Computer-Use Agent Taxonomy中文
R06.04.01.03 · TASK CLASS

Multicam & cutting to music

Sync and switch multicam angles and cut on musical beats.

20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.

Typical applications
Premiere Pro · Final Cut Pro · DaVinci Resolve
Class GUI-need label
G2 · GUI-essential

Task proposals · 20

Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.

R06.04.01.03-T01 Cut a three-angle panel discussion to follow the speaker

Original instruction

Premiere Pro project fixtures/panel.prproj has three camera clips (wide, host, guest) with matching timecode. Create a timecode-synced multicam source sequence, then cut sequence 'Panel_Cut' so the active angle shows the current speaker on ≥ 85% of frames, using the wide for overlaps, with no shot shorter than 2 s.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click av · Audio/video over time vis · Visual judgment
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
4 / 5

Proposed verifier

Parse .prproj: multicam source synced by timecode (angle offsets 0); per-frame active angle vs GT diarization map (host/guest/overlap→wide) agreement ≥ 85%; min shot length ≥ 2 s; total duration equals source. ExtendScript/UXP panels are disabled, and .prproj writes by other processes or new XML/EDL files on the VM are detected, so edits must be made in Premiere.

Proposed partial credit

20% multicam synced; 50% speaker agreement (scaled from 0.6 agreement); 20% min shot length; 10% full duration.

R06.04.01.03-T02 Swap four flagged angle cuts to the wide without moving cuts

Original instruction

In DaVinci Resolve timeline 'Talk_MC' (multicam clips), the producer flagged 4 shots with blue markers. Switch those 4 shots to angle 1 (Wide) without changing any cut point or other angles.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click key · Shortcuts
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Timeline export: the 4 marked shots now reference angle 1; record in/out of all shots unchanged; all other shots keep their angles. Resolve scripting (Console and external API) is disabled and any XML/EDL/OTIO/AAF file written on the VM is detected, so the timeline cannot be imported from a script.

Proposed partial credit

60% four switches (15 points each); 25% cut points unchanged; 15% others untouched.

R06.04.01.03-T03 Flatten a multicam edit and export FCPXML for conform

Original instruction

Final Cut Pro project 'Keynote_MC' uses a 4-angle multicam clip. Duplicate the project as 'Keynote_Flat', break apart/flatten the multicam so each shot references its original angle media with the same in/out, and export out/keynote_flat.fcpxml.

Platforms
mac · macOS
Applications
Final Cut Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click key · Shortcuts
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
2 / 5

Proposed verifier

FCPXML: no mc-clip elements in 'Keynote_Flat'; each asset-clip references the angle's media matching the original active angle per shot; source ranges and record positions equal the multicam version; original project unchanged. Final Cut Pro has no scripting API; the FCPXML must be written by the FCP process (writes logged by process) and match 'Keynote_Flat' in the library.

Proposed partial credit

40% no multicam references; 40% correct media and ranges per shot; 20% original preserved and export present.

R06.04.01.03-T04 Cut 20 B-roll shots on the beat of a 120 BPM track

Original instruction

Resolve project 'City_Beat' has a 120 BPM music track (first downbeat at 00:00:00:12, 24 fps) and 30 B-roll clips. Build timeline 'Beat_20' with 20 shots where every cut lands within ±1 frame of a beat and each shot lasts at least one bar (4 beats).

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
drag · Drag & drop av · Audio/video over time nav · Pan/zoom/scroll
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

OTIO: 20 video shots; each cut frame within ±1 of the GT beat grid (beats every 12 frames from frame 12); all shots ≥ 48 frames; music unaltered; no repeated source range. Resolve scripting (Console and external API) is disabled and any XML/EDL/OTIO/AAF file written on the VM is detected, so the timeline cannot be imported from a script.

Proposed partial credit

55% cuts on beat (scaled); 20% shot count; 15% min bar length; 10% music intact.

R06.04.01.03-T05 Shorten a music bed to end exactly on the last frame

Original instruction

Premiere Pro sequence 'Promo_45' is 45:00 (25 fps) but its music runs 58 s. Make a musical edit at a bar boundary so the music's natural ending lands on the last frame of the sequence, with a crossfade ≤ 6 frames at the edit and no audible jump.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
drag · Drag & drop av · Audio/video over time
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
4 / 5

Proposed verifier

Render audio: music ending (GT final hit) at 44:24 ± 1 frame; edit point at a GT bar boundary on both sides (same beat phase); spectral-flux spike at the edit below 1.5× local median; no silence gap. ExtendScript/UXP panels are disabled, and .prproj writes by other processes or new XML/EDL files on the VM are detected, so edits must be made in Premiere.

