← ClaudeAtlas

editlisted

Edit a pre-recorded talking-head video into a finished, brand-themed cut — overlay beats timed to the spoken words, speaker zoom punch-ins, preview/final render ladder. TRIGGER when the user provides recorded footage and asks to edit/enhance/caption it.
marsmike/feinschmiede · ★ 3 · Data & Documents · score 76
Install: claude install-skill marsmike/feinschmiede
# Edit — pre-recorded footage pipeline Takes a finished recording (the user's voice and cut are untouched) and adds a planned layer of visuals. You author **data** (`edit_plan.json`), never React code — the shared engine in `edit-engine/` renders it. ## Workflow 1. **Probe** — `ffprobe` aspect + duration (portrait <30s = short; landscape = intro/longform). 2. **Transcribe** — `feinschnitt edit transcribe <video>` → `words.json` in the workdir (`feinschnitt edit workdir <video>` prints the path). Read it; it is the timing source of truth. 3. **Author the plan** — write `edit_plan.json` next to the video. Schema: `skills/edit/schema/edit-plan.schema.json`. The template library: | kind | class | required fields | use for / NOT for | |---|---|---|---| | hook_title | overlay | title | cold-open lockup at 0.0s / never mid-video | | word_pop | overlay | items[text+appear_sec] | enumerations + emphasis lines / not card-styled lists | | stat_punch | takeover | value, caption | THE hero number / not for every number mentioned | | quote_pull | takeover | quote_text | the takeaway line; anchor must cover the spoken quote (cps + 2s dwell are computed) / not for the speaker's own passing remarks | | static | takeover, image | image_path | hero screenshots/photos shown full-frame (contain, never cropped) / not the default for b-roll | | image_card | overlay, image | image_path | DEFAULT for b-roll images — speaker stays visible / not for text-heavy scree