Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →The shared brief-to-prompt convention behind the media point tasks — parse the brief, select the model or tool, construct the prompt, then QA the output against the brief. Use when prompting for image, video, speech, or music generation or editing.
.claude/skills/a5c-ai-generative-media-prompting/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -56% | 0% |
All six media point tasks open with the same four-step shape, visible verbatim in their @description headers. Writing it down once removes six copies of the same tacit convention.
for a generation task, or the source asset plus the requested operation for an editing task.
the parsed brief. Each point task names the candidates it selects among; the selection is part of the task, not a caller input.
supports them, the structured parameters that accompany it).
point task ends in a validation step, and what it validates is modality-specific.
Limited to what the existing files already state:
image-generation.js — generates variants in parallel; validates technical andcreative quality; organises outputs with metadata.
image-editing.js — selects among named editing tools; validates edge quality, colorconsistency, and artifact absence.
video-generation.js — the parsed brief includes the request mode (text-to-video,image-to-video, video-to-video); prompt construction carries camera, lighting, and composition parameters; a low-quality output is retried with a fallback model.
video-editing.js — selects among named editing tools and runs a per-operationpipeline; validates frame consistency and audio sync.
speech-generation.js — the brief includes language, style, emotion, and SSML;validates naturalness, pronunciation, and audio specs.
music-generation.js — the brief includes genre, mood, duration, and instruments;mastering and stem separation are applied only if requested; validates musical coherence and technical audio.
This skill describes prompt construction only. Publication, review, and licensing decisions are out of scope and belong to ../../media-production-pipeline.js. No model lists or vendor guidance beyond what the point-task files themselves name.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 12,607 | 18,664 | +48% | 1 | 1 | 0% | 1,265 | 2,774 | +119% | 0 | 0 | — |
case-02 | fail→pass | 22,406 | 13,718 | -39% | 1 | 1 | 0% | 2,597 | 2,023 | -22% | 0 | 0 | — |
case-03 | pass→pass | 19,536 | 8,873 | -55% | 1 | 1 | 0% | 2,387 | 1,143 | -52% | 0 | 0 | — |
case-04 | pass→pass | 19,336 | 7,859 | -59% | 1 | 1 | 0% | 1,817 | 943 | -48% | 0 | 0 | — |
case-05 | pass→pass | 17,267 | 8,989 | -48% | 1 | 1 | 0% | 1,836 | 1,072 | -42% | 0 | 0 | — |
case-06 | pass→pass | 7,837 | 7,605 | -3% | 1 | 1 | 0% | 514 | 993 | +93% | 0 | 0 | — |
case-07 | pass→pass | 18,205 | 7,452 | -59% | 1 | 1 | 0% | 1,985 | 873 | -56% | 0 | 0 | — |
case-08 | pass→pass | 11,645 | 7,560 | -35% | 1 | 1 | 0% | 1,117 | 940 | -16% | 0 | 0 | — |
case-09 | pass→pass | 10,951 | 7,408 | -32% | 1 | 1 | 0% | 903 | 813 | -10% | 0 | 0 | — |
case-10 | pass→pass | 16,324 | 8,383 | -49% | 1 | 1 | 0% | 1,760 | 1,020 | -42% | 0 | 0 | — |
case-11 | pass→pass | 15,399 | 7,260 | -53% | 1 | 1 | 0% | 1,538 | 919 | -40% | 0 | 0 | — |
case-12 | pass→pass | 19,256 | 8,630 | -55% | 1 | 1 | 0% | 2,241 | 1,181 | -47% | 0 | 0 | — |
case-13 | fail→pass | 22,344 | 7,902 | -65% | 1 | 1 | 0% | 2,600 | 1,016 | -61% | 0 | 0 | — |
case-14 | pass→pass | 12,491 | 7,847 | -37% | 1 | 1 | 0% | 1,063 | 945 | -11% | 0 | 0 | — |
case-15 | pass→pass | 17,285 | 8,801 | -49% | 1 | 1 | 0% | 1,901 | 1,144 | -40% | 0 | 0 | — |
case-16 | fail→pass | 20,760 | 8,466 | -59% | 1 | 1 | 0% | 2,798 | 1,195 | -57% | 0 | 0 | — |
case-17 | fail→pass | 21,458 | 9,688 | -55% | 1 | 1 | 0% | 2,495 | 1,317 | -47% | 0 | 0 | — |
case-18 | fail→pass | 20,484 | 7,942 | -61% | 1 | 1 | 0% | 2,421 | 1,073 | -56% | 0 | 0 | — |
case-19 | pass→pass | 18,985 | 9,538 | -50% | 1 | 1 | 0% | 2,442 | 1,239 | -49% | 0 | 0 | — |
case-20 | pass→pass | 21,170 | 13,884 | -34% | 1 | 1 | 0% | 2,715 | 1,877 | -31% | 0 | 0 | — |
case-21 | pass→pass | 14,446 | 8,864 | -39% | 1 | 1 | 0% | 1,900 | 1,104 | -42% | 0 | 0 | — |
case-22 | pass→pass | 17,232 | 8,108 | -53% | 1 | 1 | 0% | 1,906 | 1,014 | -47% | 0 | 0 | — |
case-23 | pass→pass | 21,230 | 13,868 | -35% | 1 | 1 | 0% | 2,450 | 1,679 | -31% | 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. 23 cases were attempted. The headline lift of +22 percentage points is the difference between those two pass rates over the 23 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.