多 Agent Cron 模型一致性：避免 custom provider 被改写后 `model not allowed`

解决非主 Agent 的 cron 任务运行时报 `model not allowed`：对齐 provider 名称与 allowlist，并用手动触发回归验证。

GITHUBDiscovered 2026-02-14Author widingmarcus-cyber

Prerequisites

You run cron jobs in non-primary agents (`sessionTarget=isolated`).
Model/provider allowlist is configurable and observable in current gateway config.

Steps

Pick one failing cron job and record current payload.model plus actual runtime-resolved model/provider.
Normalize provider naming and keep explicit model IDs/aliases consistent across job config and allowlist.
Run `cron run` manually on the same non-primary agent job; compare run logs before/after change.
After passing manual run, observe at least 2 scheduled cycles to ensure no silent model remap reappears.

Commands

openclaw cron list

openclaw cron run --job <jobId>

openclaw cron runs --job <jobId>

openclaw logs --local-time

Verify

Cron runs on non-primary agents use the intended model/provider and no longer fail with `model not allowed`.

Caveats

Mixed alias/full-id usage can hide mapping bugs; keep one canonical style per environment（需验证）.
Do not mass-edit all cron jobs before validating one canary job first.

Source attribution

This tip is aggregated from community/public sources and preserved with attribution.