claude-code - 💡(How to fix) Fix Regression in v2.1.140: Windows absolute file-path paste no longer attaches as image [1 comments, 2 participants]

Environment

• Windows 11 Pro
• Claude Code CLI v2.1.140 (npm global install, not WSL)
• Windows Terminal

Previous behavior (through v2.1.139)

Pasting a Windows absolute file path like C:\path\to\screenshot.png into the Claude Code prompt would attach the referenced file as [Image #1]. The model received the image as a multimodal attachment and could see it.

Workflow:

1. Snipping Tool rectangle capture
2. Save to a folder under the project cwd
3. Copy the full filename during the Save dialog
4. Switch to Claude Code, Ctrl+V → [Image #1] chip appears

Current behavior (v2.1.140)

Same Ctrl+V paste inserts the absolute path as literal text. No attachment, no chip. Claude only sees the path string.

Sanity checks ruling out other causes

• Binary clipboard paste (Snipping Tool → Ctrl+V the bitmap directly, without saving) also produces nothing in Claude Code on Windows — appears to be the same class of issue as #22068.
• Same clipboard contents paste correctly into Outlook, confirming the bitmap is actually on the Windows clipboard.
• @<path> syntax is text-only — inlining a binary PNG corrupts it; not a viable substitute.

Net effect

With both binary-paste and filename-paste broken, there is currently no working path to attach a screenshot on Windows without WSL. Filename-paste was effectively the only working Windows workaround and v2.1.140 removed it.

Likely cause

The v2.1.140 changelog mentions a fix for "pasting/dropping multiple images only inserting the last one." That change appears to have tightened image-clipboard detection in a way that also dropped the filename-text fallback that Windows users were relying on.

Suggested fix

Either:

• Restore the filename-text-as-image-path heuristic on Windows, or
• Make Windows binary clipboard paste work (the more general fix; would also close #22068), or
• Add explicit @<path> attach syntax for images that recognizes image extensions and attaches as multimodal rather than inlining as text.