Computer-Use Agent Taxonomy中文
R06.12.01.01 · TASK CLASS

Streaming & recording scene composition

Build scenes, sources, audio and outputs for a stream or recording.

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

Typical applications
OBS Studio · Streamlabs · vMix
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.

R06.12.01.01-T01 Build Starting Soon, Live and BRB scenes to a layout spec

Original instruction

In OBS Studio create scene collection 'Weekly Stream' at 1920x1080 with scenes Starting Soon (image bg/starting.png + countdown text), Live (Display Capture full frame, webcam 'Cam' 480x270 at bottom-right with 24 px margin, overlay/frame.png on top) and BRB (bg/brb.png + muted mic). Set the default transition to Fade 300 ms.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
drag · Drag & drop geom · Precise geometry click · Point & click
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Parse basic/scenes/Weekly_Stream.json: 3 scenes with listed sources and z-order; Cam bounds 480x270 at (1416,786)±2 px; frame.png topmost; BRB mic muted; transition Fade 300 ms. Comparison with control pair T17 output.

Proposed partial credit

25% scenes and sources; 35% Cam geometry; 20% z-order and mute; 20% transition

R06.12.01.01-T02 Configure multitrack 1080p60 recording and remux to MP4

Original instruction

In OBS Studio (Advanced output mode) set recording to MKV, 1920x1080 at 60 fps, hardware or x264 encoder at 12 Mbps CBR, keyframe 2 s, with audio tracks 1 = mix, 2 = Mic only, 3 = Desktop only. Record 10 s of scene Live, then remux it to recordings/test_rec.mp4.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click type · Text & forms
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

ffprobe test_rec.mp4: 1920x1080, 60 fps, video bitrate 12 Mbps ±10%, GOP 120, 3 audio streams. Track 2 correlates with mic test tone only, track 3 with desktop tone only. profile basic.ini fields match.

Proposed partial credit

25% video settings; 35% three tracks routed correctly; 20% recording exists; 20% remuxed MP4

R06.12.01.01-T03 Build a broadcast mic filter chain to loudness spec

Original instruction

In OBS Studio, on audio source 'Mic' (fed by a looping test voice file through a virtual device), add filters in order: Noise Suppression (RNNoise), Compressor (4:1, threshold -18 dB, attack 6 ms, release 60 ms), Gain, Limiter (-1 dB). Tune Gain so a 30 s recording of Mic alone integrates to -16 LUFS ±1. Save recordings/mic_check.wav.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
AP · API-sufficient
Interaction primitives
click · Point & click av · Audio/video over time
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Scene JSON filter list order and params exact. mic_check.wav: integrated loudness -16±1 LUFS, true peak ≤ -1 dBTP, noise floor in silent gaps ≤ -60 dBFS (noise suppression active).

Proposed partial credit

30% filter order; 20% compressor params; 35% loudness target; 15% peak and noise floor

R06.12.01.01-T04 Key out a green screen webcam cleanly

Original instruction

In OBS Studio the source 'GreenCam' plays fixtures/greenscreen_take.mp4 (person on uneven green). Add Crop/Pad and Chroma Key filters and tune similarity, smoothness and spill reduction so the person is cleanly keyed over bg/office.png, with no green halo and hair kept. Record 10 s to recordings/key_test.mkv.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment click · Point & click drag · Drag & drop
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Compare recording with GT alpha mattes for sampled frames: foreground IoU ≥0.95; residual green pixels (hue band) in composited background ≤0.3%; spill: mean g-(r+b)/2 in edge band ≤ threshold. Filter params exist in scene JSON. obs-websocket is disabled, no Lua/Python scripts may be loaded (Tools > Scripts list checked), and scene JSON writes by other processes are caught by a file-write audit.

Proposed partial credit

45% key IoU; 25% no residual green; 20% spill suppressed; 10% recording

R06.12.01.01-T05 Stream to a local RTMP test server with platform-compliant settings

Original instruction

In OBS Studio configure a Custom stream server rtmp://127.0.0.1/live with key test_key, 1920x1080 at 30 fps, CBR 6000 kbps, keyframe interval 2 s, AAC 160 kbps 48 kHz. Go live with scene Live for 60 s, then stop.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
AP · API-sufficient
Interaction primitives
click · Point & click type · Text & forms
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Local nginx-rtmp records the ingest: ffprobe shows 1920x1080, 30 fps, video bitrate 6000 kbps ±10%, keyframe spacing 2.0 s, AAC 160 kbps 48 kHz; stream duration 60±5 s; service.json has the server and key.

