Computer-Use Agent Taxonomy中文
R06.04.02.05 · TASK CLASS

Footage logging & continuity notes

Watch footage and log shots, takes, timecodes and continuity issues accurately.

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

Typical applications
DaVinci Resolve · Premiere Pro · spreadsheets
Class GUI-need label
HY · Hybrid

Task proposals · 20

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

R06.04.02.05-T01 Log forty clips with shot size, movement, slate and circled takes

Original instruction

Resolve project 'Day5' bin 'Rushes' has 40 clips. Fill out/day5_log.csv with columns clip, scene, take (read from the slate), shot_size (ECU/CU/MCU/MS/WS), camera_move (static/pan/tilt/dolly/handheld) and circled (Y if the director says 'print' or 'circle that' on the audio).

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment av · Audio/video over time type · Text & forms
Work product
WP02 · Records/table
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

CSV vs GT per clip: scene/take exact; shot_size exact or adjacent class (half credit); camera_move exact; circled exact. Score per column accuracy. GT built from annotated slates, framing and transcribed director calls.

Proposed partial credit

30% scene/take; 25% shot size; 20% camera move; 25% circled flags.

R06.04.02.05-T02 Find continuity errors across six takes of a dinner scene

Original instruction

Premiere Pro project fixtures/dinner.prproj has 6 takes (3 setups) of scene 7. Watch them and log every continuity discrepancy relative to the master take 7A-2 (glass levels, napkin position, jacket button, cigarette length, hand used) in out/continuity_sc7.csv: take, timecode, item, description.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment nav · Pan/zoom/scroll type · Text & forms
Work product
WP02 · Records/table
Conditions
licensed · Licensed pro app multiapp · Multi-app
Proposed difficulty
4 / 5

Proposed verifier

Match rows to 9 GT discrepancies (same take, item category, TC within 2 s): recall and precision; description must name the changed property (keyword match against GT synonyms). The media folder is not accessible from the agent's shell (no ffmpeg, OpenCV or vision API), so footage can only be reviewed in Premiere's player.

Proposed partial credit

55% recall of 9 discrepancies; 25% precision; 20% correct property described.

R06.04.02.05-T03 Record script-line coverage for twelve takes

Original instruction

Resolve project 'Sc14' has 12 takes across 4 setups; input/sc14_script.pdf has 18 numbered dialogue lines. For each take, log first and last line covered and whether lines were dropped or ad-libbed, in out/sc14_lining.csv.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
av · Audio/video over time type · Text & forms
Work product
WP02 · Records/table
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Per take: first/last line numbers exact vs GT (from ASR alignment); dropped lines list F1 ≥ 0.8; ad-lib flag exact. Accuracy averaged over 12 takes.

Proposed partial credit

50% first/last lines; 30% dropped lines; 20% ad-lib flags.

R06.04.02.05-T04 Build a timecoded topic log of a 30-minute interview

Original instruction

Premiere Pro clip 'Int_Farmer_A' (30 min) has no transcript. Watch/listen and write out/farmer_log.xlsx with one row per topic segment: TC in, TC out, topic (from input/topic_list.txt), one-line summary. Segments must cover the whole clip.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
av · Audio/video over time type · Text & forms multi · Cross-app
Work product
WP02 · Records/table
Conditions
licensed · Licensed pro app multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

Segment boundaries vs GT topic changes (within 10 s) F1 ≥ 0.8; per-second topic label agreement ≥ 85%; coverage contiguous 00:00–30:00; summaries non-empty (LLM rubric checks factual match to segment, 10%).

Proposed partial credit

35% boundaries; 40% topic agreement; 15% full coverage; 10% summaries (LLM rubric).

R06.04.02.05-T05 Fill clip metadata fields for fifty clips in Resolve

Original instruction

In Resolve project 'Doc_Island', fill the Metadata panel fields for the 50 clips in bin 'Day1': Scene, Shot, Take, Good Take, Description (≤ 10 words) and Keywords (from input/keywords.txt), based on viewing each clip and the slate.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment type · Text & forms click · Point & click
Work product
WP02 · Records/table
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

API read of clip properties: Scene/Shot/Take exact vs GT; Good Take vs GT circled; keywords F1 ≥ 0.7 vs GT; description non-empty ≤ 10 words. Fields must be stored in the project (CSV import of metadata also allowed).

