Summary

Issue #78000 was fixed on main (commit 428cc54) — Anthropic API-key config defaults were supposed to backfill anthropic/claude-haiku-4-5 into agents.defaults.models automatically. That fix has not survived into v2026.5.28.

After upgrading from v2026.5.22 → v2026.5.28 today, all cron jobs with payload.model = anthropic/claude-haiku-4-5 are failing fleet-wide with:

FailoverError: Unknown model: anthropic/claude-haiku-4-5. Found agents.defaults.models["anthropic/claude-haiku-4-5"], but no matching models.providers["anthropic"].models[] entry. Add { "id": "claude-haiku-4-5" } to models.providers["anthropic"].models[] to register this provider model.

Environment

• OS: macOS Darwin 25.5.0 (arm64)
• Hardware: Apple Mac Mini (M4)
• Previous version: v2026.5.22 (stable, haiku-4-5 crons working fine)
• Upgraded to: v2026.5.28

Affected agents

5 of 17 agents affected: Kubera, Agastya-CAIO, Agastya-Ravindra, Agastya-Sandip, Agastya-Keshav. All running standard Anthropic API key config with claude-haiku-4-5 referenced in cron payloads.

Steps to reproduce

1. Run fleet on v2026.5.22 with cron jobs using payload.model = anthropic/claude-haiku-4-5
2. Upgrade to v2026.5.28 via npm install -g [email protected]
3. Restart all gateways
4. Observe cron failures immediately on next scheduled run

Workaround

Manually adding { "id": "claude-haiku-4-5" } to models.providers["anthropic"].models[] in each affected agent's openclaw.json resolves the failures immediately.

Expected behaviour

The backfill introduced in the #78000 fix (commit 428cc54) should prevent this automatically on upgrade — haiku-4-5 should be auto-registered whenever an Anthropic default model is configured.

Notes

This is a silent breaking change for any fleet using haiku-4-5 as a cron model override. No warning during upgrade, no migration path. Affects multi-agent deployments disproportionately — each agent instance requires the manual fix independently.