[BUG] Agent worktree subprocesses frequently require re-authentication on Max plan (OAuth)

### 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?

## Environment

• Claude Code version: 2.1.85
• OS: macOS 15 (Darwin 25.3.0)

• Auth method: claude.ai OAuth (first-party)

• Subscription: Max plan (Organization)

## Problem
Agent tool with isolation: "worktree" frequently fails with
"Not logged in · Please run /login", requiring re-authentication

~10 times per day during normal usage.

## Reproduction

1. Log in via /login (OAuth, Max plan)

1. Run a workflow that spawns multiple agents with isolation: "worktree"

1. After some time (sometimes within minutes), the agent subprocess

returns "Not logged in" without executing any work

## Expected behavior

OAuth token should persist across agent subprocesses within a session,
or token refresh should propagate to child processes automatically.

## Actual behavior
Each worktree agent spawn can independently fail auth, even immediately

after a successful /login in the main session. The token seems to

expire frequently and is not shared/refreshed for subprocesses.

## Auth status (redacted)

```json

{

"loggedIn": true,
"authMethod": "claude.ai",
"apiProvider": "firstParty",
"subscriptionType": "max"
}

```

## Notes

• No ANTHROPIC_API_KEY is set (company account, no individual API key access)

• This is a significant friction point for workflows that rely heavily

on agent worktrees (e.g., processing multiple Jira tickets in parallel)

### What Should Happen?

OAuth token should remain valid across agent subprocesses spawned

within the same session. Token refresh should propagate automatically
to child processes (worktree agents), so that a single /login at

session start is sufficient for the entire session duration.

### Error Messages/Logs

```shell
Agent worktree returned the following instead of executing the task:

"Not logged in · Please run /login"

No stack trace or additional error logs are produced.

The agent simply exits without performing any work.
```

### Steps to Reproduce

1. Authenticate via /login (OAuth, claude.ai Max plan, Organization account)

1. Confirm auth is valid: claude auth status shows loggedIn: true
2. Start a workflow that uses the Agent tool with isolation: "worktree":

• Example: any multi-step task that delegates subtasks to worktree agents

1. The first agent may succeed, but subsequent agents (or agents after

some elapsed time) return:

"Not logged in · Please run /login"

1. Run /login again — agent works

1. Repeat steps 3-5 throughout the day (~10 times)

Note: Exact token expiry timing is unpredictable. Sometimes fails

within minutes of login, sometimes after longer intervals.

### Claude Model

Opus

### Is this a regression?

I don't know

### Last Working Version

_No response_

### Claude Code Version

2.1.85 (Claude Code)

### Platform

Claude App (claude.ai)

### Operating System

macOS

### Terminal/Shell

Terminal.app (macOS)

### Additional Information

• This occurs on a company Organization account (not personal)
• No ANTHROPIC_API_KEY environment variable is set — the organization

does not provide individual API keys, so OAuth is the only auth method

• The workflow that triggers this most frequently uses multiple

sequential Agent calls with isolation: "worktree", each spawning

a new subprocess that needs to read the OAuth token

• Workaround: running /login each time the agent fails, but this

breaks automated pipelines and requires constant user intervention