Proposed partial credit

20% server config; 40% encoder settings verified at ingest; 25% audio; 15% duration

R06.12.01.01-T06 Reuse an overlay via nested scenes and add scene hotkeys

Original instruction

In OBS Studio create scene 'Overlay' (alerts area, ticker text, logo) and nest it as a source in scenes Gameplay, Just Chatting and Desktop at identical position. Bind hotkeys Ctrl+Alt+1/2/3 to switch to those scenes and Ctrl+Alt+M to toggle Mic mute.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
drag · Drag & drop key · Shortcuts click · Point & click
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Scene JSON: Overlay referenced as scene source in 3 scenes with identical transform; hotkey bindings in the collection/profile match. Synthetic key events via OS automation switch program scene (obs-websocket read-only observer).

Proposed partial credit

40% nested overlay in 3 scenes; 20% identical transforms; 40% hotkeys work

R06.12.01.01-T07 Duck music under the mic with a sidechain compressor

Original instruction

In OBS Studio add a Compressor filter on source 'Music' with sidechain source 'Mic' so music drops about 12 dB while the host speaks and recovers within 500 ms after. Record 40 s of the provided test feed (voice with pauses) on a single mix track to recordings/duck_test.wav.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
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
4 / 5

Proposed verifier

Separate stems are known: estimate music gain per 50 ms window by projecting the mix onto the music reference. During speech gain reduction 12±3 dB; within 500 ms of speech end recovers to ≥-2 dB; no ducking during silence.

Proposed partial credit

25% sidechain configured; 40% ducking depth; 25% recovery time; 10% recording

R06.12.01.01-T08 Enable a 30-second replay buffer with a save hotkey

Original instruction

In OBS Studio enable Replay Buffer with maximum time 30 s and memory 512 MB, set file name format Replay %CCYY-%MM-%DD %hh-%mm-%ss in recordings/replays/, bind Save Replay to F9, start the buffer, wait 40 s and save one replay.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click key · Shortcuts
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Profile basic.ini replay settings exact; hotkey binding present. One file in recordings/replays/ matching the name pattern with duration 30±1 s.

Proposed partial credit

40% replay settings; 20% hotkey; 40% replay file correct

R06.12.01.01-T09 Set up alerts and chat widgets in Streamlabs Desktop

Original instruction

In Streamlabs Desktop (sandbox account) add scene 'Stream' with game capture, an Alert Box widget positioned top-centre (800x300) with follower alerts using sound alerts/ding.mp3 at 40% volume and 5 s duration, and a Chat Box widget 400x700 at the right edge with font size 18. Trigger a test follower alert.

Platforms
mac · macOS win · Windows
Applications
Streamlabs Desktop
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
drag · Drag & drop click · Point & click type · Text & forms
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Scene collection JSON: widget sources with transforms ±2 px. Streamlabs sandbox widget settings via test API: follower alert sound, volume 40, duration 5. Test alert event logged; frame capture shows alert rendered in top-centre box.

Proposed partial credit

30% widget placement; 40% alert settings; 15% chat box style; 15% test alert renders

R06.12.01.01-T10 Build a vMix preset with four NDI cameras, a title and multiview

Original instruction

In vMix add four NDI inputs (test sources CAM1–CAM4), a GT Title using template LowerThird.gtzip with Name 'Dr. Sample' and Role 'Guest', set multiview to show CAM1–4 and the title, and set CAM1 as Program and CAM2 as Preview. Save the preset as presets/panel.vmix.

Platforms
win · Windows
Applications
vMix
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
click · Point & click type · Text & forms drag · Drag & drop
Work product
WP08 · App/system state
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

Parse panel.vmix XML: 4 NDI inputs with source names, title input with field values exact, multiview overlay assignments, active=CAM1 and preview=CAM2. vMix HTTP API XML state equals preset after reload.

Proposed partial credit

35% NDI inputs; 25% title fields; 20% multiview; 20% program/preview

R06.12.01.01-T11 Add a vertical 9:16 canvas for simultaneous short-form output

Original instruction

In OBS Studio with the Aitum Vertical plugin installed, configure a 1080x1920 vertical canvas with scene 'V-Live': webcam top half (1080x810 crop of Cam), gameplay crop bottom half, captions text bottom. Set vertical recording to recordings/vertical/ and record 15 s alongside the normal 1920x1080 recording.

