Menu

1 / 2

[Bug]: Reasoning settings and auxiliary output-budget overrides do not reach requests

GitHub Public Timeline Feed·gcp·about 10 hours ago
#TdrMXcFB
Reading 0:00
15s threshold

Bug description

Some explicit reasoning/output-budget settings cannot reach the request that needs them. This makes reasoning-only models difficult to use for titles and short smart-approval verdicts, and prevents a user from correcting a provider rejection through configuration.

Verified on main b9cb7a243a0a76bf76287b75a1065fe1e851e391 :

agent/title_generator.py::generate_title always passes reasoning_config={"enabled": False} . That caller value overrides auxiliary.title_generation.reasoning_effort , including an explicit configured level.

Titles and smart approval hardcode max_tokens=512 and 16 respectively. A per-task max_tokens setting does not override those caller budgets. On routes where reasoning consumes the output budget, it can leave no visible title/verdict.…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.