Skip to content

bug: Utilizing experiment_use_tmux prevents the ability to troubleshoot Claude Code #21

@bearded-bytes

Description

@bearded-bytes

Is there an existing issue for this?

  • I have searched the existing issues

Current Behavior

Tmux is a better option visually than using screen. To turn it on, you use the experiment_use_tmux = true option within your template.

When using this option, you are no longer to access the claude code logs located at .claude-code.log. When attempting to look at the logs, the only entries into the log file are "Attaching to existing Claude Code tmux session".

We are currently working with Anthropic on an issue related to Bedrock, where we are getting API Error: 400 Input is too long for requested model but unable to provide Anthropic with the logs due to the bug above. See Screen shot for result in attempting to cat the .claude-code.log.

Image

Relevant Log Output

coder@coder-5b6fc268-a48f-48e1-a17b-2c0c30639b34-6fb5bd54c-n7tg2:~$ cat .claude-code.log
Attaching to existing Claude Code tmux session.
Attaching to existing Claude Code tmux session.

Expected Behavior

.claude-code.log should contain claude output for debugging purposes, like it does when using screen

Steps to Reproduce

  1. create a template with claude code. Within the claude code module, include the experiment_use_tmux = true attribute
  2. Run the template and request Claude Code to perform an action
  3. Open up web terminal and cat .claude-code.log

Environment

Slim build of Coder, does not support the server subcommand.

Additional Context

No response

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions