[BUG] Add/Update Skill intermittently forces full re-login instead of retrying transient auth 503

Status Open
Maintainer reply None cached
Activity 0 comments · opened Sep 1, 2026

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?

Clicking Add Skill or Update Skill in the Claude Desktop app's skill marketplace panel intermittently logs the user out of the app entirely and prompts for full re-authentication, instead of showing a retryable error. This happens across all skill sources — marketplace skills, self-authored/custom skills, and skills Claude itself generated during a session.

Log evidence shows the underlying cause is a transient 503 from a backend auth RPC used by the skills-list fetch, which the client mishandles as a hard logout rather than a retryable error.

What Should Happen?

If the skills-list fetch or install/update call fails due to a transient backend issue, the app should show a retryable error (e.g. a toast/banner with "try again"), not clear the session and force a full re-login.

Notably, the app's own background periodic skill sync already handles this same error class correctly — it retries and completes normally without logging the user out. Only the interactive Add/Update Skill action path appears to escalate it into a forced logout.

Error Messages/Logs

From %APPDATA%\Claude\logs\main.log:

2026-08-16 23:09:38 [info]  [SkillsPlugin] Window focused — polling now
2026-08-16 23:09:38 [info]  [SkillsPlugin] Starting skills sync
2026-08-16 23:09:38 [error] Failed to fetch skills list: 503 {
  "type": "error",
  "error": {
    "type": "overloaded_error",
    "message": "Authentication service was unavailable. You can try again.",
    "details": {
      "error_visibility": "user_facing",
      "error_code": "auth_rpc_unavailable"
    }
  },
  "request_id": "req_011Ce7C4famNoUWPAQmjkhj6"
}

A second, identical occurrence follows ~4 minutes later at 23:13:16 (request_id: req_011Ce7CLiSrbi826hukFqWsw).

Subsequent background syncs after these errors complete normally (Sync complete: 0 downloaded, 0 failed, 0 removed, 0 orphans cleaned, all skills intact) — the client clearly can recover from this error class, just not via the Add/Update Skill button path.

Steps to Reproduce

  1. Open Claude Desktop.
  2. Go to the skills panel.
  3. Click Add Skill (on a new skill) or Update Skill (on an existing one) — reproduces with marketplace skills, self-authored/custom skills, and Claude-generated skills alike.
  4. Occasionally (tied to backend auth-service load — not reproducible on demand), instead of the skill installing/updating, the app immediately logs out and shows the sign-in screen, requiring full re-authentication.

Claude Model

None

Is this a regression?

I don't know

Last Working Version

N/A / unknown

Claude Code Version

Claude 1.40609.1 (69aac0) 2026-08-30T23:12:49.000Z

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

This Bug happens through the desktop application, havent experienced it via CLI

View original on GitHub ↗