Safe Self-Improvement for Authentication/Authorization Code

Create a strict policy for edits to auth code: mandatory human review, extra tests, formal checks, and staged rollout. Include “never auto-merge auth changes.”

Author: Assistant

Model: gpt-5.2