SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsslides-video

slides-video

Design & creation

View repository

Updated 10-16 06:17

About this skill

Slides-Video · Slide-driven narration video

Create structured videos where "one PPT slide corresponds to one narration section". What this skill captures is theproduction method and proofreading workflow, not any particular visual style — the style is decided by the user and implemented by downstream skills.

Pre-flight · Dependency check and slides skill selection

This skill is theorchestration + methodologylayer; it does not generate slides itself — it calls aslides generation skill + video-planner。

Dependency Function
Aslides generation skill Generates a single-file HTML horizontal page-turning deck · handles all visual styles
video-planner Generates script.md / youtube.md / bilibili.md / x.md and other scripts and publishing materials

Not tied to any single slides skill — choose from the currently available ones

Before starting, first determine which slides skill to use:

  1. Identify candidates — any skill in the session that can produce a single-file HTML horizontal page-turning deck is a candidate. Common ones include magazine-web-ppt、guizang-ppt-skill、frontend-slides — butthe ones actually listed in this session take precedence; do not assume a particular one is certainly present, and do not hard-code any one.

  2. 0 available → Do not write the deck logic by hand yourself. Tell the user that there is currently no available slides generation skill, andrecommend installing onebefore continuing, for example:

    After installation, return to step 1 and identify again.

  3. Exactly 1 → use it directly, and tell the user which one you are using when starting work.

  4. Multiple available → You must use AskUserQuestion to confirm with the user which one to choose AskUserQuestion (list the candidates + each one's style characteristics); do not make the decision for the user.(list the candidates + each one's style characteristics); do not make the decision for the user.

Once selected, all later references to "calling the slides skill" mean this selected one;video-planner is fixed for scripts and publishing materials.If any required dependency is unavailable, stopand tell the user — do not rewrite the deck / script generation logic yourself (that would lose consistency with the ecosystem).

It is recommended to call in parallel personal-chinese-writing-style to ensure the language style matches the author.

Applicable scenarios

Suitable — any video that needs slides paired with narration:

  • Release interpretation (new models, new products, new versions)
  • Product reviews / technical explanations
  • Breakdown of papers / reports / industry data
  • Multi-subject side-by-side comparison
  • Trend observation / commentary on phenomena

Not suitable — pure tutorials (use the general video-planner is enough) · pure screen demonstrations (slides are not the main subject) · short videos / Shorts.

Core principles · The methodology of this skill

These 4 are the methodological principles running through the entire workflow.They do not involve specific styles, only prescribethe way of doing things。

1. PPT-script 1:1 synchronization

Each PPT page = one script section. When recording the video, turning the page = switching sections.

  • Each script section starts with a page-turn marker:【PPT 切到 Slide N · 页名】
  • Total video duration ≈ number of pages × average 30-50 seconds per page
  • Page count budget reference:
    • 5-6 minutes → about 9 pages
    • 7-8 minutes → about 13-15 pages
    • 3-4 minutes → about 6-7 pages

This 1:1 constraint is this skill's core increment relative to general video-planner and is non-negotiable.

2. Language is aimed at the target audience, not internal experts

Whether the audience is AI enthusiasts, developers, or ordinary users:

  • Terms appearing for the first time must be explained in plain language in one sentence, and jargon must not be used bare
  • Use analogies / metaphors to replace abstract nouns (so the audience can form an image in their minds)
  • At the end of each section, point out "what it means for the audience" — translate technical points into scenarios / value the audience can feel
  • Numbers must come with a frame of reference — do not pile up bare numbers like "30%" or "1.6T"; give a reference ("equivalent to X", "Y times the previous generation")

Who the audience is should be aligned with the user in Step 1. Different audiences require different depth of explanation and choice of metaphors.

3. Choose one of two narrative frameworks

Must be decided before starting:

  • Single protagonist — deeply break down one topic/product · pages unfold linearly
  • Multiple protagonists — multiple topics/products compared in the same frame · pages are organized as "an independent act for each protagonist"

This determines the page structure. Do not start writing vaguely; it is hard to change midway.

4. QA review is mandatory after generation

After the PPT is generated, it must be verified:

  • Each page's contentdoes not overflow the foot area (overflow audit)
  • Whether language style / terminology density / landing on the user perspective are in place (language review)

See references/overflow-audit.md。

Fix principle — change the content, not the template. The template (CSS / components) isthe selected slides skill 's maintenance scope; this skill does not touch it.

Workflow

Each step points to the corresponding method guidance in references/ .

Step 1 · Requirements clarification

Ask the user (skip what has already been given):

  1. Topic — what is it about? (required)
  2. Narrative framework — single protagonist vs multiple protagonists? (default: single protagonist)
  3. Target duration — about how many minutes? (default 7-8 minutes)
  4. Target audience — AI enthusiasts / developers / ordinary users / ...? (default: AI enthusiasts)
  5. Reference materials — tweets / papers / blogs / source code repositories?
  6. Visual style — inherit the style of an existing project? (give URL / project path) · or start completely fresh?

