菜单
Answer multiple-choice questions about video content using embeddings
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Compositional pixel analysis patterns using kuavi_eval for counting, motion detection, change tracking, and visual comparison. Use when questions require precise measurements, object counting, or frame-by-frame comparison.
Full end-to-end video analysis with KUAVi
Multi-video corpus indexing and cross-video search
Deep multi-pass video analysis with parallel sharding and zoom
Multi-pass search with query refinement for hard-to-find video content. Use when initial search returns low-confidence results or no matches.
Index a video file for KUAVi analysis
| name | kuavi-vqa |
| description | Answer multiple-choice questions about video content using embeddings |
| argument-hint | <question> -- <choice1> | <choice2> [| choice3] [| choice4] |
| disable-model-invocation | true |
Parse $ARGUMENTS:
-- is the question-- is a pipe-separated list of choicesExample: What color is the car? -- red | blue | green | black
Call kuavi_discriminative_vqa with:
question: the parsed question text (trimmed)candidates: list of parsed choices (each trimmed)Present the answer with confidence scores for each candidate.