feat(cron): configure reasoning per job

[the3asic]

Summary

Adds an optional reasoning_effort override to each cron job, so scheduled work can choose its own quality/latency tradeoff without mutating global agent configuration.

Runtime precedence is:

1. valid per-job override
2. per-model reasoning policy for the model that actually survives provider-auth fallback
3. global reasoning policy
4. provider/model default

Behavior

• Missing or null inherits the configured policy.
• false / none explicitly disables reasoning.
• Supported levels come from the canonical parser: minimal, low, medium, high, xhigh, max, ultra.
• Invalid new writes are rejected with a clear error.
• Malformed hand-edited/legacy stored values warn and inherit instead of crashing the scheduler.
• hermes cron edit <job_id> --reasoning-effort inherit clears an override.
• Script-only (no_agent) jobs preserve the field but never construct an agent, so it has no runtime effect.
• Each run receives its own reasoning_config; no job can leak its override into another job or global config.

Surfaces

The field round-trips through cron storage, scheduler execution, the cronjob tool, standalone CLI, authenticated dashboard API, web UI, desktop UI/types/translations, and cron documentation.

Existing jobs require no migration and retain their current behavior.

Current-main compatibility

This is rebased onto current main and preserves the newer auth-fallback ordering: reasoning is resolved only after the actual fallback model is known. It also covers the gaps called out on earlier proposals: all current effort levels, raw YAML/JSON false, malformed-record fallback, standalone CLI set/clear, and both web and desktop surfaces.

Related: #7382, #63327, #23524.

Validation

• 829 Python tests collected and passed across tests/cron, cron tool, CLI parser/lifecycle, and dashboard profile API suites

• Web + desktop cron model Vitest: 16 passed

• Web TypeScript typecheck: passed

• Desktop renderer + Electron TypeScript typechecks: passed

• ESLint on every changed frontend file: passed

• Ruff on changed Python files: passed

• py_compile: passed

• git diff --check: passed

• Added-line private identifier, IP, credential-literal, and control-character scans: zero hits

• Tested on Ubuntu 26.04 (x86_64)

[alt-glitch]

This was generated by AI during triage.

Duplicate of #7382 — this implements the same per-job reasoning_effort storage and scheduler precedence. The additional UI and validation coverage matches the existing cluster rather than a separate mechanism.

[tonydwb]

Code Review Summary

Verdict: Comment

Overview

• Cron feature: configure reasoning per job
• 732 additions, 20 deletions — moderate size

Assessment

• Per-job reasoning configuration is useful for resource management
• No obvious security concerns

Note

• Given the moderate size, recommend verification that the reasoning configuration is properly validated

---

Reviewed by Hermes Agent

[the3asic]

Closing in favor of #7382, which covers the same per-job reasoning-effort mechanism and remains the active implementation. Thanks for the triage.