openclaw - 💡(How to fix) Fix [Bug]: Cron payload.model does not resolve configured model aliases before allowlist validation [2 comments, 3 participants]

Summary

After upgrading to OpenClaw 2026.4.29, cron jobs with payload.model set to configured model aliases such as easy, medium, or hard fail model allowlist validation. The aliases are configured and visible via openclaw models aliases list, but cron appears to validate the raw payload string as a provider/model ref before alias resolution.

Environment

• OpenClaw: 2026.4.29 (a448042)
• OS: Linux x64, kernel 6.17.0-23-generic
• Node: 25.8.0
• Gateway: systemd service, local loopback reachable

Config

Aliases:

hard   -> openai-codex/gpt-5.5
medium -> lmstudio/qwen3.6-35b-a3b
easy   -> ollama/bcluzel/LFM2.5-1.2B-Instruct:Q4_K_M

Allowlist:

{
  "openai-codex/gpt-5.4": {},
  "openai-codex/gpt-5.5": { "alias": "hard" },
  "lmstudio/qwen3.6-35b-a3b": { "alias": "medium" },
  "ollama/bcluzel/LFM2.5-1.2B-Instruct:Q4_K_M": { "alias": "easy" }
}

Reproduction

1. Configure a model alias, e.g.:

   openclaw models aliases add medium lmstudio/qwen3.6-35b-a3b

2. Create or update a cron job with:

   {
     "payload": {
       "kind": "agentTurn",
       "model": "medium",
       "message": "..."
     }
   }

3. Run the cron job.

Actual behavior

The job fails before execution with allowlist validation errors like:

cron payload.model 'medium' rejected by agents.defaults.models allowlist: openai-codex/medium

Similarly:

cron payload.model 'easy' rejected by agents.defaults.models allowlist: openai-codex/easy
cron payload.model 'hard' rejected by agents.defaults.models allowlist: openai-codex/hard
cron payload.model 'openai-codex/medium' rejected by agents.defaults.models allowlist: openai-codex/medium

This suggests cron is either not resolving aliases at all, or is resolving bare strings through the default provider before checking configured aliases.

Expected behavior

Cron payload.model should follow the same model ref resolution rules as documented for model selection:

• alias match first
• then unique configured-provider match
• then provider/default fallback behavior if applicable

So payload.model: "medium" should resolve to lmstudio/qwen3.6-35b-a3b before allowlist validation.

Alternatively, if cron intentionally does not support aliases, the cron update/create path should reject aliases immediately with a clear validation error instead of accepting the job and failing later at runtime.

Workaround

Patch cron jobs to use full model refs instead of aliases:

easy   -> ollama/bcluzel/LFM2.5-1.2B-Instruct:Q4_K_M
medium -> lmstudio/qwen3.6-35b-a3b
hard   -> openai-codex/gpt-5.5

This works but makes future model migrations brittle because every cron job must be manually rewritten whenever aliases change.

Related

Possibly related to model allowlist / cron model validation regressions such as #75753 and older alias-resolution issue #54166.