AgentSkillsCN

kuavi-vqa

利用嵌入式向量回答关于视频内容的多项选择题。

SKILL.md
--- frontmatter
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:

  • •Text before -- is the question
  • •Text after -- is a pipe-separated list of choices

Example: 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.