Bug

When a specific model (e.g. anthropic/claude-sonnet-4-6) hits a rate limit on Claude Max, the cooldown is applied to the auth profile (provider-level), not the specific model. This prevents automatic failover to a different model on the same provider (e.g. anthropic/claude-opus-4-6) even though that model is available and not rate-limited.

Steps to Reproduce

1. Configure agents.defaults.model with primary anthropic/claude-sonnet-4-6 and fallback anthropic/claude-opus-4-6
2. Use Claude Max (token auth) which has per-model rate limits
3. Hit the Sonnet rate limit
4. Expected: automatic failover to Opus (same provider, different model, not capped)
5. Actual: entire Anthropic profile enters cooldown, Opus is also blocked

Workaround

Manually switching with /model anthropic/claude-opus-4-6 works because user overrides bypass cooldown logic.

Suggestion

Cooldowns should be tracked per provider + model, not just per auth profile, since providers like Anthropic (especially Claude Max) enforce rate limits per model, not per account.

Environment

• OpenClaw 2026.3.8
• Auth: Claude Max token (single profile)
• macOS, local gateway