Proposed partial credit

35% ending on last frame; 35% edit on bar boundary; 20% no audible jump; 10% crossfade ≤ 6 frames.

R06.04.01.03-T06 Speed-ramp three actions onto musical downbeats

Original instruction

Final Cut Pro project 'Skate_Edit' has 3 skate tricks marked with markers (board pop frames) and a music track with downbeat markers. Apply speed ramps (between 50% and 200%) so each pop frame lands exactly on its assigned downbeat, keeping the clips' start frames fixed.

Platforms
mac · macOS
Applications
Final Cut Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
drag · Drag & drop av · Audio/video over time geom · Precise geometry
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
4 / 5

Proposed verifier

FCPXML timeMap: for each clip the source pop frame maps to its downbeat record frame (±1); speeds within 50–200%; clip record in unchanged; ramp keyframes smooth (no instantaneous speed jump > 50%). Final Cut Pro has no scripting API and any FCPXML written on the VM is detected (file-write log), so the edit must be made in the GUI, not imported.

Proposed partial credit

60% three pops on downbeats; 20% speed limits; 10% starts fixed; 10% smooth ramps.

R06.04.01.03-T07 Sync four cameras by audio waveform into a multicam clip

Original instruction

Resolve media pool bin 'Day1_Band' has 4 camera clips without common timecode and a separate board audio WAV. Create a multicam clip synced by waveform with the board audio as the audio angle; the offsets must be frame accurate.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click av · Audio/video over time
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Inspect multicam via timeline of angles: each angle's offset vs board audio within 1 frame of GT cross-correlation offsets; all 4 video angles + board audio present. Resolve scripting (Console and external API) is disabled and any XML/EDL/OTIO/AAF file written on the VM is detected, so the timeline cannot be imported from a script.

Proposed partial credit

80% four angle offsets (20 points each); 20% board audio angle used.

R06.04.01.03-T08 Cut a five-camera concert song to musical events

Original instruction

Premiere Pro project fixtures/live_song.prproj: 5 synced angles (wide, vocal, guitar, drums, crowd) and board mix. Cut 'Song3_Cut' so the guitar angle is on during the solo bars, drums on the 3 drum fills, vocal during verses, crowd for the final chorus downbeat; cuts on beats (±1 frame), each shot ≥ 1 bar.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
av · Audio/video over time vis · Visual judgment click · Point & click
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

Per-frame angle vs GT section map (solo, fills, verses, final chorus downbeat window): agreement ≥ 85% in those sections; cuts within ±1 frame of GT beat grid; min shot ≥ 1 bar; board audio unedited. ExtendScript/UXP panels are disabled, and .prproj writes by other processes or new XML/EDL files on the VM are detected, so edits must be made in Premiere.

Proposed partial credit

25% solo on guitar; 15% fills on drums; 30% verses on vocal and crowd shot; 30% cuts on beat with min length.

R06.04.01.03-T09 Fix six wrong switches in an auto-generated podcast multicam

Original instruction

Premiere Pro sequence 'Pod_Auto' was cut by Auto Multi-Camera Editing for a 2-person podcast (cams A, B, wide). Review it and fix the 6 places where the angle shows the non-speaker for more than 2 s; leave correct sections untouched.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
av · Audio/video over time vis · Visual judgment click · Point & click
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

Per-frame speaker agreement in the 6 GT error windows ≥ 90%; outside them the angle sequence equals the original (≥ 98% frames identical); min shot ≥ 1 s. ExtendScript/UXP panels are disabled, and .prproj writes by other processes or new XML/EDL files on the VM are detected, so edits must be made in Premiere.

Proposed partial credit

60% six fixes (10 points each); 30% rest untouched; 10% min shot.

R06.04.01.03-T10 Sync mixed-frame-rate wedding cameras and cut the first dance

Original instruction

Resolve project 'Wedding_Dance': cam A 23.976, cam B 29.97, phone 30.00 fps (drifts) plus a recorder WAV. Sync all to the WAV (fix the phone's drift by retiming), build a multicam, and cut 'First_Dance' following input/dance_notes.md (wide on spins, close on kiss, phone for the crowd).

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
av · Audio/video over time vis · Visual judgment drag · Drag & drop
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Sync error vs WAV ≤ 1 frame at both start and end of each angle (drift corrected); angle choice agrees with GT note windows ≥ 80%; cuts not inside GT spin moves; full song length covered. Resolve scripting (Console and external API) is disabled and any XML/EDL/OTIO/AAF file written on the VM is detected, so the timeline cannot be imported from a script.

