Unrecoverable 500 errors requiring /clear or new session

Description

Experiencing 500 Internal Server Errors that do not resolve with retries. The only workaround is /clear or starting a new session. This happens 5-6 times daily.

Behavior

1. Claude Code is performing an operation (e.g., Read)
2. Extended "Cogitated" period (1m 51s in example below)
3. 500 error occurs
4. Retrying (even after waiting several minutes) immediately fails with another 500
5. Only /clear or new session resolves it

Example

⏺ Read(lib/stores/announcement-store.ts)
  ⎿  Read 82 lines
  ⎿  Loaded CLAUDE.md
  ⎿  Loaded CLAUDE.local.md
  ⎿  API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_011CXHRosDxMhpKEJUZrma8V"}

✻ Cogitated for 1m 51s

❯ continue
  ⎿  API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_011CXHSGb1989uqZPLCiJzd3"}

The retry ("continue") was sent after waiting several minutes - not immediately.

Request IDs

• req_011CXHRosDxMhpKEJUZrma8V
• req_011CXHSGb1989uqZPLCiJzd3

Expected Behavior

• Transient 500 errors should resolve with retry
• Conversation state should not become permanently corrupted

Actual Behavior

• 500 errors persist indefinitely for the affected conversation
• Only /clear or new session fixes it
• Happens multiple times daily (5-6 times reported)

Environment

• Date: 2025-01-19
• Model: claude-opus-4-5-20251101
• Status page shows all systems operational during these failures