Adobe Edit Quick Cut

Produces 3 AI-edited sizzle reel variations from a source video, all at the same duration and style — giving the user options to pick from.

Surface note: The default flow below uses Adobe's MCP App widgets — the asset_add_file file picker (Step 2) and the side-by-side asset_preview_file preview (Step 7). Follow it as written. Only if a widget tool is not available on this surface (e.g. Codex), use the No-widget fallback attached to that step.

Tool Reference

Workflow

Step 0 — Initialize Adobe Tools

Call adobe_mandatory_init first. This returns file handling rules and tool routing guidance required for the rest of the workflow.

{ "skill_name": "adobe-edit-quick-cut", "skill_version": "2.1.0" }

Step 1 — Entitlement Check

Now that adobe_mandatory_init confirmed that the "Adobe for creativity" connector is live, check which tools are available through the "Adobe for creativity" connector by cross checking against the Tool Reference table above.

This also tells you which widgets this surface supports: if asset_add_file and asset_preview_file are available, follow the default flow (Steps 2 and 7 as written). If one is not available (e.g. Codex), use that step's No-widget fallback. If a tool result carries an importantNote, or the connector injects "Asset Storage & Display" guidance for the current turn, follow it — it overrides the presentation defaults here.

Step 2 — Open the File Picker

Open the picker immediately with this message:

"Let's create a punchy sizzle reel from your video. Start by selecting your file:"

asset_add_file()

Once the user selects a file, extract assetId (the CC asset ID) from the widget context.

video_create_quick_cut requires a CC asset ID (assetId), not presignedAssetUrl.

No-widget fallback (only if asset_add_file is unavailable on this surface, e.g. Codex) — don't ask the user to pick; get the assetId from where the file is. Staging a local file requires egress — check egress status from adobe_mandatory_init first; if egress is disabled and no picker is available on this surface, tell the user staging isn't possible here. Otherwise:

Step 3 — Confirm Upload

Once the file is selected, confirm with:

"Got it — filename is ready. Now let's set up your cut."

Then immediately present the Q&A form below.

Step 4 — Q&A Form (via AskUserQuestion)

Wait for the user's answers before proceeding; present the questions via AskUserQuestion (not plain text) so the user gets tappable buttons.

No-widget fallback (only if AskUserQuestion is unavailable, e.g. Codex) — ask the same labeled options as a plain-text message and wait for the user's typed reply. The option set is identical.

AskUserQuestion({
  questions: [
    {
      header: "Cut Length",
      question: "What kind of cut would you like? (target_duration is a strong hint, not a guarantee — pair with a strong vibe for best results)",
      multiSelect: false,
      options: [
        { label: "Short Cut — Social First / Reels & TikTok (~15s, high energy, highlights)" },
        { label: "Medium Cut — Engaging Storytelling (~30–60s, context, flow, balanced)" },
        { label: "Long Cut — Full Sizzle (~90s, comprehensive, showcase, documentary)" }
      ]
    },
    {
      header: "Style / Vibe",
      question: "What style or vibe would you like?",
      multiSelect: false,
      options: [
        { label: "Action & Energy" },
        { label: "Key Talking Moments" },
        { label: "Cinematic & Dramatic" },
        { label: "No Preference" }
      ]
    }
  ]
})

Wait for the user's selections before proceeding to Step 5.

Step 5 — Acknowledge and Run

Once the user answers, respond with:

"Got it — cut type, style vibe. Creating 3 variations at that length — let me preview them for you."

Map their answers to parameters:

Q1 duration map:

Q2 style map:

⚠️ The user_prompt is the primary lever for output quality — it does more work than target_duration. Pass the prompts below verbatim — abbreviating them weakens the output. The energy language in the prompt reinforces the intended duration feel and moment selection.

Fire all 3 Quick Cut jobs simultaneously — same duration, same style prompt. The AI will naturally select different moments on each run, giving the user 3 genuine options to pick from:

// Variation A
video_create_quick_cut({
  assetIds: [assetId],
  target_duration: <mapped_seconds>,
  user_prompt: "<mapped style prompt>"
}) // → statusId_A

// Variation B
video_create_quick_cut({
  assetIds: [assetId],
  target_duration: <mapped_seconds>,
  user_prompt: "<mapped style prompt>"
}) // → statusId_B

// Variation C
video_create_quick_cut({
  assetIds: [assetId],
  target_duration: <mapped_seconds>,
  user_prompt: "<mapped style prompt>"
}) // → statusId_C

Step 6 — Poll Until All 3 Complete

Poll all 3 in each round using quickCutPoll(statusId). Show progress % after each round. Repeat until all 3 return jobStatus: "completed".

Typical pattern: 0% → 7% → 78% → done. Usually 3–5 poll rounds.

