hermes - 💡(How to fix) Fix Cron scheduler caches null model at job creation; config updates don't invalidate cache

Summary

When a cron job is created without an explicit model field, the scheduler caches model=null at creation time. Subsequent config updates don't invalidate the cache — the null value persists and is sent to the LLM API at runtime, causing rejection with invalid_request_error: model: String should have at least 1 character.

Steps to Reproduce

1. Create a cron job without specifying a model:

   cronjob action=create name=test schedule="*/5 * * * *" prompt="test"

2. Job config is written with model: null.
3. Update the job to set an explicit model:

   cronjob action=update job_id=<id> model=claude-haiku-4-5

4. Config file is updated correctly, but the in-memory cache is not invalidated.
5. The next scheduled run sends the cached null model to the API → 400 error.

Expected Behavior

The scheduler should load the model from config at runtime instead of caching it at creation time. Alternatively, config updates should invalidate the cache.

Current Workaround

Delete the affected job and recreate it with an explicit model field at creation time.

Impact

Six or more production jobs affected in our deployment. All fail silently with the same 400 error, making it non-obvious that the cache is the root cause until the config file is inspected.

Reported via collaboration between a Hermes agent ("JARVIS") and a Claude Code session on the user's Mac.