Proposed partial credit

35% scene/shot/take; 20% good take; 30% keywords; 15% descriptions present.

R06.04.02.05-T06 Mark animal appearances in wildlife footage

Original instruction

Premiere Pro clip 'Waterhole_Cam3' (20 min, fixed camera) contains several animal visits. Add a clip marker spanning each visit (in when the animal enters frame, out when it leaves), named with the species from input/species.txt. Save the project.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment nav · Pan/zoom/scroll click · Point & click
Work product
WP10 · Annotations
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

.prproj markers vs 14 GT visits: temporal IoU ≥ 0.7 counts a match; species name exact; precision ≥ 0.85; markers are range markers (duration > 0).

Proposed partial credit

50% visits found; 25% species correct; 15% precision; 10% spans with durations.

R06.04.02.05-T07 Log every goal, foul and substitution in a match recording

Original instruction

From fixtures/u19_match.mp4 (90 min, single wide camera with scoreboard), produce out/match_log.xlsx with event type (goal, foul, yellow card, substitution), match clock, file timecode and team, for every event.

Platforms
mac · macOS win · Windows
Applications
DaVinci Resolve · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment nav · Pan/zoom/scroll type · Text & forms
Work product
WP02 · Records/table
Conditions
long · Long-horizon multiapp · Multi-app
Proposed difficulty
4 / 5

Proposed verifier

Rows vs GT events: match by type and file TC within 5 s; recall and precision per type; match clock exact ± 1 min; team correct.

Proposed partial credit

30% goals/cards recall; 30% fouls/substitutions recall; 20% precision; 20% clock/team correctness.

R06.04.02.05-T08 Tag B-roll with a controlled vocabulary in a spreadsheet

Original instruction

Folder fixtures/broll_city/ has 60 short clips. In out/broll_tags.xlsx, give each clip location (street/park/market/harbor/interior), time_of_day (day/golden/night) and subjects (from input/vocab.txt, multi-value) after viewing it in any player.

Platforms
mac · macOS win · Windows
Applications
Excel · QuickTime Player
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment type · Text & forms
Work product
WP02 · Records/table
Conditions
multiapp · Multi-app
Proposed difficulty
2 / 5

Proposed verifier

Per clip vs GT: location exact; time_of_day exact; subjects F1 ≥ 0.7; values restricted to vocabulary.

Proposed partial credit

35% location; 30% time of day; 25% subjects; 10% vocabulary compliance.

R06.04.02.05-T09 Flag technical problems in rushes with timecodes

Original instruction

Resolve bin 'Day3_Rushes' (35 clips). Log every technical issue (soft focus on subject, boom/mic in frame, lens flare, clipped highlights on faces, dropped frames) in out/tech_flags.csv with clip, TC in/out and issue type.

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
vis · Visual judgment nav · Pan/zoom/scroll type · Text & forms
Work product
WP02 · Records/table
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Match vs 21 GT issues (same clip and type, TC overlap ≥ 50%): recall and precision; per-type F1 reported; dropped frames GT from frame-duplication analysis. Resolve scripting is disabled and the media folder is not accessible from the agent's shell (no ffmpeg/OpenCV), so footage can only be reviewed in Resolve's viewer.

Proposed partial credit

55% issue recall; 25% precision; 20% TC accuracy.

R06.04.02.05-T10 Log screen direction and flag 180-degree rule breaks

Original instruction

Premiere Pro bin 'Sc22_Coverage' has 10 setups of a two-person dialogue. For each, log which way each actor looks (screen-left/right) and camera side of the line in out/sc22_axis.csv, and flag setups that break the 180° rule relative to the master.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment type · Text & forms
Work product
WP02 · Records/table
Conditions
licensed · Licensed pro app multiapp · Multi-app
Proposed difficulty
4 / 5

Proposed verifier

Per setup: actor gaze direction exact for both actors vs GT; camera side exact; flagged set equals GT violators (F1). The media folder is not accessible from the agent's shell (no ffmpeg, OpenCV or vision API), so footage can only be reviewed in Premiere's player.

Proposed partial credit

40% gaze directions; 25% camera side; 35% violations flagged correctly.

