Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Start and stop screen recording using ffmpeg.
.claude/skills/sonichi-screen-record/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -44% | 0% |
Start and stop screen recording using ffmpeg.
Usage: /screen-record start or /screen-record stop
The user asks to record the screen to a video over time (demo, bug repro, walkthrough). For a single still frame, use the screen-capture path instead.
If argument is start:
bashpython3 skills/screen-record/scripts/record.py start
If argument is stop:
bashpython3 skills/screen-record/scripts/record.py stop
.mov written by record.py to /tmp/sutando-recording-<ts>.mov (it prints the path on stop).start while already recording → the script no-ops and returns the existing pid + path; a stale PID file (previous run crashed) is detected and cleared, so it self-heals rather than wedging; stop with nothing running → no-op./opt/homebrew/bin/ffmpeg (hardcoded in three places — record.py:59 device-listing, :152 the recording call, :204 the audio check. Apple Silicon brew prefix; on an Intel Mac brew install ffmpeg lands in /usr/local and the skill stays broken until all three are fixed — patching only the recording call still fails at device-listing before recording starts). No frames / black video → the invoking terminal lacks macOS Screen Recording permission (same grant the screen-capture server uses).| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 6,723 | 3,658 | -46% | 1 | 1 | 0% | 900 | 512 | -43% | 0 | 0 | — |
case-01 | fail→pass | 4,336 | 8,571 | +98% | 1 | 1 | 0% | 621 | 521 | -16% | 0 | 0 | — |
case-03 | fail→pass | 4,308 | 2,855 | -34% | 1 | 1 | 0% | 592 | 495 | -16% | 0 | 0 | — |
case-04 | pass→pass | 3,504 | 5,857 | +67% | 1 | 1 | 0% | 526 | 1,227 | +133% | 0 | 0 | — |
case-05 | pass→pass | 3,849 | 14,760 | +283% | 1 | 1 | 0% | 557 | 1,488 | +167% | 0 | 0 | — |
case-06 | pass→pass | 4,591 | 6,010 | +31% | 1 | 1 | 0% | 739 | 1,324 | +79% | 0 | 0 | — |
case-07 | fail→pass | 11,880 | 2,733 | -77% | 1 | 1 | 0% | 1,777 | 831 | -53% | 0 | 0 | — |
case-08 | pass→pass | 10,562 | 2,248 | -79% | 1 | 1 | 0% | 1,662 | 670 | -60% | 0 | 0 | — |
case-09 | pass→pass | 7,173 | 1,820 | -75% | 1 | 1 | 0% | 1,050 | 654 | -38% | 0 | 0 | — |
case-10 | fail→pass | 8,524 | 2,009 | -76% | 1 | 1 | 0% | 1,317 | 680 | -48% | 0 | 0 | — |
case-11 | fail→pass | 9,038 | 2,373 | -74% | 1 | 1 | 0% | 1,321 | 737 | -44% | 0 | 0 | — |
case-12 | fail→pass | 2,823 | 2,254 | -20% | 1 | 1 | 0% | 322 | 740 | +130% | 0 | 0 | — |
case-13 | fail→pass | 13,571 | 4,246 | -69% | 1 | 1 | 0% | 2,074 | 1,019 | -51% | 0 | 0 | — |
case-14 | pass→pass | 11,139 | 2,572 | -77% | 1 | 1 | 0% | 1,530 | 742 | -52% | 0 | 0 | — |
case-15 | pass→pass | 5,733 | 1,542 | -73% | 1 | 1 | 0% | 863 | 559 | -35% | 0 | 0 | — |
case-16 | fail→pass | 8,310 | 1,603 | -81% | 1 | 1 | 0% | 1,300 | 529 | -59% | 0 | 0 | — |
case-17 | fail→fail | 9,150 | 1,352 | -85% | 1 | 1 | 0% | 1,451 | 552 | -62% | 0 | 0 | — |
case-18 | pass→pass | 13,517 | 3,051 | -77% | 1 | 1 | 0% | 1,956 | 586 | -70% | 0 | 0 | — |
case-19 | pass→pass | 10,792 | 2,687 | -75% | 1 | 1 | 0% | 1,635 | 751 | -54% | 0 | 0 | — |
case-20 | pass→pass | 13,829 | 2,552 | -82% | 1 | 1 | 0% | 2,157 | 778 | -64% | 0 | 0 | — |
case-21 | pass→pass | 5,247 | 2,085 | -60% | 1 | 1 | 0% | 850 | 708 | -17% | 0 | 0 | — |
case-22 | pass→pass | 7,757 | 1,347 | -83% | 1 | 1 | 0% | 1,192 | 540 | -55% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +36 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.