Proposed partial credit

35% sync incl. drift fix; 40% notes followed; 15% no cuts mid-spin; 10% full coverage.

R06.04.01.03-T11 Sync lip-sync performance takes to a music-video playback track

Original instruction

Final Cut Pro event 'MV_Takes' holds 7 performance takes shot to playback (scratch audio) and the master song. Sync each take to the master song with frame accuracy, then build project 'MV_Perf' alternating takes every 2 bars across the first chorus.

Platforms
mac · macOS
Applications
Final Cut Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
av · Audio/video over time drag · Drag & drop
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

Per-take offset vs master within 1 frame (scratch-audio cross-correlation GT); project: chorus covered with cuts at 2-bar boundaries (±1 frame), consecutive shots from different takes, master audio unaltered. Final Cut Pro has no scripting API and any FCPXML written on the VM is detected (file-write log), so the edit must be made in the GUI, not imported.

Proposed partial credit

50% seven takes synced; 30% 2-bar alternation; 20% master audio intact.

R06.04.01.03-T12 Cut a montage on snare hits with a faster chorus rhythm

Original instruction

Premiere Pro project fixtures/sports_reel.prproj has a 96 BPM track. Cut 'Reel_v1' so the first 16 bars have a cut every 2 beats landing on snare backbeats, and the chorus (bars 17–24) cuts on every beat; use any of the 60 clips without repeats.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
drag · Drag & drop av · Audio/video over time
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

Cut frames vs GT snare onsets for bars 1–16 (every backbeat) and all beats for 17–24, tolerance ±1 frame; no repeated source range; music unedited. ExtendScript/UXP panels are disabled, and .prproj writes by other processes or new XML/EDL files on the VM are detected, so edits must be made in Premiere.

Proposed partial credit

45% verse cuts on snares; 35% chorus cuts on beats; 10% no repeats; 10% music intact.

R06.04.01.03-T13 Fix one angle's sync in the Angle Editor and recut

Original instruction

Final Cut Pro multicam clip 'Lecture_MC' has angle C (slides capture) 12 frames late. Correct it in the Angle Editor, then in project 'Lecture_Edit' make sure every slide-change cut lands within 2 frames of the slide change visible in angle C.

Platforms
mac · macOS
Applications
Final Cut Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
drag · Drag & drop av · Audio/video over time vis · Visual judgment
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

FCPXML: angle C offset corrected (within 1 frame of GT); project cuts to angle C occur within 2 frames of GT slide-change times; other angle cuts unchanged. Final Cut Pro has no scripting API and any FCPXML written on the VM is detected (file-write log), so the edit must be made in the GUI, not imported.

Proposed partial credit

45% angle sync fixed; 40% slide-change cuts aligned; 15% other cuts unchanged.

R06.04.01.03-T14 Live-switch a multicam during playback to follow the speaker

Original instruction

In DaVinci Resolve's multicam viewer, live-switch the 3-angle debate clip 'Debate_MC' during real-time playback (no stopping) to follow the speaker, then refine at most 10 switches afterwards. Save the timeline as 'Debate_Live'.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
rt · Real-time control av · Audio/video over time key · Shortcuts
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Timeline: speaker agreement ≥ 80% vs GT diarization; switch events log (Resolve undo history) shows ≥ 70% of cuts made during a single playback pass and ≤ 10 post-hoc adjustments; min shot ≥ 1 s. Resolve scripting (Console and external API) is disabled and any XML/EDL/OTIO/AAF file written on the VM is detected, so the timeline cannot be imported from a script.

Proposed partial credit

50% speaker agreement; 30% cuts made live in one pass; 20% minimal refinements and shot minimum.

R06.04.01.03-T15 Mark every downbeat for 64 bars and snap a cut to them

Original instruction

In Premiere Pro, add sequence markers on every downbeat of the first 64 bars of 'track_08.wav' (tempo drifts slightly; no fixed BPM) in sequence 'Lyric_Edit', then place the 16 lyric-card clips so each starts on the downbeat of its bar from input/lyric_bars.csv.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
av · Audio/video over time key · Shortcuts drag · Drag & drop
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

.prproj markers: 64 markers each within ±1 frame of GT downbeats (from annotated tempo map); 16 clips start on the listed bars' downbeats (±1 frame). Scripted beat detection allowed if markers end up in the sequence.

Proposed partial credit

50% downbeat markers (scaled); 40% lyric cards placed; 10% marker count exact.