Platforms
mac · macOS win · Windows
Applications
OBS Studio · Aitum Vertical
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
drag · Drag & drop geom · Precise geometry vis · Visual judgment
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Two recordings with overlapping timestamps: 1920x1080 and 1080x1920. Vertical frames: webcam region top 810 px matches Cam crop (SSIM ≥0.9 vs reference frame), gameplay crop bottom region matches; plugin config JSON has V-Live scene. obs-websocket is disabled, no Lua/Python scripts may be loaded (Tools > Scripts list checked), and config writes by other processes are caught by a file-write audit.

Proposed partial credit

25% vertical canvas config; 40% layout correct in frames; 20% both recordings; 15% captions source

R06.12.01.01-T12 Add a local HTML overlay as a browser source

Original instruction

In OBS Studio add a Browser source 'Ticker' to scene Live using local file overlay/ticker.html at 1920x80, positioned at the bottom edge, with custom CSS 'body{background:rgba(0,0,0,0)}' and 'Shutdown source when not visible' enabled. Refresh cache once.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click type · Text & forms drag · Drag & drop
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Scene JSON: browser_source with is_local_file true, path, width 1920, height 80, css string, shutdown true; transform y=1000±1. Program screenshot via websocket shows ticker text at bottom with transparent background.

Proposed partial credit

40% source settings; 30% position; 30% renders transparent

R06.12.01.01-T13 Record a tutorial that captures only one app window

Original instruction

In OBS Studio create scene 'Tutorial' that captures only the Calculator window (window capture, cursor on) scaled to fit 1280x720 canvas, with a mic source. Record 20 s while performing 12 × 34 = in Calculator. No other window, notification or desktop may appear in the recording. Save recordings/calc_tutorial.mp4.

Platforms
mac · macOS win · Windows
Applications
OBS Studio · Calculator
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
click · Point & click vis · Visual judgment multi · Cross-app
Work product
WP05 · Time-based media
Conditions
multiapp · Multi-app interrupt · Interruptions
Proposed difficulty
3 / 5

Proposed verifier

Recording 1280x720, 18–25 s; OCR shows '408' in Calculator display by the end; per-frame check: no wallpaper template or test notification (fired mid-task) pixels visible; scene JSON uses window capture, not display capture. obs-websocket is disabled, no Lua/Python scripts may be loaded, the agent has no terminal or ffmpeg, and scene JSON writes by other processes are caught by an audit log.

Proposed partial credit

30% window-only capture; 25% canvas and duration; 25% action recorded; 20% no leak of notification

R06.12.01.01-T14 Measure and correct webcam audio-video offset

Original instruction

In OBS Studio the webcam 'Cam' (test feed: clapper video) arrives later than the 'Mic' audio. Record a clap test, measure the offset, set Sync Offset on Mic (or a Render Delay on Cam) to fix it, and record recordings/sync_after.mkv showing ≤10 ms residual offset.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
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 click · Point & click
Work product
WP05 · Time-based media
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Hidden injected delay is 120 ms. sync_after.mkv: detect clapper close frame (visual) vs audio transient onset over 5 claps; mean |offset| ≤10 ms. Scene JSON sync offset or render delay within 120±15 ms.

Proposed partial credit

20% before-test recorded; 40% correct offset set; 40% residual ≤10 ms

R06.12.01.01-T15 Migrate a Streamlabs scene collection into OBS and fix broken sources

Original instruction

Import fixtures/streamlabs_collection.overlay into OBS Studio as collection 'Migrated'. Several sources reference missing file paths and Streamlabs-only widgets. Relink media to the files in assets/, replace widgets with Browser sources to the provided URLs in widget_urls.txt, and make all 5 scenes render without missing-source warnings.

Platforms
mac · macOS win · Windows
Applications
OBS Studio · Streamlabs Desktop
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click type · Text & forms vis · Visual judgment
Work product
WP08 · App/system state
Conditions
multiapp · Multi-app hidden · Hidden/stale state
Proposed difficulty
4 / 5

Proposed verifier

Collection JSON: 5 scenes; every file source path exists; widget sources replaced by browser sources with listed URLs; transforms preserved within 2 px of original; OBS log on load shows no 'missing files'.

