reflect

To stop repeated AI behavior mistakes, you can convert recurring user corrections into a single targeted update to the underlying rule, skill, or agent definition file. This changes the behavior at the source instead of requiring manual correction each time.

Updating a CLAUDE.md file to fix recurring prompt behavior involves detecting repeated corrections, routing the issue to the correct file, and generating a unified diff proposal. The proposed edit is then presented for user approval before being applied.

A two-strike correction handling workflow for prompt governance detects when a current correction matches a prior correction in memory. It then treats this repetition as a behavioral pattern worth addressing structurally through a targeted rule update.

You can refine existing agent rules without duplicating guidance by performing a full-file review and collision checking for existing coverage. The system evaluates whether related guidance already exists before proposing either a refinement or a focused addition.

Memory matching for tuning AI writing style and workflow rules works by comparing a current user correction against prior corrections stored in memory. When a match is found, it triggers a structural rule proposal rather than a temporary behavioral fix.