R06.04.02.05-T11 Reconcile the planned shot list against what was shot

Original instruction

input/shotlist_day2.xlsx lists 28 planned shots (description, size, angle). Resolve bin 'Day2' has 46 clips. Add columns to a copy out/shotlist_reconciled.xlsx: captured (Y/N), clip names covering it, and notes for partial coverage.

Platforms
mac · macOS win · Windows
Applications
Excel · DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment type · Text & forms multi · Cross-app
Work product
WP02 · Records/table
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

Per planned shot: captured flag exact vs GT; clip set F1 ≥ 0.8; 4 GT partial shots have non-empty notes. Original columns unchanged.

Proposed partial credit

40% captured flags; 40% clip mapping; 10% partial notes; 10% sheet integrity.

R06.04.02.05-T12 Cross-check sound reports against picture rolls

Original instruction

input/sound_report_day6.pdf lists 52 sound takes (scene, take, track names, notes like 'wild line', 'room tone'). Resolve bin 'Day6' has the picture clips and WAVs. Produce out/sound_reconcile.csv matching each sound take to its picture clip (or 'wild'/'room tone'), and flag mismatches (missing files, wrong scene/take metadata).

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
av · Audio/video over time type · Text & forms multi · Cross-app
Work product
WP02 · Records/table
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

Matches vs GT pairing (by TC overlap and slate audio): accuracy ≥ 95% for full credit; wild/room tone correctly labeled; 5 GT mismatches flagged with correct reason.

Proposed partial credit

55% pairing accuracy; 15% wild/room tone labels; 30% mismatches flagged.

R06.04.02.05-T13 Pull ten best soundbites for five themes with exact timecodes

Original instruction

Resolve project 'Founders_Doc' has 4 interview clips (2 h total). Pull 10 soundbites (2 per theme from input/themes.md), each a complete thought of 8–25 s, logged in out/bites.xlsx with clip, TC in, TC out, theme and verbatim text.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
av · Audio/video over time type · Text & forms nav · Pan/zoom/scroll
Work product
WP02 · Records/table
Conditions
multiapp · Multi-app
Proposed difficulty
4 / 5

Proposed verifier

Each bite: TC range inside a GT acceptable bite span for its theme (overlap ≥ 80%), 8–25 s, starts/ends in pauses; verbatim text WER ≤ 5% vs ASR of the range; 2 per theme.

Proposed partial credit

50% bites in acceptable theme spans; 15% durations; 15% clean in/out; 20% verbatim text.

R06.04.02.05-T14 Fill custom bin columns in Avid from camera reports

Original instruction

Avid Media Composer bin 'Day9_A' has 60 clips. Using input/camera_report_A.pdf, fill custom columns Scene, Take, Camroll, Lens and Circle for each clip, then export the bin as out/day9_A.ale.

Platforms
mac · macOS win · Windows
Applications
Avid Media Composer
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
HY · Hybrid
Interaction primitives
type · Text & forms click · Point & click multi · Cross-app
Work product
WP02 · Records/table
Conditions
licensed · Licensed pro app
Proposed difficulty
2 / 5

Proposed verifier

Parse ALE: columns exist; per clip values equal camera report (matched by clip/TC); Circle flags exact; 60 rows.

Proposed partial credit

20% columns created; 50% values correct; 20% circles; 10% ALE export.

R06.04.02.05-T15 Build a continuity sheet with end-frame grabs per take

Original instruction

For Premiere Pro bin 'Sc31' (9 takes), create out/sc31_continuity.xlsx with one row per take (take, last good TC, wardrobe notes) and the take's last good frame inserted as an image in column D, grabbed with Export Frame.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click nav · Pan/zoom/scroll multi · Cross-app vis · Visual judgment
Work product
WP02 · Records/table
Conditions
licensed · Licensed pro app multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

xlsx: 9 rows; each embedded image matches the take frame at the given TC (pHash distance ≤ 6) and that TC is within 1 s of GT last-good frame (before 'cut'); wardrobe notes mention GT items (keyword match).

Proposed partial credit

20% rows; 40% correct frames embedded; 25% last-good TC; 15% wardrobe notes.

