publish
Publish a finished video to YouTube
Install / Use
npx skills add digitalsamba/claude-code-video-toolkitInstalls into whichever agent you are using.
Claude Commands
Claude Code slash commands
Quality Score
Category
MarketingSupported Platforms
Our assessment of publish
publish scores 86/100 on our quality scale, 380th of 646 Marketing skills we index.
Its Claude Commands is 11 KB long, well organised into 14 sections with 8 code examples: a thorough specification that gives an agent plenty to work with.
With 2,544 GitHub stars, it is one of the more widely adopted skills in the catalogue.
Maintenance, license and trust
- The repository was last updated 5 days ago, so publish is actively maintained.
- It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
- Its trust signals score 100/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
publish compared with similar skills
All 4 of these similar skills score higher than publish; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| publish (this skill)by digitalsamba | 86 | 2.5k | 5d ago | Claude Commands |
| claude-memby thedotmack | 100 | 117.9k | today | CLAUDE.md |
| taste-skillby Leonxlnx | 100 | 117.6k | 9d ago | CLAUDE.md |
| Agent-Reachby Panniantong | 100 | 112.7k | 9d ago | CLAUDE.md |
| Understand-Anythingby Egonex-AI | 100 | 93.0k | 2d ago | CLAUDE.md |
Frequently asked questions
- How do I install publish?
- Run
npx skills add digitalsamba/claude-code-video-toolkit. The install tabs above show the steps for each supported agent. - Which AI agents does publish work with?
- It is written for Claude Code, as a Claude Commands file. Other agents that read the same format can often use it too.
- Is publish safe to use?
- It is MIT-licensed and scores 100/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is publish still maintained?
- The repository was last updated 5 days ago, so publish is actively maintained.
Skill content
View source on GitHubdescription: Publish a finished video to YouTube
Publish to YouTube
Upload a rendered project to YouTube, auto-filling the metadata from project.json.
Wraps tools/youtube_upload.py (OAuth 2.0 + Data API v3, resumable upload).
check setup → project.json + rendered MP4 → publish kit (subtitles, chapters) → metadata draft → dry-run
→ upload → write back videoId/URL
One-time setup required. YouTube uploads need OAuth (not an API key). If the user hasn't set this up, point them at
docs/youtube-upload.mdand stop untilYOUTUBE_CLIENT_SECRETS_FILEis in.envanduv run tools/youtube_upload.py --authhas been run once. Don't attempt an upload without a cached token.
Step 0: Check the setup
cd /path/to/claude-code-video-toolkit && uv run --extra youtube tools/youtube_upload.py --check --account NAME --json-out
Read-only, about 1 quota unit. It reports the channel the login belongs to (say it to the user: personal and
default are different channels), the scopes, and whether the Data and Analytics APIs are switched on. Each
problem comes with a fixes line: a login command for the user to run with !, or the exact Cloud Console page
that switches an API on. Stop until success is true. Analytics being off doesn't block an upload; mention it.
Entry Point
Step 1: Locate the project and its rendered video
If the user named a project, use it. Otherwise scan for candidates:
cd /path/to/claude-code-video-toolkit && ls projects/*/project.json
Read the chosen project.json defensively — real files carry render, format,
and publish blocks beyond the lib/project/types.ts schema. Resolve the video file:
render.fileif present (e.g.out/ai-agent-short.mp4), relative to the project dir.- Else scan the project's
out/*.mp4and pick the most recent. - Confirm the resolved path exists and is non-empty before continuing.
If the project's phase isn't complete, warn the user and confirm they still want to publish.
Step 1b: Publish kit (optional; on by default when the project has scene scripts)
cd /path/to/claude-code-video-toolkit && uv run tools/publish_kit.py --project projects/NAME --video out/FILE.mp4
Transcribes the render once (Scribe, $0.40 an audio hour, about $0.01 for 90 s; --engine whisper is free) and
aligns it to the scene scripts, so it writes out/publish/video.srt for the whole video, spelled as the scripts
spell it, and out/publish/chapters.txt, a chapter at each scene's cut. Show the chapters: titles come from
project.json scenes or the scene file names ("Narrate"), so offer to rename them; put the names in
project.json scenes[].title and re-run (it reuses the saved transcript, free). Notes say when a scene was too
short for its own chapter or the scripts don't match the render. Skip the kit when the user doesn't want
subtitles or chapters, or the video has no narration.
