agentleFS
berkayburakkberkayburakk/tubecrew/skills/produce-video/SKILL.md

produce-video

Skill2 starsChanged 30 days agoMITView on GitHub

Run a YouTube video from idea to a private upload as a gated production line - topic, sourced research, script, voiceover, render, review, upload, then retention follow-up. Use when someone wants to make a video end to end, asks "what's the next step for this video", or starts a new video project. Each stage has its own skill; this one keeps the order and the gates.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Before the first video, run the setup skill once.

    line 26
  2. Very specific:

    Never invent a number, quote, result or source: If a figure is missing, ask for it or write the line without it.

    line 55
  3. Distinctive:

    No automated engagement on your own videos: No script or browser agent watches, likes or comments on your public videos.

    line 68
  4. Distinctive:

    After each station: what was produced (path), the gate result (exit code or verdict line), and the single next action.

    line 82

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “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

Produce a video

A video moves through stations. Each station writes a file, and the next station starts only when the previous one passed its gate. Thinking (topic choice, story, taste) is yours; counting and measuring is done by the scripts in each skill; reviewing is done by an agent that did not make the thing.

The project folder

One folder per video. Create it first:

<video>/
  facts.json   script.md   upload.json
  voice/       render/     qa/

qa/ holds every gate result. See CONTRIBUTING.md in the plugin root for the exact file contract.

The creator's tools

Before the first video, run the setup skill once; it saves the creator's tools and subscriptions in ~/.tubecrew/stack.json. At every station, use the tool chosen there instead of guessing. If the file is missing, run setup first.

The line

#StationSkill / agentGate (do not move on until)
1Topicfind-topic (demand.py for demand), then the topic-skeptic agentat least one candidate is ALIVE with evidence
2Researchresearch-factsfacts_check.py facts.json exits 0
3Scriptwrite-script, then the script-auditor agentlint gate passes, A/B verdict merged
4Number checkfacts_check.py --script, the fact-checker agent, and Codex when the stack's verifier is codex0 unsourced numbers, every checker PASS
5First sliceyou + the creatorthe creator approved the first 60-90 seconds
6Voiceoverthe stack's voice (ElevenLabs, CosyVoice or a recording), then check-voiceovervoice_check.py exits 0
7Visuals and editthe stack's motion engine (Remotion, HyperFrames or morph-clips) and clips (Wan)text on screen follows the dose rule below
8Render checkcheck-render (mix with normalize_audio.py), then the render-auditor agentrender-check.json PASS and RENDER-AUDIT PASS for the same file
9Packagetitle, description, check-thumbnail, optional localize for other languagesthumbnail-check.json PASS; thumbnail text does not repeat the title
10Uploadthe upload-checker agent, then upload-privateUPLOAD-GATE GO, uploaded as private (scheduled only with --schedule)
11Publishthe creator, in YouTube Studioa human pressed publish or chose the schedule
12Learnchannel-pulse weekly, retention-by-scene after 72 hone lesson in lessons.md; due decisions.md rows filled from tool output

Shorts: once the long video passed station 10, make-shorts turns it into Shorts with the same gates; each Short goes public only after its long video.

Station 5 exists because the most expensive mistake is a whole video built on an opening nobody liked. Render or mock the first 60-90 seconds and ask before you build the rest.

Rules that hold at every station

  • Never invent a number, quote, result or source. If a figure is missing, ask for it or write the line without it. Every number on screen or in the narration lives in facts.json with a link.
  • The maker never reviews their own work. Reviews go to the agents above, with the files, not a summary. Say what level of check you want (every figure, or totals only) and read which level the report ran.
  • Reviewers are for what ships. Drafts, tests and v1 experiments skip the review agents; the final render and anything published do not.
  • Measured problems are fixed, not argued. A script that exits 1 means fix the work. If a check is wrong for this video, the creator decides, and you write down why.
  • On-screen text is punctuation. Full-screen text cards stay under ~20% of the runtime, never two in a row, and never repeat what the narration is saying. Big text is for the hook question, a chapter turn, or the one number to remember.
  • Nothing publishes itself. Uploads are private. Publishing, and anything irreversible, is the creator's decision; a schedule is used only when the creator writes publish_at and passes --schedule.
  • No automated engagement on your own videos. No script or browser agent watches, likes or comments on your public videos. Fake engagement breaks YouTube's rules and can cost the monetization account that all of the creator's channels share.

Lessons become checks

Copy this skill's templates/ into the channel folder (decisions.md, lessons.md, gate-counter.md; costs.md goes into each video folder) and follow references/learning-loop.md. Lessons are numbered and append-only; check for duplicates before adding one. At the start of each video, put the lessons that apply into <video>/preflight.md. When the same lesson shows up twice, turn it into a script check instead of a longer rule. Log every real catch in gate-counter.md; after a month, a code gate that caught nothing is loosened and an agent gate that caught nothing is removed. The creator decides.

What to report back

After each station: what was produced (path), the gate result (exit code or verdict line), and the single next action. Do not hand over a list of twenty things.

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.