Install any skill in seconds. Free to start, no credit card required.
Get Started Free →专为InfiniteTalk项目设计的小省导购员数字人带货提示词生成技能,基于四大智能体协同(提示词生成师、质量管控师、知识库运维师、跨环节适配师),生成适配Image-to-Video模式的结构化提示词(角色固定特征+动作时序+场景环境+音频匹配+光影氛围+技术约束),支持9:16竖屏、5s/幕、音频同步(Suno+chinese-wav2vec2-base)、一致性管控(角色/视觉/情绪),直接对接模型推理流程
.claude/skills/anbeime-infinitetalk-shopping-avatar/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 119% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 72% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 72% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 74% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 91% | 0% |
提示词生成师负责生成适配InfiniteTalk的结构化提示词:
输入信息:
生成流程:
输出格式:
prompt参数)质量管控师负责核查提示词质量:
核查维度:
核查结果:
知识库运维师负责归档提示词和知识库更新:
归档内容:
归档格式:
跨环节适配师负责跨工具集成和音频/图片适配:
音频集成:
图片集成:
init_image参数输入InfiniteTalk推理对接:
功能说明:生成产品推荐场景的五幕提示词
执行方式:提示词生成师(智能体自然语言生成)
关键参数:
输出:五幕中文提示词、参数配置表
功能说明:核查提示词质量
执行方式:质量管控师(智能体分析推理)
核查维度:技术参数匹配度、角色一致性、音频匹配性、光影可实现性
输出:质检报告(pass/fail)、问题清单(如不通过)
功能说明:Suno音乐→InfiniteTalk音频对接
执行方式:跨环节适配师(智能体理解+自然语言指导)
关键步骤:
输出:音频编码文件、视频生成结果
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 22,520 | 28,629 | +27% | 1 | 1 | 0% | 3,515 | 7,697 | +119% | 0 | 0 | — |
case-02 | fail→fail | 11,221 | 7,085 | -37% | 1 | 1 | 0% | 1,517 | 3,587 | +136% | 0 | 0 | — |
case-03 | fail→pass | 23,434 | 31,640 | +35% | 1 | 1 | 0% | 4,300 | 7,398 | +72% | 0 | 0 | — |
case-04 | fail→fail | 22,305 | 29,960 | +34% | 1 | 1 | 0% | 3,210 | 6,626 | +106% | 0 | 0 | — |
case-05 | fail→pass | 17,795 | 9,055 | -49% | 1 | 1 | 0% | 2,302 | 3,967 | +72% | 0 | 0 | — |
case-06 | fail→pass | 15,099 | 10,303 | -32% | 1 | 1 | 0% | 2,311 | 4,014 | +74% | 0 | 0 | — |
case-07 | pass→pass | 15,645 | 15,352 | -2% | 1 | 1 | 0% | 2,824 | 5,531 | +96% | 0 | 0 | — |
case-08 | fail→pass | 27,964 | 29,336 | +5% | 1 | 1 | 0% | 4,278 | 8,160 | +91% | 0 | 0 | — |
case-09 | fail→pass | 12,533 | 4,891 | -61% | 1 | 1 | 0% | 2,175 | 3,264 | +50% | 0 | 0 | — |
case-10 | pass→pass | 8,213 | 2,542 | -69% | 1 | 1 | 0% | 1,349 | 2,850 | +111% | 0 | 0 | — |
case-11 | fail→pass | 19,084 | 10,857 | -43% | 1 | 1 | 0% | 2,637 | 4,353 | +65% | 0 | 0 | — |
case-12 | fail→pass | 12,368 | 4,116 | -67% | 1 | 1 | 0% | 2,067 | 3,236 | +57% | 0 | 0 | — |
case-13 | fail→pass | 7,231 | 3,932 | -46% | 1 | 1 | 0% | 1,195 | 2,983 | +150% | 0 | 0 | — |
case-14 | fail→pass | 16,149 | 11,061 | -32% | 1 | 1 | 0% | 2,499 | 4,389 | +76% | 0 | 0 | — |
case-15 | fail→pass | 15,944 | 15,456 | -3% | 1 | 1 | 0% | 2,694 | 5,036 | +87% | 0 | 0 | — |
case-16 | fail→pass | 19,157 | 16,570 | -14% | 1 | 1 | 0% | 3,397 | 5,515 | +62% | 0 | 0 | — |
case-17 | fail→pass | 13,539 | 18,262 | +35% | 1 | 1 | 0% | 2,852 | 5,663 | +99% | 0 | 0 | — |
case-18 | fail→pass | 16,300 | 15,379 | -6% | 1 | 1 | 0% | 2,651 | 4,970 | +87% | 0 | 0 | — |
case-19 | fail→pass | 16,015 | 13,759 | -14% | 1 | 1 | 0% | 2,653 | 4,501 | +70% | 0 | 0 | — |
case-20 | fail→pass | 15,396 | 2,730 | -82% | 1 | 1 | 0% | 2,304 | 2,881 | +25% | 0 | 0 | — |
case-21 | pass→pass | 24,087 | 18,968 | -21% | 1 | 1 | 0% | 4,494 | 6,226 | +39% | 0 | 0 | — |
case-22 | pass→pass | 10,730 | 8,910 | -17% | 1 | 1 | 0% | 1,828 | 3,702 | +103% | 0 | 0 | — |
case-23 | pass→pass | 14,833 | 17,170 | +16% | 1 | 1 | 0% | 2,910 | 5,414 | +86% | 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 +70 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.