Turn the selected angle into a format brief that can actually be completed, rather than listing every format.
Confirm Inputs First
Extract from the given data: topic, selected angle, platform/purpose, willingness to appear on camera, existing footage or screen material, time available, what needs to be proven, and the desired viewer next step. When data is missing, ask 1–3 questions at a time about what most affects the choice; do not fill gaps with assumptions.
- First confirm whether there is material to support "what needs to be proven"; if not, recommend a format that can honestly express the limitations, and do not fabricate results.
- Do not call a format more advanced just because it is complex; when resources are insufficient, prioritize formats that can be completed and that communicate clearly.
- Carousel image-and-text is a content format, but it is not video; text-only Reels are still dynamic short-form video.
- For any format that will include third-party video, screenshots, or audio, first save the source and confirm authorization or the applicable scope of citation. If rights are unclear, do not use the material at all; being publicly visible does not equal being authorized.
Workflow
- Restate the inputs and constraints in one sentence so the user can correct them first.
- Read
references/八種呈現形式.mdandreferences/選擇矩陣.md, compare only among the eight fixed formats; check in order platform publishability, third-party rights, proof material, willingness to appear on camera, existing footage, time, and viewer next step. - Use the matrix's tie-breaking rules to select one primary and one backup. The backup must take another completable material path, not just rename the same option.
- Read
references/輸出模板.md, and fill in for each: fit rationale, first frame, 30–90 second or image-and-text rhythm, necessary materials, shooting/production checklist, and risks. - Show the result in the conversation first; only create a file if the user explicitly asks to save it, and confirm the location first.
Boundaries and Guardrails
- Picture-in-picture: two recognizable layers of footage exist at the same time in key segments, for example your own main screen plus a small window of a person explaining; a single screen recording does not count as picture-in-picture.
- Tutorial demonstration: the main subject is operating steps, objects, or a single screen recording, so viewers can understand how to do it; switch to picture-in-picture only when a second layer of footage must continuously help with interpretation.
- Response/remix: when information or rights are insufficient, do not put third-party clips, screenshots, or audio into any format; instead use your own talking-head delivery, your own screen that does not show third-party content, self-made diagrams, or text-only paraphrase.
- Carousel image-and-text: use page order, swipe rhythm, and cover description; do not force it into a 30–90 second video rhythm.
Stopping Point and Quality Check
Stop at the format brief; do not write the full script, verbatim transcript, hook, shot list, editing instructions, or publishing schedule for the user. The "shooting/production checklist" should list only equipment, materials, location, people, and production preparation; it should not list shot order, lines, transitions, or segment-by-segment editing methods. Use natural Traditional Chinese as used in Taiwan; do not guarantee traffic, reach, completion rate, or commercial results.