claude-youtube-editor
CLAUDE.md368 starsChanged 11 days agoMITView on GitHub
You record the talking head. Claude Code does the rest. Every screen moment — UI walkthroughs, full-screen statements, diagrams, terminal mockups — is built as Remotion TSX and composited over your cut. No screen recording, no video editor. The right skill is picked from the request: The pipeline order is: cut → visuals → voice → SFX → packaging → upload. brand.md is the style contract every step-2+ skill reads (palette, motion, delivery specs, SFX taste). remotion/src/brand.ts + fonts.ts are…
- Reads credentials
- Installs packages
In this repository9 other files it gives its agents.
- brand-setupSkill
- clean-audioSkill
- clean-cutSkill
- fake-screencastSkill
- make-tsxSkill
- youtube-packagingSkill
- suggest-sfxSkill
- youtube-thumbnailSkill
- vidtsx-2d-generatorSkill
The file
We split this CLAUDE.md into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 3
You record the talking head. Claude Code does the rest: Every screen moment — UI walkthroughs, full-screen statements, diagrams, terminal …
- Very specific: line 79
The Remotion registry is generated: after adding/renaming a shot, cd remotion && npm run gen (the render/frames scripts do NOT run it …
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md 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.
Files
CLAUDE.md — claude-youtube-editor
You record the talking head. Claude Code does the rest. Every screen moment — UI walkthroughs, full-screen statements, diagrams, terminal mockups — is built as Remotion TSX and composited over your cut. No screen recording, no video editor. The right skill is picked from the request:
| The user asks to… | Skill | Output |
| set up / change their brand, colors, fonts, wordmark | /brand-setup | brand.md + remotion/src/{brand.ts,fonts.ts} + a proof render |
| cut the raw footage / tighten pacing / remove fillers | /clean-cut | master cut + videos/<project>/work/{analysis/cuts.json, edited-transcript.json} |
| build the visual beats / add overlays | /make-tsx (+ /fake-screencast) | shots in remotion/src/shots/<project>/ + a baked preview |
| write a crash-free TSX shot | /vidtsx-2d-generator | the low-level Remotion authoring rules |
| remove background noise / isolate voice | /clean-audio | a cleaned master (levels preserved) |
| add SFX / sound-design a beat | /suggest-sfx | videos/<project>/work/sfx-plan.json + an audition mix |
| package a video / titles + thumbnails | /packaging | videos/<project>/packaging/ (1 title × 3 thumbnail bets + rendered thumbs) |
| make / iterate thumbnails only | /thumbnail | rendered + verified frames in videos/<project>/packaging/thumbs/ (asks which style elements you want) |
| upload it | tools/yt_upload.py | a private draft on YouTube |
| track it / put it in Notion | tools/notion_sync.py | the video's row in the Notion content tracker |
The pipeline order is: cut → visuals → voice → SFX → packaging → upload.
brand.md is the style contract every step-2+ skill reads (palette, motion, delivery specs, SFX
taste). remotion/src/brand.ts + fonts.ts are the same contract as code — /brand-setup owns all
three together; they drift silently if you hand-edit one.
Layout
tools/ Python tools (see requirements.txt); the cut-editor UI in tools/editor/;
RNNoise models in tools/models/rnnoise/
remotion/ the Remotion project — src/lib/ (kit, browser, screencast, vscode),
src/shots/<project>/; registry is GENERATED (npm run gen)
media/ Remotion's public root: library/ (reusable: sfx, music, logos, faces)
+ projects/<project>/ (media for ONE video — via staticFile('projects/<p>/x'))
videos/ per-video project data — EMPTY until you make one. See videos/README.md
brand.md the style contractThe example
remotion/src/shots/example/ is 37 real shots from a published video — the worked example of the
kit. They render standalone with no footage: cd remotion && npm run studio. Read a few before
authoring new ones; that's the fastest way to learn remotion/src/lib/.
remotion/src/shots/brand/BrandProof.tsx is not a video beat — it's a utility shot that renders the
current brand (wordmark, palette, type) so you can see it. /brand-setup uses it.
Conventions (hard rules)
- Run everything from the repo root. Tools resolve engine paths (media/library, catalogs,
remotion/out) against their own location, but project paths against the CWD — so pass the project
as
videos/video-1and run from the repo root. Example:python tools/render_cuts.py videos/video-1 --style natural --mode preview. - Python: use a venv at the repo root. These tools have real dependencies. Create it once:
Then either activate the venv or callpython -m venv venv venv/Scripts/python -m pip install -r requirements.txt # Windows ./venv/bin/pip install -r requirements.txt # macOS/Linuxvenv/Scripts/python tools/<tool>.py. A barepythonthat resolves to a system interpreter will hitModuleNotFoundError(requests, Pillow, google-*) — that error means you're not on the venv.ffmpeg/ffprobeandnode/npxmust be on PATH (not pip). - API keys live in
.envat the repo root (copy.env.example). Never commit.env.ASSEMBLYAI_API_KEY= transcription ·ELEVENLABS_API_KEY= voice-isolate + SFX + music ·GEMINI_API_KEY= thumbnails ·NOTION_TOKEN+NOTION_LONGS_PAGE_ID= the content tracker. YouTube uses OAuth, not a key:tools/yt_upload.py(upload) andtools/yt_stats.py(stats) authorize via a browser and cache tokens under.youtube/(git-ignored). One-time setup: seetools/yt_upload_SETUP.md, thenpython tools/yt_upload.py auth. - Notion is the tracker, the repo is the source of truth.
python tools/notion_sync.py videos/<project> --applypushes a project's row: properties fromvideos/<project>/notion.json(committed) and the page body from itsscript/script.md. It is a dry run without--apply, and it matches an existing row before creating one — ideas get parked in that database months early under a rough working name, so a blind create makes a duplicate. Never hand-edit the script in Notion; edit the repo and re-run with--resync.--listshows every row and its stage. - The Remotion registry is generated: after adding/renaming a shot,
cd remotion && npm run gen(the render/frames scripts do NOT run it themselves).npm run gendiscoverssrc/shots/**and writessrc/registry.gen.tsx+src/shots.manifest.json(both git-ignored). - Media rules:
media/library/is for CROSS-VIDEO reusable assets only (each with a catalog). Anything generated FOR ONE video (screen-caps, example art) goes inmedia/projects/<project>/, referenced asstaticFile('projects/<project>/x'). Reuse before you generate — check the catalogs. The face-reference kit for thumbnails ismedia/library/faces/— you supply your own images (git-ignored; see its README). No face kit, no face thumbnails. - Committed vs git-ignored: the repo carries the reproducible pipeline —
cuts.json, edit plans,timeline.json, transcripts, TSX shots, SFX/music plans, catalogs. Raw footage and master cuts NEVER go to git (videos/*/*.MP4,videos/*/reference/), and neither do rendered previews/outputs, audio work files, or rendered face thumbnails (.gitignorecovers all of it). The SFX/music libraryclips/*.mp3ARE committed (there is deliberately no blanket*.mp3ignore). - QA is not optional: render frames and READ them before declaring a shot done; run
verify_cut.pyon every cut render (it catches ghost speech + A/V drift); audit the SFX cue sheet before mixing. Scratch renders/frames go in a scratch dir, not the project. - The brand contract is three files.
brand.md,remotion/src/brand.ts,remotion/src/fonts.ts. Nothing errors when they disagree — the docs just stop describing the videos. Change them together, or run/brand-setup.
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
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.

