this decidable angle confirmation card.
When the manuscript's entry benchmark matches a candidate main line, list the original sentence and its historical performance framing after that row or table; when the original sentence belongs to another main line, let it undergo the same acceptance under that main line. The user's choice of main line does not equal automatic approval or rejection of the original sentence; at the text title stage, also check whether it matches the final body text and platform display position.
The following situations can skip confirmation:
- The user has already clearly specified the core proposition, point of contention, explanatory framework, or level of abstraction;
- It is only compressing word count, adjusting colloquial wording, or adapting to publishing fields;
- The material supports only one direction that can be delivered, and other candidates have been eliminated due to distortion or insufficient evidence.
Skipping confirmation does not mean skipping the body text delivery check.
Generate the required surfaces
text title
After the angle is determined, follow the module handoff to retain the selected main line, manuscript entry benchmark, factual boundaries, and production status, and read display position and title. Only execute title diagnosis, direct generation, and quality thresholds; do not repeat angle discovery or confirmation. Organize the text according to the platform's actual filling position; different platforms can share the same sentence, and adjust only when field length, search recognition, display context, or user goals change.
By default, generate two sets of text first: the Xiaohongshu title, including punctuation, must not exceed 20 characters, and the Bilibili title is used directly; the Douyin work copy, including punctuation, must not exceed 30 characters, and the WeChat Channels work description is used directly. Rewrite separately and explain the reason only when the actual display conditions change.
cover title
When the user needs a cover, read cover title, and share the selected angle with the text title. Deliver only the cover title text and line breaks, and do not provide cover image suggestions.
video opening
Read opening and body text connection. Before shooting, you can write an opening that can be spoken directly and the next sentence that connects to the body text; when footage already exists, prioritize selecting, moving forward, cutting, and combining original speech, and clearly define the footage positioning, editing action, return-to-body-text position, and handling of the original position.
Unless the user clearly is willing to reshoot, do not mix newly written sentences into existing original speech. Only reliable timecodes can provide precise cut points; when only subtitles have been seen, do not claim to have verified expressions, pauses, voice, and shot continuity.
Deliver the three types of text separately
- text title: the title, work copy, or work description the user fills in on the platform publishing interface.
- cover title: the title placed directly on the content cover; it is an independent title surface and is not split into "large text/small text".
- video opening: the spoken part, original interview speech, action, or cold open segment that appears after the final video starts playing.
Chapter text and explanatory text within the video画面 belong to another type of delivery. Do not generate them proactively when the user has not explicitly requested them, and do not use platform publishing fields to replace the video opening.
default delivery
First use one sentence to explain the adopted content direction. Before shooting and when the angle is already determined, first deliver the angle brief and the necessary minimal changes to the body text, then deliver the required title and a directly speakable opening; when only one text surface is needed, keep it as quick delivery. By default, deliver 1 preferred set; when there are substantially different and deliverable options, then provide up to 2 alternative sets. When the user requests batch or matrix production, deliver according to the specified quantity.
text title and cover title
For video-type content, delete table rows according to the actual platform:
| Platform | Text title placement | Text title that can be copied directly | Cover title usage location | Cover title that can be used directly |
|---|---|---|---|---|
| Xiaohongshu | Note title | …… | Image-text cover | …… |
| Bilibili | Video title | …… | Image-text cover | …… |
| Douyin | Work copy | …… | Homepage index cover | …… |
| WeChat Channels | Work description | …… | Image-text cover | …… |
WeChat Official Account usage:
| Platform | Text title placement | Text title that can be copied directly |
|---|---|---|
| WeChat Official Account | Article title | …… |
Only when the user explicitly requests a WeChat Official Account cover title, add a separate "Cover Title" field and follow the cover template provided by the user. If the user only wants the text title, delete the cover title column; if only the cover title is wanted, delete the text title column. When the text is the same across multiple platforms, still write it out completely item by item; do not use "same as above." When changing only one platform or one title, the table may be omitted, but "platform + usage location + title" must be clearly stated.
Opening for a newly written or reshot video
- Recommended opening: the first few sentences that can be spoken directly to camera;
- Next sentence / transition point: the exact connecting sentence or original passage;
- Reason to stay: explain, based on the current material, why viewers will continue;
- Body fulfillment: point out how the existing content supports the opening, and if there are gaps, list only the necessary additions.
Cold open in already shot footage
| Opening order | Footage location | Original words / visuals used | Editing action |
|---|---|---|---|
| 1 | Timecode or first and last sentences | Keep original words | Move earlier / trim / keep |
After the table, clearly state the position for returning to the main text, how the original position is handled, the new content, and the audio-visual edit points that still need to be checked.
When the user requests a text title, cover title, and opening at the same time, deliver the title table first, then the opening; all three share the same direction, each adds effective information, and they should avoid verbatim repetition and dependence on each other to be understood.
Default platform scope
- When the user specifies platforms, output only the specified platforms.
- When the user does not specify platforms and the input is a video, the text title and cover title cover Xiaohongshu, Bilibili, Douyin, and WeChat Channels by default.
- When the input is an article, generate a WeChat Official Account text title by default; generate a cover title only when the user explicitly needs one.
- When the user only requests a video opening, do not add multi-platform text titles and cover titles.
Acceptance and boundaries
Read content fulfillment check, correct distortion, unfulfillable claims, confused sources, and wrong usage locations, then recommend the finished product.
Do not promise increases in traffic, click-through rate, or retention rate; do not fabricate numbers, experiences, authority, causation, motives, or first-person quotations; do not delete key conditions in order to expand the audience. When the material lacks strong contrast, specific, useful, and accurate expression can still be recommended.
Use natural, direct Chinese expression and follow the Chinese Copywriting Guidelines. Avoid empty binary reversals and abstract packaging words.
After completing the current task, end directly. Only when the user explicitly asks about the next step and the current environment already has /dbs installed, briefly prompt: "When unsure about the next step, you can enter /dbs。」