400 diagnostics.previous_message_id error after usage-limit resume

Summary

After a Claude Code session hits the 5-hour usage limit and resumes, every subsequent turn fails with:

API Error: 400 diagnostics.previous_message_id: must be the `id` from a prior /v1/messages response (starts with `msg_`)

The session is stuck until restarted.

Reproduction

1. Active Claude Code session hits the usage limit and pauses ("sautéed").
2. Limit resets, session resumes.
3. Send any user message after resume.
4. API returns 400 on diagnostics.previous_message_id.
5. Every subsequent turn fails identically until the session is restarted.

Expected behaviour

On the first post-resume turn, Claude Code should either:

• Replay a valid msg_… ID from before the pause, or
• Omit diagnostics.previous_message_id entirely so the API doesn't reject the request.

Workaround

Exit the Claude Code session and start a new one. /clear may also work but I haven't fully verified.

Environment

• OS: macOS (Darwin 25.5.0)
• Model: claude-opus-4-7
• Channel: terminal CLI

Happy to provide session IDs or transcript snippets if useful.