R06.04.02.05-T16 Track character looks across fifteen scenes and catch story-day errors

Original instruction

Resolve project 'Feature_Assy' covers 15 scenes. input/story_days.csv gives each scene's story day. Build out/looks_matrix.xlsx (scene × 4 characters: hair up/down, jacket on/off, glasses) from the footage, and flag the scenes where a look changes within the same story day without a scripted reason.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment nav · Pan/zoom/scroll type · Text & forms
Work product
WP02 · Records/table
Conditions
long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

Matrix cells vs GT (scene, character, attribute) accuracy ≥ 90%; flagged scenes vs 3 GT continuity errors (F1); flags reference the attribute that changed. Resolve scripting is disabled and the media folder is not accessible from the agent's shell (no ffmpeg/OpenCV), so footage can only be reviewed in Resolve's viewer.

Proposed partial credit

50% matrix accuracy; 35% errors flagged; 15% attribute named correctly.

R06.04.02.05-T17 Log archival footage with on-screen dates and rights

Original instruction

Folder fixtures/archive_reels/ has 25 archival clips plus input/licenses.pdf. In out/archive_log.xlsx record per clip: source, date visible on screen (if any), places/people identified from on-screen captions, license type and usage restrictions from the PDF.

Platforms
mac · macOS win · Windows
Applications
Excel · DaVinci Resolve
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
HY · Hybrid
Interaction primitives
vis · Visual judgment type · Text & forms multi · Cross-app
Work product
WP02 · Records/table
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

Per clip vs GT: on-screen date exact (or blank when none); captions-derived places/people F1 ≥ 0.8; license type and restriction exact from PDF.

Proposed partial credit

30% dates; 30% places/people; 30% license fields; 10% all 25 rows present.

R06.04.02.05-T18 Control: generate the technical log columns with ffprobe

Original instruction

Control pair of T01 (CLI half, technical fields only). For the 40 clips in fixtures/rushes_day5/, write out/day5_tech.csv with clip, start TC, duration, fps, resolution, codec, audio channels using ffprobe/MediaInfo and a script.

Platforms
lin · Linux mac · macOS win · Windows
Applications
ffprobe · MediaInfo
Interfaces
term · Terminal
GUI need · task setting
CL · CLI-sufficient
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands
Work product
WP02 · Records/table
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Every field equals GT from container metadata (exact). Paired with T01 to separate metadata extraction (CLI-solvable) from viewing-based logging (shot size, circled takes).

Proposed partial credit

80% fields correct (scaled); 20% all 40 rows.

R06.04.02.05-T19 Log a full shooting day with circled takes and best-take picks

Original instruction

Resolve project 'Day11' has 120 clips (6 scenes). Produce out/day11_log.xlsx: scene, take, setup, shot size, circled (from audio), continuity notes, issues, and for each setup a recommended best take based on performance clarity (line accuracy vs script) and technical quality.

Platforms
mac · macOS win · Windows lin · Linux
Applications
DaVinci Resolve · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
HY · Hybrid
Interaction primitives
vis · Visual judgment av · Audio/video over time type · Text & forms
Work product
WP02 · Records/table
Conditions
long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

Field accuracies vs GT as in T01; issues recall ≥ 0.8 vs GT; best take per setup ∈ GT acceptable set (no technical flaw, line accuracy ≥ 95%). Score aggregates per column.

Proposed partial credit

30% slate fields; 30% shot sizes and circled; 15% issues; 25% best-take picks.

R06.04.02.05-T20 Log story beats across a three-hour reality-show timeline

Original instruction

Premiere Pro sequence 'House_Day3' (3 h, 2 cams synced). Log every story beat in out/day3_beats.xlsx: TC in/out, characters present, location, beat summary, and tag beats that pay off earlier setups listed in input/storylines.md.

Platforms
mac · macOS win · Windows
Applications
Premiere Pro · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
av · Audio/video over time vis · Visual judgment nav · Pan/zoom/scroll type · Text & forms
Work product
WP02 · Records/table
Conditions
licensed · Licensed pro app long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

Beat spans vs 46 GT beats (temporal IoU ≥ 0.5): recall/precision; characters present F1 ≥ 0.85; location exact; storyline payoff tags vs GT (F1). Summaries LLM-rubric checked (10%).

Proposed partial credit

40% beat recall/precision; 30% characters and locations; 20% payoff tags; 10% summaries.

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