image_vision_read

Capability

• •读取并描述一张或多张图片。支持图片在线链接（http/https）或本地绝对路径；结合文本提示进行识别并返回结果。支持常见图片格式包括 JPG、PNG、GIF、WebP 等。

Real-world impact

• •Read-only: does not modify local files or external systems.

Typical scenarios

• •Use when the user explicitly requests this capability and the required inputs can be extracted from context or asked via a follow-up question.

Non-goals

• •Do not use when inputs are missing and cannot be reliably inferred. Ask a follow-up question instead.
• •Do not fabricate IDs, paths, URLs, tokens, or example values.

Input

• •

  Required fields:

  • •prompt
• •

  Conditional required (anyOf/oneOf): satisfy at least one group:

  • •[prompt]
  • •[image]
  • •[images]
• •

  Prefer batch/array fields when the schema provides both singular and plural versions (e.g., query/queries, file/files, keyword/keywords).

• •

  Prefer real values extracted from conversation history or prior tool results; do not use placeholders.

Output

• •The tool returns structured data. If it produces local files, paths must be absolute paths.

Failure modes

• •If execution fails, explain the reason and provide actionable next steps (e.g., correct inputs, retry later, narrow scope).