What problem are you trying to solve
The reasoning effort parameter is hardcoded to "xhigh" for agent investigations, which can be overkill and increase cost. There is currently no flag to configure this without changing code, even though the Codex backend already reads config.reasoningEffort when provided.
What would you want deepsec to do
Add a flag to configure reasoning effort for agent backends. It should support values like "low", "medium", "high", and "xhigh", then pass the selected value into config.reasoningEffort. The effort value is already sent to the Codex SDK as modelReasoningEffort when starting investigation threads.
Workaround you're using today
I am using different model tiers, such as gpt-5.4 instead of gpt-5.5, to indirectly control cost. This does not change the reasoning effort setting, which remains "xhigh". The refusal follow-up already uses "low" effort, showing the system supports multiple effort levels.
What problem are you trying to solve
The reasoning effort parameter is hardcoded to "xhigh" for agent investigations, which can be overkill and increase cost. There is currently no flag to configure this without changing code, even though the Codex backend already reads config.reasoningEffort when provided.
What would you want deepsec to do
Add a flag to configure reasoning effort for agent backends. It should support values like "low", "medium", "high", and "xhigh", then pass the selected value into config.reasoningEffort. The effort value is already sent to the Codex SDK as modelReasoningEffort when starting investigation threads.
Workaround you're using today
I am using different model tiers, such as gpt-5.4 instead of gpt-5.5, to indirectly control cost. This does not change the reasoning effort setting, which remains "xhigh". The refusal follow-up already uses "low" effort, showing the system supports multiple effort levels.