agentleFS
Sign inStart free
kathan3009kathan3009/video-studio/SKILL.md

video-studio

Skill2 starsChanged 6 months agoMITView on GitHub

Edit videos with Claude Code. End-to-end: write scripts, generate voiceovers, record product screens with Playwright, transcribe locally with Whisper, plan cuts with word-level precision, render with ffmpeg, self-evaluate, publish. Free and local by default; OpenAI TTS opt-in for voiceovers.

9 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

8 rules. 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 3

    Distinctive

    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:

    When the user makes a request, always:

    line 24
  2. Very specific:

    Read the corresponding formats/<name>.md — its rules override defaults below.

    line 33
  3. Very specific:

    Check for product memory: Before deciding what to film, scan ~/.claude/projects/<project-slug>/memory/MEMORY.md for a *-memory.md …

    line 35
  4. Very specific:

    Ask → confirm → execute → self-eval → persist: Never start cutting without strategy approval.

    line 43
  5. Distinctive:

    Crop-to-fill on every panel: Never letterbox unless the user explicitly asks.

    line 92
  6. Distinctive:

    Self-eval has to pass before you say done: If after 3 fix loops it still fails, report the failures to the user explicitly — don't claim …

    line 95
  7. Very specific:

    project.md persists across sessions: Read it on startup; write to it after key decisions (chosen format, voice, music, style overrides).

    line 98
  8. Very specific:

    If the user hasn't installed yet, point them at install.md.

    line 110

“Common” is measured: kept by at least 6 in 10 of the 8 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

video-studio

Open-source, free-by-default video pipeline. One repo, one skill, many formats.

How to use this skill

When the user makes a request, always:

  1. Pick a format profile based on the request:

    • reel — 9:16 vertical (Instagram Reels, TikTok, Shorts)
    • podcast_clip — multi-speaker 9:16 or 1:1 highlight
    • tutorial — 16:9 fullscreen + corner head-cam, long-form how-to
    • explainer — 16:9 annotated product demo, mid-form
    • square — 1:1 for LinkedIn / Twitter / feed posts

    Read the corresponding formats/<name>.md — its rules override defaults below.

  2. Check for product memory. Before deciding what to film, scan ~/.claude/projects/<project-slug>/memory/MEMORY.md for a *-memory.md describing the product/platform being recorded (URL map, danger buttons, auth flow). Read it before planning shots.
  3. Always read helpers/ — that's where the editing scripts live. Call them by path; don't re-invent.
  4. Ask → confirm → execute → self-eval → persist. Never start cutting without strategy approval. Always run self-eval before showing the user. Persist session state to project.md in the working directory.

Pipeline (canonical order)

1. Script ──┬─ TTS  (helpers/tts.py)
            └─ user records ──┐
                              ▼
2. Record (scripts/record_*) ──> 3. Clean audio (helpers/clean_audio.py)
                                              │
                                              ▼
                          4. Transcribe (helpers/transcribe.py)
                                              │
                                              ▼
                          5. Refine SRT (helpers/refine_srt.py)
                                              │
                                              ▼
                          6. Pack takes (helpers/pack_takes.py) → takes_packed.md
                                              │
                                              ▼
                          7. LLM reasons → EDL (helpers/edl_planner.py)
                                              │  • filler word cuts
                                              │  • silence trimming
                                              │  • multi-take selection
                                              │  • b-roll matching
                                              │  • color grade per segment
                                              │  • zoom & cursor planning
                                              │  • 30ms fade placement
                                              │  • caption chunking
                                              │  • animation triggers
                                              ▼
                          8. Animations (helpers/animation/) ─── parallel sub-agents
                                              │
                                              ▼
                          9. Render (helpers/compose.py + formats/<name>.md)
                                              │
                                              ▼
                          10. Self-eval (helpers/self_eval.py) ──┐ max 3 fix loops
                                              │                  │
                                              ▼ ◄────────────────┘
                          11. Publish bundle (helpers/publish.py)

Defaults you must not silently change

  • Free, local, offline by default. Whisper, ffmpeg, Playwright, demucs — all local. Paid APIs (OpenAI TTS, ElevenLabs) only when the user has set the env var AND opted in.
  • Crop-to-fill on every panel. Never letterbox unless the user explicitly asks.
  • Brand-name refinement on every transcript before captions burn. Never ship captions built from a raw Whisper SRT.
  • Self-eval has to pass before you say done. If after 3 fix loops it still fails, report the failures to the user explicitly — don't claim success.
  • Output goes to <cwd>/edit/. Sources stay untouched.
  • project.md persists across sessions. Read it on startup; write to it after key decisions (chosen format, voice, music, style overrides).

Memory hooks

  • Per-project session state: <cwd>/project.md
  • Per-product mind-maps: ~/.claude/projects/<slug>/memory/<product>-memory.md
  • Style presets: ~/.video-studio/styles/*.yaml (user-saved overrides)
  • Music library: ~/.video-studio/music/ (indexed by mood + BPM)

Setup

If the user hasn't installed yet, point them at install.md. If they have but something's broken, common failures live in each formats/*.md "Lessons learned" section.

Where to find more

  • formats/*.md — per-format authoritative rules
  • helpers/*.py — Python editing brain
  • scripts/* — Node/shell capture scripts
  • templates/*.md — script teleprompter templates
  • examples/sentinel_reel/ — flagship worked example

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.

Check your setup, free