Always something to change when Claude reviews its own work?
retained text
Over the past year I have been hesitant to lock myself into one AI provider. However, recently I have leaned on Claude more and more. With this I experience a concerning trend. For each review there is something wrong, missing, to be modified etc. Today, I am in what appears to be an endless loop with Claude when using Opus 5.5 for specifying, planning, and preparing for implementation. Decisions made are suddenly not clear in the documents, reiterated challenges pop up in repeated iterations, and so on. I spent 3 days going through the specifications, and planned out each implementation step and conversion from the previous version, today. Now I am on the 3rd iteration of the revised implementation plan and it is still going on! Yeah, manually reviewing and typing is not respected much either. New sessions, CLI vs app vs IDE harness does not change this for me. Any idea what I could do differently (besides one-shot it :)? Using Opus 5.5 high
document fetched 2026-11-12 00:42 · canonical https://reddit.com/r/ClaudeCode/comments/1xtf5f5/always_something_to_change_when_claude_reviews