Problem

Embedded agent sub-sessions repeatedly timeout on kimi/kimi-code with no fallback configured, causing silent failures in isolated cron job sessions.

Steps to reproduce

Run any isolated cron job that uses the embedded agent (e.g., daily-q, nightly review) — these run as sessionTarget: "isolated" with payload.kind: "agentTurn".

Expected behaviour

When the primary embedded model times out, the system should failover to a configured fallback model (sub_agents fallbacks in model_stack include openai-codex/gpt-5.2, openai-codex/gpt-5.4, zai/glm-5).

Actual behaviour

Log shows:

"embedded_run_failover_decision",
"failoverReason": "timeout",
"profileFailureReason": "timeout",
"provider": "kimi",
"model": "kimi-code",
"profileId": "sha256:cf78ee31c9b1",
"fallbackConfigured": false,
"timedOut": true,
"aborted": true

The embedded agent profile for kimi/kimi-code has fallbackConfigured: false despite global model_stack having fallbacks configured. No fallback fires — the job just fails.

Environment

• OpenClaw: 2026.4.27 (cbc2ba0)
• OS: Linux x64
• Node: v24.14.1
• Session type: isolated (cron job)