clean-recording

name	clean-recording
description	Clean up a raw recording in Descript — remove filler words, cut silences, add captions — and export the result to out/. Use when the user has a recording in raw/ to polish.
disable-model-invocation	true

Clean a recording with Descript

Input: $ARGUMENTS = path to a recording (e.g. raw/episode1.mp4). If empty, ask which file in raw/ to use.

Heads up to the user before importing: the Descript MCP uploads the file to Descript's servers (it leaves the machine). Mechanical-only edits (trim/concat) can be done locally with ffmpeg instead — offer that if they don't want an upload.

Steps

1. Confirm the file exists at the given path.
2. Import into Descript via the Descript MCP (import_media):
   • Local file → use direct upload (content_type + file_size → PUT to the returned upload_url).
   • If the user gave a URL / Drive / Dropbox link, pass it as url.
   • Include add_compositions so it lands on the timeline.
   • Use wait_for_job to wait for import to finish; capture the project + composition id via get_project.
3. Run cleanup with prompt_project_agent on that composition. Default cleanup:
   • remove filler words ("um", "uh", "like", "you know")
   • shorten/remove long silences and dead air
   • remove obvious false starts
   • add captions Ask the user if they want anything different before running.
   • Background: the user films themselves and wants their real background replaced with an on-brand one. Offer to remove the background and drop in the branded Background (render Background-Light / Background-Dark via /render first, then use it as the background image in Descript's background-removal feature).
4. Wait for the agent job (wait_for_job), then export the cleaned result (export_transcript for text, or trigger a video export) and save the video to out/<original-name>-clean.mp4.
5. Report what changed (filler words removed, time trimmed) and the output path.

Next step

Offer to add on-brand animations (/add-animation) or render an intro (/render).