What happened?
When a session is created via Claude Code running inside the VS Code
extension, the IDE prepends an <ide_opened_file>...</ide_opened_file>
context block to the user's prompt. The CLI surfaces this injected block
verbatim in both the session/checkpoint title and the rendered prompt
body, instead of showing only the user's actual message. The result is a
"dirty" title and prompt cluttered with IDE context that the user never typed.
Steps to reproduce
- Open any file in VS Code with the Claude Code extension active.
- Send a prompt to Claude Code (e.g. "rewrite these docs as one plan").
- The IDE silently prepends an
<ide_opened_file>...</ide_opened_file>
block to the message before it's sent.
- Open the session in the web viewer.
Entire CLI version
Entire CLI 0.7.5; Go version: go1.26.4
OS and architecture
OS/Arch: linux/amd64
Agent
Claude Code
Terminal
No response
Logs / debug output
Additional context
No response
What happened?
When a session is created via Claude Code running inside the VS Code
extension, the IDE prepends an
<ide_opened_file>...</ide_opened_file>context block to the user's prompt. The CLI surfaces this injected block
verbatim in both the session/checkpoint title and the rendered prompt
body, instead of showing only the user's actual message. The result is a
"dirty" title and prompt cluttered with IDE context that the user never typed.
Steps to reproduce
<ide_opened_file>...</ide_opened_file>block to the message before it's sent.
Entire CLI version
Entire CLI 0.7.5; Go version: go1.26.4
OS and architecture
OS/Arch: linux/amd64
Agent
Claude Code
Terminal
No response
Logs / debug output
Additional context
No response