agentleFS
godot-fungodot-fun/gai/.agents/skills/video-publish/SKILL.md

video-publish

Skill210 starsChanged 2 days agoMITView on GitHub

From user materials, generates multi-platform video publish pack: Zhihu Chinese article, Reddit English post, 3 landscape + 3 portrait Google image-gen cover prompts, and title/description/tags for Bilibili, Douyin, Xiaohongshu, Weibo, YouTube, X, TikTok, Instagram — each in separate files, platform-native and click-worthy. Use when the user wants video-publish, video publishing, publish copy, cover prompts, title/description/tags, Zhihu/Reddit posts, or multi-platform upload copy.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

17 rules. About 0 are common, 12 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not invent product facts, prices, logos, or claims not in the materials.

    line 21
  2. Distinctive:

    If goal/audience is ambiguous, state one short assumption and proceed.

    line 22
  3. Distinctive:

    Always overwrite these filenames in the output directory (create if missing).

    line 31
  4. Very specific:

    Prefer plain ` with no language

    line 53
  5. Distinctive:

    Do not put multiple platforms or multiple fields in one fence

    line 55
  6. Distinctive:

    Write all 5 files (articles as normal MD.

    line 61
  7. Distinctive:

    Prefer “too much useful detail” over thin summary.

    line 76
  8. Distinctive:

    If materials include code, snippets, scripts, API usage, or repo paths: pull in the user’s own code, not generic invented samples

    line 81
  9. Distinctive:

    Prefer several short, real excerpts over one toy rewrite.

    line 82
  10. Very specific:

    Use fenced code blocks with the right language tag (gdscript, python, cpp, …)

    line 83
  11. Distinctive:

    If materials only mention code indirectly (e.g.

    line 85
  12. Distinctive:

    No code in materials and none findable → skip code sections.

    line 86
  13. Very specific:

    No hard sell, no keyword stuffing.

    line 98
  14. Distinctive:

    Do not use one universal tag format.

    line 157
  15. Distinctive:

    No irrelevant viral spam, no competitor-brand stuffing unless in materials

    line 172
  16. Very specific:

    Use exact section headings ## Bilibili … ## Instagram (stable for copy-paste / tooling).

    line 259
  17. Very specific:

    If materials are thin, still ship all files.

    line 268

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Video Publish

From user materials only, produce a full publish pack and write files (do not only print in chat).

Inputs

Any mix of: video/storyboard, title ideas, product/brief text, stills, constraints. Optional: one input file path (used only to locate the output directory).

Do not invent product facts, prices, logos, or claims not in the materials. If goal/audience is ambiguous, state one short assumption and proceed.

Output directory

ModeWrite files to
User provides an input fileThat file’s parent directory
No input fileProject root

Always overwrite these filenames in the output directory (create if missing).

Required output files

FileContent
zhihu.md1 Chinese Zhihu article
reddit.md1 English Reddit post
covers-landscape.md3 landscape Google image-gen prompts
covers-portrait.md3 portrait Google image-gen prompts
platforms.mdTitle / description / tags for all 8 platforms

Chat after write: short confirm only (output dir + file list). Do not dump full copy into chat unless the user asks.

Copy-friendly MD

Articles are normal Markdown — zhihu.md and reddit.md stay readable long-form (headings, paragraphs). Do not wrap the whole article in a code fence.

Paste fields use fences — platform Title / Description / Tags, and each cover prompt: one field = one fence so one-click copy works.

Fence rules (covers + platforms.md only):

  • Headings/labels outside the fence; only the text to paste inside
  • Prefer plain ``` with no language
  • Inside: raw publish text only — no extra structural markdown
  • Do not put multiple platforms or multiple fields in one fence

Workflow

  1. Resolve output directory (file parent vs project root)
  2. Inventory materials; infer topic, hook, audience, tone
  3. Write all 5 files (articles as normal MD; covers + platforms with paste fences)
  4. Consistency pass (same core claim/hook across platforms; no invented facts)
  5. Confirm paths in chat

Platform styles (must match)

Eye-catching first; still native to each platform — not the same paragraph pasted everywhere.

Articles

Both articles must be substantive, not teaser blurbs. Expand from the user’s materials: explain context, walk through key beats, add concrete detail, and give readers something useful even if they never open the video.

Depth (required)

  • Zhihu: typically 1500–3500 Chinese characters (or more if materials are dense); multiple ## sections
  • Reddit: typically 600–1500 words; clear sections, still conversational
  • Prefer “too much useful detail” over thin summary; do not pad with empty hype
  • Structure idea: hook → background / problem → what we built or showed → how it works → takeaways → soft CTA
  • Reuse facts, names, numbers, and claims only from materials; when expanding prose, stay faithful

Code from the user (required when present)

  • If materials include code, snippets, scripts, API usage, or repo paths: pull in the user’s own code, not generic invented samples
  • Prefer several short, real excerpts over one toy rewrite; keep enough context to be readable
  • Use fenced code blocks with the right language tag (gdscript, python, cpp, …)
  • Lightly annotate why each snippet matters; do not “improve” the code into something they didn’t provide
  • If materials only mention code indirectly (e.g. storyboard about a feature), open related project files when a clear path/repo is in scope and quote their implementations
  • No code in materials and none findable → skip code sections; do not fabricate demos

