fix(cron): wire max_tokens from per-job config into AIAgent constructor

[AlexFucuson9]

Summary

run_job() in cron/scheduler.py reads several job-config keys (model, reasoning_config, toolsets, etc.) into the AIAgent(...) constructor but never reads max_tokens. A cron job that sets max_tokens: 4096 gets it silently ignored.

Impact

Any cron job whose expected output exceeds the model's default max-output behavior fails with RuntimeError: Response remained truncated after 3 continuation attempts — and setting max_tokens in that job's config to fix it does nothing, silently.

Fix

Added max_tokens=job.get(max_tokens) to the AIAgent() constructor call in run_job().

Root Cause

AIAgent.__init__ accepts max_tokens: int = None and the plumbing exists everywhere downstream (agent_init.py → build_api_kwargs()). The only missing link was the constructor call in cron/scheduler.py.

Fixes #58423

[alt-glitch]

This was generated by AI during triage.

Duplicate of #58440 — both fix #58423 by wiring max_tokens from the per-job config into the AIAgent(...) constructor in run_job() (cron/scheduler.py). #58440 was opened earlier and is slightly more robust (adds int-validation + a warning on a bad value), so it is the canonical fix.

[teknium1]

Thanks for tracing the missing constructor forwarding; the diagnosis and one-line implementation are clear.

This is an automated hermes-sweeper review. This change is being closed under the standing max-tokens-knob policy: Hermes does not add user-facing max_tokens configuration surfaces, including per-cron-job overrides.

• Current main constructs cron agents at cron/scheduler.py:3171-3202 without a per-job max_tokens path; adding job.get("max_tokens") would create the prohibited per-job knob.
• The supported configuration is the global model.max_tokens setting, documented at website/docs/integrations/providers.md:1032-1036 and validated in agent/agent_init.py:1646-1672.
• The related canonical PR fix(cron): read per-job max_tokens from job config and pass to AIAgent #58440 was closed unmerged on the same policy, and the later duplicate fix(cron): honor per-job max tokens #61149 was also closed under max-tokens-knob.

For truncation cases, please use or improve the documented global model.max_tokens configuration and investigate provider context/output-cap settings rather than adding a per-job override.

---

Closed as not-planned per standing maintainer policy (max-tokens-knob). This is a design-direction decision, not a code-quality judgment — see the Contribution Rubric in AGENTS.md for what the project is looking for. If you believe this policy was misapplied to your change, comment here and a maintainer will take a look.