Item 6 is crucial — this skill does not prescribe style; the style is specified by the user here. If inheriting an existing project, pass the corresponding configuration as-is in Step 5 toThe chosen slides skill。

Step 2 · Research

See references/research-method.md. Core:

  • Use WebFetch / WebSearch to pull tweets, announcements, reports
  • Use Read to read paper PDFs (supports pages parameter to extract specific pages)
  • Use Explore agent / Grep / Glob to explore the code repository
  • Key data comes with source records — each benchmark score / price / date has its source noted, for easy citation in the foot line

Step 3 · Plan the structure

See references/planning.md。First draw up the page rhythm table, confirm with the user, then start writing。

Planning artifact: a table of N pages × each page's topic + topic class (light/dark/hero light/hero dark).

Step 4 · Create the output directory

Create per project convention:

{output-dir}/{YYYYMMDD}-{slug}/
├── ppt/
│   ├── index.html         # 由所选 slides skill 生成
│   └── images/            # (可选)插图
├── script.md              # 由 video-planner 生成,本 skill 加 1:1 标记
├── youtube.md
├── bilibili.md
└── x.md

Step 5 · Call the chosen slides skill to generate the PPT

Use Skill tool to call the slides skill selected in Pre-flight, passing in the configuration gathered in Step 1 and Step 3:

  • Theme color (user-specified / inherited from the reference project)
  • Page rhythm table
  • Content (per the Step 3 plan + Step 2 research)

This skill does not dictate the theme color, nor the cover / masthead style — these are chosen by the user andthe chosen slides skill is responsible for implementing them. Different slides skills have different page type / template conventions, so follow its own.

Step 6 · Call video-planner to generate the script + publishing assets

Use Skill tool to call, and apply this skill's method enhancements:

  • At the start of each paragraph in script.md, add a page-cut marker 【PPT 切到 Slide N · 页名】
  • Add a PPT sync note comment at the top of script.md
  • End each paragraph on "what it means for the audience"
  • See references/script-method.md

For detailed conventions for publishing assets (youtube/bilibili/x), see references/publishing-method.md。

Step 7 · QA review

See references/overflow-audit.md。

Two types of review:

  • overflow review — use chrome-devtools MCP to run JS to scan each page; only ≤ 5px counts as passing
  • language review — read through the script + PPT, check terminology density, user perspective, and reasonableness of metaphors

Step 8 · Iterate

Adjustments after user feedback:

  • All modifications use the Edit tool (surgical edit), not whole-file rewrite, so the user can clearly see each change
  • Prioritize content rewrites, leave templates untouched
  • After iterating, re-run the Step 7 overflow review

Step 9 · Completion report

Brief checklist:

视频制作完成,产物在 {目录}:
├── ppt/index.html         — N 页 PPT(全部 0 overflow)
├── script.md              — N 段口播(带切页标记)
├── youtube.md / bilibili.md / x.md — 发布素材

PPT 已在浏览器打开,可以开始录制。

Engineering notes

Style choices are left to the user

This skill's principle is fixed method · open style:

  • Theme color, brand name, masthead copy, cover design — all decided by the user in Step 1 or inherited from the reference project
  • This skill does not dictate a specific visual scheme

If the user wants to follow the style of a previous episode's video, in Step 5 pass that project's PPT configuration tothe chosen slides skill(theme color, cover structure, brand elements, etc.). Note: inherited styles should ideally come from the same slides skill; cross-skill inheritance may require adaptation due to different template systems.

Do not auto-publish

Only generate files,do not call any publishing API. The user publishes manually.

Reuse personal promotion info

video-planner will read the author's fixed promotion block from auto memory's video-promo.md . This skill trusts this mechanism and does not reinvent it.

Folder naming

Uniform {YYYYMMDD}-{slug} format. The date defaults tovideo production/publish date(not the product release date), unless the user specifies otherwise.

Examples

Existing outputs can serve as reference, butshould not be treated as a style that must be copied — the next video can keep the same style (as a series) or start a completely different visual set:

  • Single-protagonist deep-dive version reference —— src/content/videos/20260424-deepseek-v4/
  • Multi-protagonist comparison version reference —— src/content/videos/20260424-frontier-releases/

Refer to their structural approach(page pacing, 1:1 sync, audience-perspective landing, overflow control), notthe specific visuals(theme color, brand name, masthead copy).

Critical Rules

  1. Explicitly invoke the chosen slides skill + video-planner — do not rewrite the deck / script generation logic yourself; the slides skill is not fixed, choose from those currently available, and when there are multiple candidates use AskUserQuestion to confirm with the user
  2. Plan before acting — the page pacing table in Step 3 must be confirmed with the user
  3. 1:1 sync is non-negotiable — each PPT slide corresponds to one script segment, and each script segment begins with a page-turn marker
  4. Every segment must have an audience perspective — technical points must be translated into value the audience can feel
  5. Terms must be explained on first appearance — do not assume the audience understands
  6. Style is decided by the user — this skill does not lock down theme color / brand name / specific page design
  7. PPT must undergo QA review after generation — overflow ≤ 5px counts as done
  8. Fix content, do not change the template — solve overflow / terminology issues by changing content, not by changing CSS
  9. Do not auto-publish — only produce files, the user publishes manually