Anthropic OAuth token spontaneously expires/fails to refresh

Environment

• Clawdbot version: 2026.1.5-3
• OS: macOS Tahoe 26.1
• Auth mode: OAuth (anthropic:default with mode: "oauth")

Description

The Anthropic OAuth token randomly fails to refresh without any user action. The error appears spontaneously:

Error: Error: Error: OAuth token refresh failed for anthropic: Failed to refresh OAuth token for anthropic. Please try again or re-authenticate.

This requires running pnpm clawdbot update to reset the OAuth, which is disruptive and happens frequently enough to be a significant pain point.

Steps to Reproduce

1. Configure clawdbot with Anthropic OAuth authentication
2. Use normally
3. Wait (unclear trigger - happens spontaneously)
4. OAuth stops working with the refresh error

Expected Behavior

• OAuth token refreshes automatically in the background
• Or graceful degradation with clear user prompt to re-auth
• Ideally: persistent auth that doesn't randomly break

Actual Behavior

• Token refresh fails unpredictably
• Requires manual intervention (clawdbot update) to fix
• No clear trigger or warning before failure

Config (relevant section)

{
  "auth": {
    "profiles": {
      "anthropic:default": {
        "provider": "anthropic",
        "mode": "oauth"
      }
    }
  }
}

Questions for debugging

• Is there a token expiry log or timestamp I can check?
• Are there specific conditions that trigger this (time-based, request-based, network-based)?
• Is there a way to manually refresh without running update?

Frequency

~4 times in the first 24 hours of use. Highly disruptive.

Log entries

2026-01-07T03:49:35.410Z - [gws] ⇄ res ✗ chat.send 316ms 
  errorCode=UNAVAILABLE 
  errorMessage=Error: OAuth token refresh failed for anthropic: Failed to refresh OAuth token for anthropic. Please try again or re-authenticate.

2026-01-07T03:52:15.491Z - [gws] ⇄ res ✗ chat.send 393ms 
  errorCode=UNAVAILABLE 
  errorMessage=Error: OAuth token refresh failed for anthropic: Failed to refresh OAuth token for anthropic. Please try again or re-authenticate.

Note: Two failures within 3 minutes of each other, suggesting the token was already bad and multiple requests hit it before the user could re-auth.

Workaround

Running pnpm clawdbot update resets the OAuth and fixes the issue temporarily.