[BUG][Cowork] Model effort level leaks across projects, scoped globally per-model instead of per-conversation

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

What's Wrong?

In Cowork, changing the model effort level in one conversation silently changes it in every other open conversation across every other project that uses the same model — even though model selection stays independent per conversation. Effort appears to be stored globally per model family, not per conversation.

No notification, no confirmation — it just changes.

Steps to Reproduce

1. In Project A start a conversation, set model Sonnet 5, effort High.
2. In a different Project B start a new conversation, set model Sonnet 5, effort Medium.
3. Switch back to the Project A conversation.
4. Its effort has silently changed to Medium.

Notes

• Model choice is unaffected — different conversations can run different models fine.
• Only effort leaks, and it leaks per model family (all open Sonnet 5 conversations share one effort value app-wide; Opus conversations share a separate one).
• Started ~3-4 weeks ago, coinciding with the Chat/Cowork merge.

Environment

• Product: Cowork (Claude Desktop, not Claude Code CLI)
• OS: macOS Tahoe 26.5.2
• Claude Desktop version: 1.26832

Related

Adjacent but not exact matches — [#66407](https://github.com/anthropics/claude-code/issues/66407) (Cowork, model silently changed + locked), [#40731](https://github.com/anthropics/claude-code/issues/40731) (Cowork Global Instructions overwritten by another session — same "shared state clobbered" pattern, different field), [#49166](https://github.com/anthropics/claude-code/issues/49166) (effort global across sessions, but Claude Code CLI, pre-dates the merge). None describe effort leaking per-model-family across Cowork projects specifically.

What Should Happen?

Expected Behavior

Effort should be scoped per conversation (like model already is). Changing it in one conversation shouldn't touch another.

Error Messages/Logs

Steps to Reproduce

Steps to Reproduce

1. In Project A start a conversation, set model Sonnet 5, effort High.
2. In a different Project B start a new conversation, set model Sonnet 5, effort Medium.
3. Switch back to the Project A conversation.
4. Its effort has silently changed to Medium.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

1.26832

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_