Store:

• url_A — Variation A presignedAssetUrl
• url_B — Variation B presignedAssetUrl
• url_C — Variation C presignedAssetUrl

Step 7 — Preview All 3

asset_preview_file({
  assets: [
    { name: "Variation 1 — <cut_type> <style>.mp4", presignedAssetUrl: url_A, source: "acp" },
    { name: "Variation 2 — <cut_type> <style>.mp4", presignedAssetUrl: url_B, source: "acp" },
    { name: "Variation 3 — <cut_type> <style>.mp4", presignedAssetUrl: url_C, source: "acp" }
  ]
})

No-widget fallback (only if asset_preview_file is unavailable on this surface, e.g. Codex) — present all 3 variation URLs directly in the message, labeled by variation:

Variation 1 — <cut_type> <style>.mp4 → <url_A>
Variation 2 — <cut_type> <style>.mp4 → <url_B>
Variation 3 — <cut_type> <style>.mp4 → <url_C>

UI clients that render media URLs inline will show previews automatically. In Codex or other non-UI agents, download each to the workspace (curl -L -o variation_1.mp4 "<url_A>", etc.) and reference those local paths instead.

Step 8 — Deliver Summary + Download Prompt

After preview, present:

✅ 3 variations ready — same length, different moment selection. Pick your favorite!

| Variation | Cut Type | Style   | Target | Status |
| --------- | -------- | ------- | ------ | ------ |
| 1         | <type>   | <style> | ~<Xs>  | ✅      |
| 2         | <type>   | <style> | ~<Xs>  | ✅      |
| 3         | <type>   | <style> | ~<Xs>  | ✅      |

Note: actual durations may vary — Quick Cut selects the best moments rather than cutting to an exact second. The prompt vibe (e.g. "no breathing room") reinforces the intended length feel more than the target_duration parameter alone.

Then prompt:

"Which variation do you want to download, or would you like all 3? You can also rerun with a different style or cut type."

The videos are available for download directly from the preview above (or from the links above on a no-widget surface).

⚠️ Known Gap — Output Cannot Feed Downstream Video Tools

video_create_quick_cut returns a temporary presigned download URL, not a CC-stored asset ID. Tools like video_resize and media_enhance_speech require a CC asset ID as input.

You cannot chain Quick Cut → Resize or Quick Cut → Enhance Speech directly.

Workaround — if user wants to resize a Quick Cut output:

1. Tell the user: "Quick Cut outputs can't be passed directly to the resize tool — you'll need to download your preferred cut first, then re-upload it and I'll resize from there."
2. Let them download from the preview.
3. Re-ingest the downloaded file exactly as in Step 2 (asset_add_file(), or the no-widget staging fallback).
4. Once re-uploaded, run video_resize on the fresh assetId with their target dimensions.

When the user asks to resize or enhance a Quick Cut output, surface the limitation proactively — the chain is known to fail.

What Quick Cut Does NOT Support

• Content-aware cuts based on speech ("remove the parts where they repeat themselves")
• Trimming to specific timestamps ("cut from 0:30 to 1:15")
• Semantic understanding of dialogue

For these, recommend a manual video-editing workflow.

Error Handling

• video_create_quick_cut returns 403 (entitlement): Retrying does not help for a 403 entitlement — stop and surface the plan requirement. Respond with:

  "I was unable to create your quick cut.

  Why: Adobe Quick Cut isn't available on your current Adobe plan.

  *Options:

  • Upgrade your Adobe plan to one that includes Quick Cut.*
  • Manually trim your video using Adobe Premiere Rush or Premiere Pro.*

  Let me know how you'd like to proceed."

• Any tool call returns 401 (not authenticated): Ask the user to re-authenticate via Adobe OAuth and retry.
• StoryBuilderNoARoll: The most common error. Means Quick Cut detected no A-roll (talking head / primary camera footage) in the clip — only B-roll. The tool requires at least some dialogue or narration to anchor the story structure. Respond with:

  "This video appears to be B-roll only — scenery, action, or product shots without anyone speaking to camera. Quick Cut needs some talking-head footage to build a story around. Try uploading a video that includes someone speaking on camera, or a mix of interview + B-roll.

  If you only have B-roll, Adobe Premiere Rush or Adobe Express let you manually assemble a highlight reel without requiring dialogue." Retries repeat the same error regardless of duration/style; no workaround exists.

• Job fails with any other error on first attempt: Retry once with the same parameters. If it fails again, report and suggest re-uploading the source video.
• Stuck at same % for 5+ poll rounds: Inform user, suggest re-uploading the source video.
• User uploads an image by mistake: Detect from mediaType — if not video/*, say so and re-open the picker (or re-run the no-widget staging fallback).
• One of the 3 variations fails (but not all): Preview and deliver the successful ones, note the failure clearly. If all 3 fail with StoryBuilderNoARoll, apply the B-roll error response above.

Constraints

• Never pass a raw local filesystem path to video_create_quick_cut or any other video tool. Local files must reach Creative Cloud first — selected via the asset_add_file picker, or (no-widget fallback) staged via asset_initialize_file_upload → PUT → asset_finalize_file_upload; only the resulting assetId is valid.
• video_create_quick_cut requires assetId (CC asset ID), not presignedAssetUrl.