Bug Description

The gateway process crashes entirely when a channel's stored claudeCliSessionId references a Claude Code CLI session that no longer exists. Instead of gracefully falling back to a new session, the unhandled error takes down the whole gateway.

Steps to Reproduce

1. Have a channel with an active CLI session (stored as claudeCliSessionId and cliSessionIds in sessions.json)
2. Wait for the CLI session to expire/get cleaned up, or trigger a heavy task that invalidates it (in my case, filing a GitHub issue via the gh skill)
3. Post a new message in that channel
4. Gateway attempts --resume {sessionId}, gets "No conversation found with session ID: {id}", and crashes

Error

Agent failed before reply: No conversation found with session ID: 26161484-7b46-486c-aa83-6d020ae3433e.

The gateway process exits entirely — no recovery, no fallback. The bot goes fully offline (no typing indicator, no response in any channel including DMs).

Expected Behavior

When --resume fails with a "session not found" error, the gateway should:

1. Clear the stale claudeCliSessionId / cliSessionIds from the session entry
2. Start a fresh CLI session for the channel
3. Continue operating normally

Workaround

Manually edit ~/.clawdbot/agents/main/sessions/sessions.json and delete the claudeCliSessionId and cliSessionIds fields from the affected channel entry, then restart the gateway.

Environment

• Clawdbot version: 2026.1.24-3
• OS: Windows 11 Pro
• Node: 22.17.0
• CLI backend: claude-cli with --resume / --session-id via wrapper script