Description

Isolated cron jobs consistently fail with LiveSessionModelSwitchError after updating to OpenClaw 2026.3.28. The error references anthropic/claude-opus-4-6 even when the cron is configured with a different model.

Affected cron jobs

• morning-briefing
• weekly-summary
• usage-sync-supabase

All three are isolated cron jobs (not running in the main session).

Error

LiveSessionModelSwitchError: ... anthropic/claude-opus-4-6

Steps to reproduce

1. Update to OpenClaw 2026.3.28
2. Have isolated cron jobs configured (with sessionKey or without)
3. Wait for cron to fire
4. Cron fails with LiveSessionModelSwitchError

What was tried (all failed)

• Clearing sessionKey from cron config
• Clearing agentId from cron config
• Restarting the gateway
• Changing the primary model to GPT-5.4
• Setting model: default on the cron job
• Various combinations of the above

Expected behavior

Isolated cron jobs should execute without model switch errors, using either the configured model or the system default.

Actual behavior

Cron jobs fail immediately with LiveSessionModelSwitchError. The error accumulates (consecutiveErrors climbing, reached 5+ for usage-sync-supabase).

Workaround

Generating briefings/summaries via sessions_spawn subagents works fine in isolated sessions. The bug appears specific to the cron runner in isolated mode.

Environment

• OpenClaw: 2026.3.28
• Host: macOS (Apple Silicon)
• Primary model: anthropic/claude-opus-4-6
• Node: v25.8.2