Feature Request

After a gateway restart (planned or crash recovery), all in-flight work is lost:

• Active agent sessions lose their conversation context
• Running cron jobs are silently dropped
• There's no mechanism to resume or retry interrupted work

Current Behaviour

1. Gateway crashes or restarts
2. Watchdog detects and restarts gateway (~5 min)
3. All in-flight sessions and cron runs are gone
4. Agent wakes up fresh with no knowledge of what was in progress
5. Cron jobs that were mid-execution are not retried until their next scheduled time

Desired Behaviour

Session recovery:

• On restart, the gateway should detect sessions that were active at shutdown
• Inject a system event into recovered sessions indicating the restart (e.g. [System] Gateway restarted. Previous session context may be incomplete.)
• Optionally: persist session state to disk so context survives restarts

Cron run recovery:

• Track in-flight cron runs in a durable store (e.g. SQLite or file)
• On restart, check for interrupted runs
• Re-queue interrupted runs with a flag indicating they're retries
• Respect a configurable retry policy (e.g. max retries, backoff)

Wake mechanism:

• After successful restart, automatically send a wake event to all agents that had active sessions
• This ensures agents can check for in-progress work rather than waiting for their next heartbeat

Workarounds Currently in Use

• External watchdog script (watchdog.sh) handles restart detection
• Manual cron wake event after restart to kick agents
• Daily memory files for manual context recovery
• openclaw-safe-restart scripts for planned restarts

Impact

This is especially important for:

• Long-running cron jobs (e.g. email cleanup that takes 9+ minutes)
• Multi-step agent workflows that get interrupted
• Users with longer heartbeat intervals (1-2h) who won't notice the gap

Environment

• macOS, LaunchAgent-based gateway
• Watchdog runs every 5 minutes
• Multiple agents (kit, cron-bot, etc.)