ADD THIS SKILL
paste into a new chat
Copy this into a blank Wideframe chat. It tells the agent to build the skill, what to confirm with you first, the steps to follow, and how to verify the result.
PROMPT
I'd like you to add a skill to Wideframe called "rough-cut-to-premiere".
Save the SKILL.md below exactly as written, then confirm I can run it by saying /rough-cut-to-premiere.
---
name: rough-cut-to-premiere
description: Turn a folder of raw footage into a first rough cut, delivered as a validated Premiere Pro project plus a review MP4. Use when someone says "rough cut", "first pass", "assemble this", "make a cut from my footage", or shares a folder of clips with a brief. This is the default path for anyone who arrives with footage rather than a project.
argument-hint: "[path-to-footage] [optional: what the video is for]"
---
# Rough Cut to Premiere
A first-time user with a folder of clips wants to see a cut they can open in Premiere within the session. This skill gets there in one pass: understand every clip, choose the keepers, build a sequence with sensible settings, add transitions where they help, export a project, validate it, and render something they can watch. It does not stop to ask ten questions first.
## Scope
A rough cut, not a finish. Story order, clean cuts, matched settings, simple transitions. No grading, no titles, no music unless it was supplied. Those come after the user has seen the first pass.
## Step 0: Resolve the input and set defaults
Accept a folder, a set of files, or a workspace. Read any brief the user gave (a sentence, a doc, a reference video). If no brief exists, infer the intent from the footage in Step 1 and state the assumption.
Defaults, used unless the user said otherwise:
- Sequence settings match the majority of the footage (frame rate and resolution from Step 1).
- Target length: 60 to 90 seconds for social or promo footage, 3 to 5 minutes for interview or event footage, or whatever the brief says.
- Keep source media where it is. Symlink into the workspace; never copy.
Ask nothing yet.
## Step 1: Understand every clip
1. `caltools media-info-batch` over the footage. Record frame rate, resolution, rotation, duration and audio presence per clip. Use rotation-corrected dimensions.
2. `caltools media-analyze-content-batch` on every clip for scene, shot type, subject and a one-line summary.
3. `caltools media-transcript-get-batch` on every clip with speech.
4. `caltools job-poll` until nothing is pending.
Evaluation: one row per file with fps, resolution, a content summary, and a transcript where there is speech. Unreadable files are listed, not silently dropped.
## Step 2: Choose the keepers
Score clips for usability (focus, exposure, stability) and for relevance to the brief. Reject only what is unusable; set aside borderline clips in a list rather than deleting anything. For speech, keep complete sentences and drop false starts and long pauses.
Write the plan as a short shot list in story order with the reason each clip is in. This is what the user will react to, so keep it readable.
Evaluation: every kept clip has a reason, every rejected clip has a measured reason, and the planned total is within the target length.
## Step 3: Build the sequence
1. `caltools sequence-create` at the majority frame rate and resolution.
2. `caltools sequence-media-set` so the sequence settings match the footage. If a minority of clips run at a different rate (for example 59.94 among 23.976), decide per clip: slow-motion intent gets `interpret_fps` to play at the sequence rate; normal-speed clips are conformed as is.
3. `caltools sequence-clips-add` in story order, then `sequence-clips-update` to trim handles to the planned in and out points.
4. `caltools sequence-transitions` only where a cut needs help: between scenes, into and out of the piece, across a time jump. Hard cuts everywhere else. State the count.
5. If music or a voiceover was supplied, add it on its own audio track. If not, leave audio as the source sound and say so.
Evaluation: render frames with `sequence-render-frames` at the start, one third, two thirds and end. No black frames, no wrong-aspect clips, no clip playing at the wrong speed.
## Step 4: Export, validate, render
1. `caltools sequence-export-premiere` to `<workspace>/renders/<name>_roughcut_v1.prproj` at the sequence settings.
2. `caltools prproj validate`. Zero dangling references and zero orphans, or fix and re-run.
3. `caltools prproj inspect` and reconcile the clip count against the sequence.
4. `caltools sequence-render-final` to `<workspace>/renders/<name>_roughcut_v1.mp4`, H.264, at the sequence resolution and frame rate. Confirm with `media-info-batch` that it plays and has audio.
Evaluation: validate is clean, clip counts match, the MP4 duration equals the sequence duration.
## Step 5: Report and ask the one question
Report in this order: the MP4 to watch, the .prproj to open, the shot list, the set-aside clips, and the assumptions you made (brief, length, settings). Then ask the one question that matters: "What should change for v2?"
Offer three follow-ups by name: tighten to a length, swap clips from the set-aside list, add captions (`caption-this`), or cut social versions (`social-clips`).
## Done when
- Every clip in the folder was read, and every keep or reject has a reason.
- The sequence settings match the footage and no clip plays at the wrong speed.
- The `.prproj` passed `prproj validate` with clip counts reconciled.
- A review MP4 exists, plays, and has audio.
- The user has been asked one question, about v2, and nothing else.
Download SKILL.md. If this workflow calls another named skill, add that recipe from the library too. Available tools and source access determine which steps can run.
RUN IT AGAIN
Once it’s saved, run it on the next project with /rough-cut-to-premiere, or just say:
Make a first rough cut from this folder and hand it back for Premiere.Run it on your own footage.
7-day free trial. Requires Apple Silicon.
RELATED SKILLS