Pasted images visible to vision but not saveable to filesystem

Bug

When a user pastes/attaches an image in a Claude Code conversation, the model can see and analyze the image (vision works), but there is no tool or mechanism to save the image data to a file on the local filesystem.

Expected behavior

If Claude can see the image pixels, it should be able to write those pixels to a file path (e.g., via the Write tool or a dedicated image-save tool). The user reasonably expects that pasting an image means Claude can use it — including saving it to disk.

Current behavior

• Image is visible to the model (vision analysis works perfectly)
• No file is created on disk — the image exists only in the conversation stream
• The Read tool can read existing image files, but there's no inverse operation
• The user has to manually save the image themselves, defeating the purpose of pasting it into the conversation

Reproduction

1. Open Claude Code CLI
2. Paste a screenshot into the conversation
3. Ask Claude to save it to a specific path (e.g., public/images/screenshot.png)
4. Claude can describe the image in detail but cannot write it to disk

Impact

This blocks common workflows like:

• "Here's a screenshot, use it as the product image"
• "Save this diagram to docs/"
• "Replace the logo with this image"

The user expects paste → save to work, and it's genuinely confusing when Claude can clearly see the image but claims it can't save it.

Environment

• Claude Code CLI (macOS)
• Model: Claude Opus 4.6