SkillAgentSearch skills...

publish

Publish a finished video to YouTube

Install / Use

npx skills add digitalsamba/claude-code-video-toolkit

Installs into whichever agent you are using.

About this skill
⚡

Claude Commands

Claude Code slash commands

Quality Score

86/100

Category

Marketing

Supported Platforms

Claude Code

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.

Substance
29/30
Structure
20/20
Description
8/15
Adoption
14/20
Freshness
15/15

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.

SkillScoreStarsUpdatedFormat
publish (this skill)by digitalsamba862.5k5d agoClaude Commands
claude-memby thedotmack100117.9ktodayCLAUDE.md
taste-skillby Leonxlnx100117.6k9d agoCLAUDE.md
Agent-Reachby Panniantong100112.7k9d agoCLAUDE.md
Understand-Anythingby Egonex-AI10093.0k2d agoCLAUDE.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.

description: 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.md and stop until YOUTUBE_CLIENT_SECRETS_FILE is in .env and uv run tools/youtube_upload.py --auth has 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:

  1. render.file if present (e.g. out/ai-agent-short.mp4), relative to the project dir.
  2. Else scan the project's out/*.mp4 and pick the most recent.
  3. 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 reports errorType: "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 shorts and upload the -9x16.mp4 it 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:

  1. Say "improve this" → Claude captures it in _internal/BACKLOG.md
  2. Edit .claude/commands/publish.md → update _internal/CHANGELOG.md
  • Issues/PRs: github.com/digitalsamba/claude-code-video-toolkit

Related Skills

View on GitHub
GitHub Stars2.5k
CategoryMarketing
Updated5d ago
Forks444

Languages

Python

Trust signals

100/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

No cautions
publish — Claude Commands: Install & Safety Check | SkillAgent