Zhihu (zhihu.md) — Chinese, normal Markdown article

  • # title as H1; body with ## / ### subheads, lists, and code fences as needed
  • Long-form, structured, credible; strong hook title + opening question/conflict
  • Fill sections with explanation, comparisons, pitfalls, and material-grounded examples
  • Light CTA at end (watch / discuss); avoid slang spam and emoji walls

Reddit (reddit.md) — English, normal Markdown post

  • # title as H1; body as readable prose (not one giant fence)
  • Discussion-first, authentic; title like a real post, not ad copy
  • Same richness bar as Zhihu (translated tone): context → build/demo → code when available → ask the room something specific
  • No hard sell, no keyword stuffing; optional ## Suggested subreddits (2–5 as a normal list)

Cover prompts

Google image generation prompts (Imagen / Gemini image). English prompts preferred (models follow them more reliably). Each prompt must be self-contained: subject, mood, composition, lighting, text-in-image rules, aspect.

covers-landscape.md — 3 variants, 16:9 thumbnail / banner

  • Safe margins for platform UI crop; readable focal point left/center
  • State: aspect ratio 16:9, no watermarks, no logos unless in materials

covers-portrait.md — 3 variants, 9:16 short-video cover

  • Vertical hero, face/product large in upper two-thirds; room for caption overlay
  • State: aspect ratio 9:16, no watermarks, no logos unless in materials

Each file format:

# Covers — landscape (16:9)

## Variant 1 — [short label]

```
[full prompt — paste-ready]
```

## Variant 2 — [short label]

```
[full prompt — paste-ready]
```

## Variant 3 — [short label]

```
[full prompt — paste-ready]
```

(Same structure for portrait with 9:16.)

Three variants should differ in composition/hook (e.g. curiosity gap / bold claim visual / emotional close-up) — not three near-duplicates.

Video platforms (platforms.md)

Generate all eight. Each section: Title, Description, Tags — each in its own fence.

PlatformLanguageStyle notes
BilibiliChineseSearchable title; description with highlights / chapters if known
DouyinChineseUltra-short hook title; first desc line = scroll-stopper
XiaohongshuChineseTitle + emoji OK; lifestyle/note tone
WeiboChineseShort punchy; @/topic friendly; desc can double as post body
YouTubeEnglishSEO title (~≤100 chars ideal); description with hook + value + CTA
XEnglishPunchy post-length energy; desc ≈ post text if needed
TikTokEnglishCaption-first hook; keep title/caption tight
InstagramEnglishReels/caption tone; line breaks in description

Tags — follow each platform’s rules (required)

Do not use one universal tag format. The Tags fence must match how that platform expects paste-in.

PlatformFormat inside Tags fenceCount / notes
BilibiliPlain keywords, comma-separated, no # (upload tags field)~8–12; mostly Chinese; mix category terms + content-specific terms; avoid unrelated hot keywords
Douyin#topic space-separated, leading # only, no trailing #~5–8; mix 1–2 trending topics + precise terms; avoid spammy # walls
Xiaohongshu#topic space-separated, leading # only, no trailing #~8–15; common in note discovery; lifestyle/tutorial terms first
Weibo#topic# space-separated (opening and closing #)~3–6; keep short; trending-friendly; do not use long sentences as topics
YouTubeComma-separated SEO phrases, no # (YouTube Studio tags field)~8–15; may include 2–4 word phrases; keep English/proper nouns as-is
X#Hashtag space-separated, with #2–4 only; stay restrained; CamelCase or short words; no hashtag walls
TikTok#hashtag space-separated, with #~4–8; 1–2 broad + rest niche; avoid empty #fyp spam (max 1, optional)
Instagram#hashtag space-separated, with #~8–15; paste into Tags or first comment; mix broad + specific

Shared tag quality:

  • Relevant to the video; mix discoverable + specific
  • No irrelevant viral spam, no competitor-brand stuffing unless in materials
  • Language matches the platform column above

platforms.md structure:

# Platform publish copy

## Bilibili

### Title

```
[title]
```

### Description

```
[description]
```

### Tags

```
Godot, GDScript, game development, …
```

## Douyin

### Tags

```
#Godot #GameDev #IndieGame …
```

## Xiaohongshu

### Tags

```
#Godot #GameDev #IndieGame
```

## Weibo

### Tags

```
#Godot# #IndieGame#
```

## YouTube

### Tags

```
Godot, GDScript, game development, …
```

## X

### Tags

```
#Godot #IndieDev
```

## TikTok

### Tags

```
#Godot #gamedev #indiedev …
```

## Instagram

### Tags

```
#Godot #gamedev #indiedev …
```

(Each platform still has Title / Description / Tags; examples above only show Tags shape per platform.)

Use exact section headings ## Bilibili … ## Instagram (stable for copy-paste / tooling).

Quality bar

  • Hook in the first line/title every time
  • Articles are rich and material-grounded; include the user’s own code when available
  • Same factual core across files; wording adapted per platform
  • Tags: per-platform format (see table); relevant + discoverable; no irrelevant viral spam
  • Covers: no unreadable tiny text; if on-image text is needed, keep to ≤5 words and say so in the prompt
  • If materials are thin, still ship all files; mark uncertain lines with (assumption: …) once in chat, not inside every file

Examples

Input: res://… storyboard + “publish this video”
Output dir: storyboard file’s folder → write the 5 files there.

Input: paste brief only, no path
Output dir: project root → write the 5 files there.

CLI

Copy-paste commands: cli/video-publish.md

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.