Typing corrupts streaming edits #13443

Before submitting your bug report

• I've tried finding an answer on the Continue docs site
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

- OS: Windows: 11
- Continue version: 2.0.0
- IDE version: {
Version: 1.131.0 (system setup)
Commit: e4c7e7b1d6d060162f4aa7f8225271b67ce1df75
Date: 2026-07-28T10:51:25Z
Electron: 42.7.0
ElectronBuildId: 14710161
Chromium: 148.0.7778.280
Node.js: 24.18.0
V8: 14.8.178.38-electron.0
OS: Windows_NT x64 10.0.26200
}
- Model: Deepseek
- config:
  
name: Main Config
version: 1.0.0
schema: v1
disableAutocomplete: true
models:

  - name: DeepSeek Coder
    provider: deepseek
    model: deepseek-coder
    apiKey: 
    contextLength: 128000
    shell: powershell.exe

  
  OR link to agent in Continue hub:

Description

When a tool call (e.g. edit_existing_file, single_find_and_replace) is still streaming and the target file is open and focused in the editor, keystrokes typed by the user are injected into the content being generated — both into the file written to disk and into the tool-call arguments themselves. This produces stray, out-of-place text fragments that belong to neither the existing file nor the intended edit.

This happens only when the user types during an active streaming tool call; it does not occur when the user is idle. It is distinct from the spontaneous argument corruption (Issue A), which occurs with no user input.

Suspected mechanism: during generation, focus/cursor is placed in the open file buffer, so keyboard input is delivered to that buffer and/or desynchronizes the tool-call stream, causing typed text to land inside generated content.

To reproduce

To reproduce

Open a source file in the editor and focus it.
Ask the agent to modify that file (e.g. "add a comment at the top of this file") — a task that emits an edit tool call.
While the tool call is still streaming, type a short distinctive phrase into the editor (e.g. ZZZTEST123).
Observe: the typed phrase (or fragments of it) appears inside the written file content and/or inside the tool-call arguments, at positions unrelated to where you typed.
Expected: user keystrokes affect only the editor as normal editing and never appear inside agent-generated file content or tool-call arguments.

Actual: typed text bleeds into the generated file content / tool-call arguments, corrupting them.

Confirm scope: repeat the same steps without typing — the contamination does not occur. This isolates the bug from Issue A.

Log output