Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Cron-executed jobs do not honor explicit model selection via either payload.model or agentId.

Despite correct configuration, all cron jobs resolve to the default/main agent model at runtime.

This prevents assigning different models per job and breaks mixed-model workflows (e.g., running most jobs on Haiku while reserving GPT-5.4 for higher-quality tasks).

Steps to reproduce

1.	Configure an agent with a specific model in openclaw.json:

{ "id": "gpt5-operations", "model": "openai/gpt-5.4" }

1. Configure a cron job in cron/jobs.json:

{ "name": "weekly-intelligence-operation", "agentId": "gpt5-operations", "payload": { "model": "openai/gpt-5.4" } }

1. Run the cron job. 4. Inspect the resulting session log:

grep "model" /data/.openclaw/agents/main/sessions/<session-id>.jsonl

Observed result

The session executes as:

{ "provider": "anthropic", "model": "claude-haiku-4-5" }

Expected behavior

The session should execute as:

{ "provider": "openai", "model": "gpt-5.4" }

Actual behavior

cron-executed jobs ignore both payload.model and agentId.

All cron jobs execute using the default/main agent model, regardless of explicit configuration.

In this case, the job configured with: • agentId: gpt5-operations • payload.model: openai/gpt-5.4

still executed as:

{ "provider": "anthropic", "model": "claude-haiku-4-5" }

OpenClaw version

OpenClaw 2026.3.28 (f9b1079)

Operating system

Ubuntu 24.04.4 LTS (VPS)

Install method

Docker (VPS deployment)

Model

claude-haiku-4-5 (actual execution) openai/gpt-5.4 (expected but not honored)

Provider / routing chain

openclaw → cloudflare-ai-gateway → minimax

Additional provider/model setup details

Issue occurs consistently in Docker deployment
•	Direct sessions_spawn with explicit model works outside cron
•	Cron-dispatched sessions appear to ignore both payload.model and agentId
•	All cron jobs resolve to the main/default agent model at runtime

Logs, screenshots, and evidence

Cron job configuration:

{
  "name": "weekly-intelligence-operation",
  "agentId": "gpt5-operations",
  "payload": {
    "model": "openai/gpt-5.4"
  }
}

Session log (actual execution):

{
  "provider": "anthropic",
  "model": "claude-haiku-4-5"
}

This behavior is consistent across multiple cron jobs.

Impact and severity

•	Affected systems: OpenClaw cron execution (all cron jobs)
•	Severity: High (breaks model routing and mixed-model workflows)
•	Frequency: Always (100% reproducible across all tested jobs)
•	Consequence:
•	payload.model is ignored
•	agentId is ignored
•	All cron jobs execute using the default/main agent model
•	Cannot assign different models per job
•	Prevents using higher-capability models (e.g. GPT-5.4) for specific tasks
•	Previously caused unintended high-cost usage (before mitigation)

Additional information

No response