Fix OpenClaw “LLM request failed”

Evidence-first sequence

1. 1

   Confirm the selected model

   Record the exact provider/model and fallback list.

   openclaw models status --json
   openclaw models list

   Copy
2. 2

   Inspect auth state

   Do not print the underlying token or API key.

   openclaw models auth list --json

   Copy
3. 3

   Run doctor

   Look for provider, SecretRef, routing, plugin, and config findings.

   openclaw doctor --json

   Copy
4. 4

   Test the narrowest path

   Use a small direct provider/local-model probe before a full tool-rich agent turn.

Interpret the category, not just the message

• 401/403: credential, account, method, or permission mismatch.
• 404/unknown model: wrong provider/model reference or unavailable catalog entry.
• 429/quota: rate, balance, concurrency, or account-plan limit.
• Timeout/network: DNS, proxy, TLS, firewall, endpoint, or provider availability.
• Local model plain-text success but agent failure: tool calling, context, or agent-capability mismatch.
• Fallback success: useful availability evidence, but the primary failure still needs an explicit decision.