R06.04.01.03-T16 Hide jump cuts by switching angles at content edits

Original instruction

Resolve timeline 'CEO_Interview' cuts a two-camera interview (A medium, B close) and has 5 content edits that show as jump cuts on angle A. Rework so every content edit coincides with an angle switch (A↔B), keeping the content edits themselves unchanged.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click vis · Visual judgment av · Audio/video over time
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

At each of the 5 GT content-edit frames, the angle before differs from the angle after; audio content identical to original (ASR + waveform hash); no new jump cuts introduced (same-angle content edits = 0). Resolve scripting (Console and external API) is disabled and any XML/EDL/OTIO/AAF file written on the VM is detected, so the timeline cannot be imported from a script.

Proposed partial credit

60% five jump cuts hidden; 25% audio content unchanged; 15% no new jump cuts.

R06.04.01.03-T17 Make a 15-second teaser cut on beats with a shortened music edit

Original instruction

Final Cut Pro project 'Album_Teaser' must be exactly 15:00 (30 fps). Edit the 2-min song down with one musical edit at a bar boundary, cut 8–12 shots from the 'Studio' event on beats (±1 frame), and end on a downbeat with the song's final chord.

Platforms
mac · macOS
Applications
Final Cut Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
drag · Drag & drop av · Audio/video over time
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

FCPXML: duration 15:00; 8–12 video shots; cuts within ±1 frame of GT beats; music edit at bar boundary (beat phase match); final frame within 1 frame of the final chord onset + its hold. Final Cut Pro has no scripting API and any FCPXML written on the VM is detected (file-write log), so the edit must be made in the GUI, not imported.

Proposed partial credit

20% duration; 35% cuts on beats; 25% musical edit; 20% ending.

R06.04.01.03-T18 Control: generate the T04 beat-cut timeline with a script

Original instruction

Control pair of T04 (CLI half). Detect beats in fixtures/city_beat.wav with librosa (or provided grid), then generate out/beat_20.otio with 20 shots from fixtures/broll/ whose cuts fall within ±1 frame of beats and each ≥ 1 bar, and render out/beat_20.mp4 with ffmpeg.

Platforms
lin · Linux mac · macOS win · Windows
Applications
Python · OpenTimelineIO · ffmpeg
Interfaces
term · Terminal
GUI need · task setting
CL · CLI-sufficient
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Same checks as T04 on OTIO and the render. Paired with T04 to measure GUI timeline cutting against scripted assembly.

Proposed partial credit

55% cuts on beat; 20% shot count; 15% min length; 10% render valid.

R06.04.01.03-T19 Conform a six-camera event edit to the director's program feed

Original instruction

Premiere Pro project fixtures/awards.prproj has 6 ISO camera recordings and a low-res program feed of the live-switched show. Build a multicam and recreate the director's cuts in 'Awards_Conform' so each shot matches the program feed's angle, then upgrade two shots flagged in notes to better ISO angles.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment av · Audio/video over time click · Point & click nav · Pan/zoom/scroll
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

Per-frame angle vs GT program angle (from image matching) agreement ≥ 95% outside the two flagged windows; cut frames within ±1 of program cuts; flagged windows use the angle named in notes. ExtendScript/UXP panels are disabled, and .prproj writes by other processes or new XML/EDL files on the VM are detected, so edits must be made in Premiere.

Proposed partial credit

50% angle agreement; 25% frame-accurate cut points; 25% two flagged upgrades.

R06.04.01.03-T20 Cut a four-minute dance performance to phrase and choreography rules

Original instruction

Resolve project 'Ballet_Gala': 4 synced angles of a 4-min piece and input/choreo_notes.md (formations, solos, 3 lifts, 8-count phrases). Cut 'Gala_Final': wide during group formations, close on the soloist during solos, cuts only on 8-count phrase starts (±2 frames), never during a lift.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
av · Audio/video over time vis · Visual judgment click · Point & click
Work product
WP05 · Time-based media
Conditions
long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

Per-frame angle vs GT section map (formation→wide, solo→soloist close) ≥ 85%; all cuts within ±2 frames of GT phrase starts; zero cuts inside GT lift windows; full duration covered. Resolve scripting (Console and external API) is disabled and any XML/EDL/OTIO/AAF file written on the VM is detected, so the timeline cannot be imported from a script.

Proposed partial credit

40% section angle agreement; 30% cuts on phrases; 20% no cuts in lifts; 10% full coverage.

Mapped research references

Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.

In this workflow family