Examples I'm seeing:
- Incomplete responses: Asked if 3 files were still needed post-refactoring; got an answer for only 1, other 2 ignored completely
- Context failures: Asked for the deploy command after modifying our Ansible script; gave wrong answer without checking obviously related files Code quality regression: Simple refactors that used to work smoothly now require multiple iterations
- Agent execution issues: Agents not triggering without explicit @ mentions. Spent 1+ hour getting a 'cypress-test-runner' agent to work despite clear instructions in CLAUDE.md. Even when triggered, it couldn't follow basic instructions (kept paraphrasing test names instead of returning them unchanged)
Has anyone else noticed the same?