Step 2: Assemble metadata (into a publish block)
Build a publish object and write it back into project.json so it's reviewable,
editable, and re-runnable. If a publish block already exists, use it as defaults.
| Field | How to derive |
|-------|---------------|
| title | Existing publish.title, else the hook/title scene's title, else the project name (humanized). Keep ≤100 chars. Name the subject in the words people would type: a film's own name alone is not searched for. |
| description | Existing publish.description, else auto-draft: a 1–2 line summary from the scene narration/titles, then the links (where the video should send people, in the first lines), and two or three hashtags on the last line (YouTube shows the first three above the title). Keep ≤5000 chars. |
| tags | Existing publish.tags, else derive 5–12 topical tags from scene titles + the brand. Comma-joined when passed to the tool. |
| category | Choose one by topic: "28" (Science & Tech), "27" (Education), "24" (Entertainment), "26" (Howto & Style). Left out, the tool uploads as "22" (People & Blogs) and says so. |
| thumbnail | Existing publish.thumbnail, else out/thumbnail.* or public/thumbnail.*. If none and the user wants one, run /thumbnail first. |
| privacy | Default private (safe — the video uploads but stays hidden until you flip it). Offer unlisted, public, or a scheduled go-live (--publish-at, e.g. next morning ~09:00 local in UTC) if the user asks. |
| playlist | Optional; only if the user has one. |
| chapters | From the kit's chapters.txt: put the lines in the description (YouTube reads them there), unless it already has a 0:00 line. |
| captions | The kit's out/publish/video.srt, passed as --captions. |
| synthetic | On (--synthetic-media) when the video shows realistic people who were generated or animated by AI (an LTX-animated cast, FLUX portraits, a SoulX presenter) or a cloned voice saying new lines. YouTube requires the disclosure; it shows a label, it doesn't limit reach. Off for UI rebuilds, charts and stylised cartoons. |
Show the assembled metadata to the user and let them edit before uploading.
Step 3: Dry-run first (no upload)
cd /path/to/claude-code-video-toolkit && uv run tools/youtube_upload.py \
--video "projects/NAME/out/video.mp4" \
--title "TITLE" \
--description-file "projects/NAME/.publish-description.txt" \
--tags "tag1,tag2,tag3" \
--category "28" \
--publish-at "2026-06-10T09:00:00Z" \
--thumbnail "projects/NAME/out/thumbnail.png" \
--captions "projects/NAME/out/publish/video.srt" \
--dry-run --json-out
Write the description to a temp file (--description-file) rather than passing a long
--description on the command line. Parse the JSON: confirm requestBody looks right and
authOk is true. Read notes too: advice on the metadata (no hashtags, no link, no tag in the title, no
category, a Short that may need its related video). Fix what applies and show the user the rest; a note never
stops the upload. If authOk is false, surface the auth error and have the user run
uv run tools/youtube_upload.py --auth first — do not proceed.
Step 4: Upload
Re-run the same command without --dry-run, keeping --json-out. Parse the result.
Step 5: Write back and report
On success, merge into the project's publish block:
"publish": {
"platform": "youtube",
"videoId": "<id>",
"url": "https://www.youtube.com/watch?v=<id>",
"privacyStatus": "<actual returned status>",
"publishAt": "<scheduled time or null>",
"uploadedAt": "<today ISO date>"
}
Append a sessions[] entry summarizing the upload, then report to the user:
Published to YouTube
Title: <title>
URL: https://www.youtube.com/watch?v=<id>
Privacy: <actual> (requested: <requested>)
Schedule: <publishAt or "—">
If privacyStatus came back private but you requested public/scheduled, tell the
user plainly: this is the unverified-app lock — the video uploaded but won't go public
until their Google Cloud OAuth app is verified. They can publish manually in YouTube Studio.