Proposed partial credit

20% import; 35% media relinked; 25% widgets replaced; 20% transforms preserved

R06.12.01.01-T16 Configure a stinger transition with a track matte

Original instruction

In OBS Studio enable Studio Mode and add a Stinger transition using transitions/swipe.webm with track matte transitions/swipe_matte.webm (stacked layout), transition point 600 ms, audio monitoring off. Make it the transition for Live→BRB only; keep Fade elsewhere. Record a cut Live→BRB to recordings/stinger.mkv.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
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
None declared
Proposed difficulty
4 / 5

Proposed verifier

Collection JSON: stinger transition with matte settings and point 600 ms; scene-specific transition override BRB←Live. Recording: swipe frames present; scene content switches at 600±33 ms from transition start; Fade used in a second test switch.

Proposed partial credit

30% stinger config; 20% track matte; 25% override only for BRB; 25% timing in recording

R06.12.01.01-T17 Create the Weekly Stream scenes via obs-websocket (control pair of T01)

Original instruction

Control pair of T01. With OBS running headless-accessible and obs-websocket v5 enabled (password in .env), write build_scenes.py that creates the same three scenes, sources, Cam transform and Fade 300 ms transition without touching the OBS window.

Platforms
mac · macOS win · Windows lin · Linux
Applications
OBS Studio · obs-websocket
Interfaces
api · API/SDK
GUI need · task setting
AP · API-sufficient
GUI need · ordinary access
AP · API-sufficient
Interaction primitives
code · Code & commands
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Same checks as T01 on the resulting collection JSON; websocket server log shows requests from the script; no UI interaction events recorded.

Proposed partial credit

50% scenes/sources; 30% transforms; 20% transition

R06.12.01.01-T18 Stream to two destinations and record program plus ISO in vMix

Original instruction

In vMix configure streaming to two local RTMP endpoints (rtmp://127.0.0.1/a and /b) at 1080p30 6 Mbps, and MP4 recording of Program plus MultiCorder ISO recordings of CAM1 and CAM2 to D:\rec\. Run for 45 s with two cuts between CAM1 and CAM2 at about 15 s and 30 s.

Platforms
win · Windows
Applications
vMix
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click type · Text & forms av · Audio/video over time
Work product
WP05 · Time-based media
Conditions
licensed · Licensed pro app
Proposed difficulty
4 / 5

Proposed verifier

Both RTMP ingests recorded with matching params; Program MP4 shows cuts near 15 s and 30 s (±2 s) via shot-boundary detection; two ISO files with continuous single-camera content and duration ≥40 s.

Proposed partial credit

30% dual streaming; 25% program recording with cuts; 30% ISO files; 15% settings

R06.12.01.01-T19 Build a complete stream setup from a producer's brief

Original instruction

From brief_stream.pdf in OBS Studio: 6 scenes and 14 sources with layout coordinates, mic and music filter chains, sidechain ducking, scene hotkeys, stream to local RTMP at the brief's encoder settings, recording track 2 = VOD mix without music, and a Starting Soon countdown that switches to Live at 0. Save collection 'Show_v1'.

Platforms
mac · macOS win · Windows
Applications
OBS Studio
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
drag · Drag & drop geom · Precise geometry click · Point & click av · Audio/video over time
Work product
WP08 · App/system state
Conditions
long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

Collection JSON vs brief: scenes/sources/transforms (±2 px), filters and params, hotkeys. 2-minute dry run: RTMP ingest params; recording track 2 contains no music (correlation ≤0.05) while track 1 does; countdown triggers switch to Live.

Proposed partial credit

30% scenes/layout; 20% audio chains; 20% outputs; 30% VOD track and countdown automation

R06.12.01.01-T20 Auto-switch to a privacy scene when sensitive windows appear

Original instruction

In OBS Studio with Advanced Scene Switcher, add macros: switch to scene 'Privacy' when the focused window title contains 'Password' or 'Bank', return to the previous scene 2 s after it loses focus, and mute Mic while in Privacy. Test by opening the provided fixtures/fake_password_manager app (synthetic data).

Platforms
mac · macOS win · Windows
Applications
OBS Studio · Advanced Scene Switcher
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click type · Text & forms multi · Cross-app
Work product
WP08 · App/system state
Conditions
multiapp · Multi-app interrupt · Interruptions
Proposed difficulty
5 / 5

Proposed verifier

Harness focuses test windows in a scripted sequence and records program scene via websocket: Privacy within 500 ms of focus on matching titles, not for others; returns to previous scene 2±0.5 s after; Mic muted only in Privacy. Macro config exported in plugin settings.

Proposed partial credit

35% switch on match; 15% no false trigger; 25% return to previous; 25% mic mute

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