[Model behavior] Told to read memory, the model writes a duplicate memory entry instead

After a failure, the user told the model to read its memory. Instead of reading the existing memory files, the model wrote a new entry restating a rule that was already recorded in several existing memory files. The user had to tell it again to read them; the model then reverted its own edit.

Steps to reproduce

1. Have a persistent memory directory with explicit rules (e.g. "PR text is a claim, not evidence").
2. Let the model violate one of those rules, then say "read the memories".
3. Observe: the model edits/adds memory content instead of reading the files that already contain the rule.

Expected

"Read the memories" results in reading them, then applying them to the task at hand; no duplicate entries.

Model: Claude Opus 5.5 (claude-opus-5-5[1m]), Claude Code CLI, session on 2026-10-05.
Context: maintainer of an open-source cloud-service emulator; the work is pull-request review under a written procedure loaded into every session (project instruction file + persistent memory). No personal data, repository names, PR numbers or handles are included.