shortform-rough-cut-ops

name	shortform-rough-cut-ops
description	Turn messy short-form video clips, creator intent, and optional format runbooks into an inspectable edit decision list and rough draft plan for CapCut or similar editors.
memory_tags	["domain:social-media","workflow:shortform-rough-cut","skill_role:operator","repo_boundary:tools","inputs:local-media","outputs:edit-decision-list","risk:medium"]

Shortform Rough Cut Ops

Trigger

Use when:

• the user wants help turning raw short-form video clips into a rough cut
• clips may be filmed out of narrative order and need to be mapped to story function
• the user has a recurring video format, hook pattern, or editing runbook
• the output should be an edit decision list, rough timeline, or CapCut draft plan

Do not use when:

• the user only wants viral-format research; use video-breakout-research-ops
• the user only wants transcription or archiving of a finished video
• the user wants final taste decisions fully automated with no human review

Inputs

• Required: local clip folder or clip manifest
• Required: intended story flow, even if rough
• Optional: format runbook, script, reference video, music track, beat markers, target duration, platform, and editor/draft tool

Core Principle

The edit follows the intended story, not filename order.

Raw short-form clips are often captured opportunistically: B-roll may belong earlier than it was filmed, A-roll may explain a previous moment, and the strongest ending may be buried in the middle of the folder. Treat chronological order as metadata, not truth.

Workflow

1. Establish the target format.
   • Ask for the intended hook, situation, progress beats, twist or ending, and target platform.
   • Capture target duration. For short-form talking/story formats, treat 55-70 seconds as a useful default window when the user has no stronger format-specific target.
   • If the user provides a recurring format runbook, read it before classifying clips.
   • If no runbook exists, create a temporary working structure from the user's rough description.
2. Inventory the source clips.
   • Capture path, filename, duration, orientation, resolution, audio presence, and visible timestamp metadata when available.
   • Create or request a contact sheet when visual selection matters.
   • Do not move, overwrite, or rename raw media unless explicitly asked.
3. Transcribe and tag clips.
   • Transcribe speech clips with available local tools.
   • Mark clips without useful speech as visual/B-roll candidates.
   • Preserve exact phrases that identify anchor moments, transitions, progress updates, jokes, or twists.
   • For voiceover or A-roll-driven edits, use a staged A-cut workflow: first choose the spoken story ranges for the hook, setup, progression beats, twist, and ending; then run a separate tightening pass for dead air, long silences, false starts, repeated takes, filler-only fragments, abandoned sentences, duplicate ideas, and weak tangents.
   • Treat narration cleanup as iterative: rough story selection first, cleanup second, compression third. Do not use B-roll, captions, effects, or music to hide an unresolved spoken story.
   • Do not speed narration up by default; only apply speed changes when the user asks for that style or a format runbook explicitly calls for it.
   • Keep the cleaned speech timeline as an edit decision list of source ranges and removed ranges, not only as a baked audio export, so captions, overlays, effects, and B-roll can be retimed to the final narration.
4. Classify by story function. Use categories such as:
   • hook / anchor
   • setup
   • A-roll explanation
   • progress update
   • B-roll / montage
   • proof / detail
   • twist / ending
   • uncertain
5. Build an edit decision list before drafting.
   • Sequence selected clips against the intended story flow.
   • Include rough trims, rationale, confidence, and alternatives.
   • Flag whether the planned sequence is likely under or over the target duration.
   • Prefer two options when ambiguity is real: story-forward and montage-forward.
   • When speech drives the edit, lock the cleaned voiceover/A-roll timeline first, then place B-roll bridges, captions, overlays, sound effects, and proof assets against that cleaned timeline.
   • Surface removed ranges and alternate takes before drafting when the narration decision is ambiguous.
6. Plan editor automation only after structure exists.
   • Use CapCut or another draft writer as the timeline assembly layer, not the editing brain.
   • Generate a draft only from selected clips and rough trims.
   • Keep final pacing, text, effects, comedic timing, and taste-heavy choices human-reviewed.

Output Contract

Return:

• clip inventory summary
• transcript/tag table or pointer to it
• clip classification table
• edit decision list with rough timing and rationale
• cleaned speech / voiceover range list when narration drives the edit
• unresolved questions or weak clips
• draft generation command or next manual editor steps, when applicable

Guardrails

• Do not dump every clip into a timeline unless the user explicitly asks for assembly-only import.
• Do not hard-code one creator's format into the skill. Put format-specific rules in reference runbooks.
• Do not assume a strong hook or twist exists just because the format expects one.
• Do not invent clip contents. If visual inspection or transcription is missing, mark the gap.
• Do not copy private footage into public repos.

References

Open only when needed:

• format runbook structure: references/format-runbooks.md