Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create or revise CompozyOS release videos with high-resolution product captures, Ultramock camera scenes, and the reusable Remotion package. Excludes publishing tweets and general video editing.
.claude/skills/compozy-cy-release-video/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 10% | 0% |
Deliver a reviewable release/launch MP4 and its reusable sources in packages/launch-video (@compozy/videos). Ultramock owns the camera treatment; Remotion owns the opening, pacing, scene overlaps, cursor interactions, and final assembly. Continue the existing project instead of rebuilding its animation stack.
observable UI action and result; a PR description or attractive panel alone is not enough to demonstrate the feature.
feedback before editing. Respect the latest requested feature order, duration, text policy, and audio choice. Preserve previous deliverables before replacement.
contain real UI. Do not add captions, title cards, logos, or an outro to the demo unless requested. The initial beta.26 image-only instruction was later amended to include the existing release opening.
The historical beta.26 final was 42.3s, not a universal duration requirement.
approach the control, show the click, reveal its result, then inspect details. A feature with fewer useful states can lead; do not give every feature equal time.
| Work being done | Read | | --- | --- | | New or inadequate screenshots, integrated feature demonstration | Capture and visual direction, all sections | | Editing or exporting the camera timeline in Chrome | Ultramock operation, camera/edit/export sections; recovery only when needed | | Opening, cursor, trim, transition, or release configuration changes | Remotion assembly, relevant section and validation | | Reusing the project's visual research or understanding a prior rejection | Lessons and evidence, matching failure or research section |
Use eng-ui-screenshot for Compozy capture mechanics and exa-search when new component discovery is needed or requested. Inspect installed components first; research is not a reason to add another framework. Browser operations follow the active browser/computer tool's supported API and the user's requested Chrome session; saved tab IDs and accessibility indices are not reusable instructions.
Verify the feature still belongs visibly to its session or parent surface.
after uploads or parameter changes, then export and verify the downloaded file.
interactions and their cursor under the same camera transform. Trigger the matching result on the click; movement alone does not communicate behavior.
contact, opening/closing states, text sharpness, and continuity; fix source captures or geometry before spending time on the complete render.
intervals and adjacent cuts at delivery resolution, and preview playback for timing. Report the distinction if only extracted frames were inspected.
and final MP4. Update package instructions when the reusable workflow changes. Remove owned temporary capture stories, stop only owned capture servers, reset temporary viewport overrides, and keep the edited project/preview available.
Delivery means an accessible final MP4 that demonstrates the requested feature, plus the sources needed to revise it. A configured timeline, successful upload, typecheck, or completed export button is not sufficient. State duration, resolution/fps, and any material limitation (for example, fixtures versus a live provider measurement). Link/show the actual artifact; publishing a release tweet is a separate action requiring user authorization. Do not require a new research corpus, subagent round, approval, or full application QA for every video revision.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 16,222 | 19,975 | +23% | 1 | 1 | 0% | 241 | 1,349 | +460% | 0 | 0 | — |
case-02 | fail→fail | 12,373 | 18,500 | +50% | 1 | 1 | 0% | 256 | 1,337 | +422% | 0 | 0 | — |
case-03 | fail→fail | 38,769 | 19,943 | -49% | 1 | 1 | 0% | 6,519 | 1,174 | -82% | 0 | 0 | — |
case-04 | pass→fail | 7,207 | 12,650 | +76% | 1 | 1 | 0% | 862 | 1,443 | +67% | 0 | 0 | — |
case-05 | fail→fail | 5,080 | 26,217 | +416% | 1 | 1 | 0% | 838 | 1,445 | +72% | 0 | 0 | — |
case-06 | fail→fail | 3,047 | 15,206 | +399% | 1 | 1 | 0% | 166 | 1,396 | +741% | 0 | 0 | — |
case-07 | fail→pass | 11,780 | 18,064 | +53% | 1 | 1 | 0% | 1,625 | 2,130 | +31% | 0 | 0 | — |
case-08 | fail→pass | 12,819 | 10,557 | -18% | 1 | 1 | 0% | 2,117 | 1,929 | -9% | 0 | 0 | — |
case-09 | pass→pass | 11,134 | 7,657 | -31% | 1 | 1 | 0% | 1,554 | 2,072 | +33% | 0 | 0 | — |
case-10 | fail→pass | 11,666 | 3,066 | -74% | 1 | 1 | 0% | 1,904 | 1,424 | -25% | 0 | 0 | — |
case-11 | fail→fail | 10,593 | 4,789 | -55% | 1 | 1 | 0% | 1,452 | 1,685 | +16% | 0 | 0 | — |
case-12 | fail→pass | 9,043 | 4,759 | -47% | 1 | 1 | 0% | 1,388 | 1,588 | +14% | 0 | 0 | — |
case-13 | pass→pass | 15,668 | 9,221 | -41% | 1 | 1 | 0% | 2,260 | 2,283 | +1% | 0 | 0 | — |
case-14 | pass→pass | 11,367 | 6,118 | -46% | 1 | 1 | 0% | 1,621 | 1,853 | +14% | 0 | 0 | — |
case-15 | pass→pass | 24,805 | 9,708 | -61% | 1 | 1 | 0% | 1,764 | 2,257 | +28% | 0 | 0 | — |
case-16 | fail→pass | 9,573 | 4,753 | -50% | 1 | 1 | 0% | 1,401 | 1,540 | +10% | 0 | 0 | — |
case-17 | pass→pass | 6,988 | 4,255 | -39% | 1 | 1 | 0% | 945 | 1,680 | +78% | 0 | 0 | — |
case-18 | pass→pass | 10,356 | 5,071 | -51% | 1 | 1 | 0% | 1,599 | 1,670 | +4% | 0 | 0 | — |
case-19 | fail→fail | 8,934 | 9,001 | +1% | 1 | 1 | 0% | 1,206 | 1,463 | +21% | 0 | 0 | — |
case-20 | pass→pass | 11,626 | 3,618 | -69% | 1 | 1 | 0% | 1,603 | 1,394 | -13% | 0 | 0 | — |
case-21 | fail→pass | 19,786 | 4,902 | -75% | 1 | 1 | 0% | 1,492 | 1,539 | +3% | 0 | 0 | — |
case-22 | fail→pass | 9,737 | 3,727 | -62% | 1 | 1 | 0% | 1,562 | 1,310 | -16% | 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 15 counted toward the lift figure. The other 7 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 +27 percentage points is the difference between those two pass rates over the 15 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
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.