If the upload is a Short cut from a longer video (the notes say it is a Short), tell the user to link the
two by hand: Studio, Content, Shorts, the Short's Details, Related video. The API can't set it, and the field
only appears once the channel's advanced features are on (a one-time verification in Studio's channel settings).
Put the long video's link in the Short's description as well, which works at once.
Then offer once, in one line: "Want to share it in the toolkit's Show and tell? /contribute option 6 drafts the post for you to check." Don't post anything from here.
Quick Mode
Direct invocation for experienced users:
/publish ai-agent-short
/publish ai-agent-short --privacy unlisted
Parse the project name and any privacy/schedule overrides, still show the metadata and run a dry-run before the real upload.
Tool Reference (tools/youtube_upload.py)
| Option | Description |
|--------|-------------|
| --video, --input | Path to the video file |
| --title | Title (≤100 chars) |
| --description / --description-file | Description text, or a file (- = stdin) |
| --tags | Comma-separated tags (combined ≤500 chars) |
| --category | Numeric category ID string (28 = Science & Tech; left out it is 22, with a note) |
| --privacy | private (default) / unlisted / public |
| --synthetic-media | Disclose realistic AI-generated or altered people, places or voices |
| --publish-at | ISO8601 UTC schedule, e.g. 2026-06-10T09:00:00Z (forces private at insert) |
| --thumbnail | Custom thumbnail (≤2MB, 1280×720) |
| --captions + --captions-language | Caption file + language code (needs the force-ssl scope; an older token asks for --auth) |
| --playlist | Playlist ID |
| --video-id | Add captions, a thumbnail or a playlist to an uploaded video, without uploading |
| --account | Channel name namespacing the cached token (default default) |
| --auth | Interactive login only — cache a token and exit |
| --check | Login, channel, scopes, Data/Analytics API on or off, with fixes; read-only |
| --dry-run | Validate + print the request body without uploading |
| --json-out | Single machine-readable JSON line on stdout |
Quota & Limits (worth knowing)
- Default API quota is 10,000 units/day; each upload costs ~1,600 units → ~6 uploads/day.
- Hitting quota returns HTTP 403
quotaExceeded(the tool reportserrorType: "quota"). - Custom thumbnails require a channel with a verified phone number.
- Shorts: YouTube makes any vertical or square upload of 3 minutes or less a Short; there is no flag for it. To publish a Short of a 16:9 video, cut it down first with
uv run tools/cutdown.py --input <render> --platform shortsand upload the-9x16.mp4it writes.
Error Handling
| Symptom (errorType) | Solution |
|-----------------------|----------|
| auth | Run uv run tools/youtube_upload.py --auth once; if the refresh token expired (7-day Testing limit), re-run --auth. |
| validation | Fix the flagged field (missing video/title, bad --publish-at). |
| copy | tools/copy_lint.py found a dash or stock AI/marketing phrasing in the title, description or tags. Reword it and show the user; --allow-copy only if they want those words. Lint drafts before showing them: uv run tools/copy_lint.py .publish-description.txt --text "TITLE". |
| config | YouTube Data API v3 isn't enabled on the Cloud project (or just enabled, still propagating). Enable it, wait 1–2 min, retry. See docs/youtube-upload.md. |
| quota | Daily quota exhausted — wait, or request more in Google Cloud. |
| forbidden | Permission denied (e.g. the authorized account lacks rights to the target channel). Check the account / --account. |
| upload / http | Transient network/server issue; the tool already retried — try again later. |
| Video uploaded but stuck private | Unverified OAuth app lock — verify the app, or publish manually. |
Evolution
This command evolves through use. If something's awkward or missing:
- Say "improve this" → Claude captures it in
_internal/BACKLOG.md - Edit
.claude/commands/publish.md→ update_internal/CHANGELOG.md
- Issues/PRs:
github.com/digitalsamba/claude-code-video-toolkit
Related Skills
claude-mem
117.9kPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
taste-skill
117.6kAgent skills that give coding agents taste: art direction, reference images and a browser check instead of generic AI pages
Agent-Reach
112.7kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
Understand-Anything
93.0kGraphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
