Allow AGENTS.md policies and Hooks to dynamically escalate/de-escalate reasoning effort during a session

What variant of Codex are you using?

codex desktop Version 26.602.40724 • Released 5 Jun 2026

What feature would you like to see?

Today AGENTS.md can express policies such as:

• low complexity
• moderate complexity
• high reasoning

However these policies cannot automatically influence the active reasoning effort.

Users must manually change:

GPT-5.5 Low
→ GPT-5.5 Medium
→ GPT-5.5 High

even when Codex can clearly determine that task complexity has changed.

Expose reasoning effort as a runtime-adjustable setting.

Allow:

• Hooks
• Future profile managers
• Agent orchestration logic

to request effort changes.

Example:

if architecture_detected:
effort = high

if repetitive_edits:
effort = low

• Lower token consumption
• Better cache reuse (same model, different effort)
• Better UX
• Supports policy-driven workflows
• Enables adaptive reasoning

Additional information

_No response_