Aleksey Nogin
Aleksey Nogin
> Found 2 possible duplicate issues: > > 1. [Feature Request: A Structured Workflow for Complex, Long-Running Tasks to Improve Agent Reliability and Focus #5996](https://github.com/anthropics/claude-code/issues/5996) > 2. [Feature Request: Advanced...
I believe this is still relevant.
Not 100% sure this is the best, but it probably is. Also as this (by necessity) resets `self._total_samples` to 0, should it be doing the same to `self._lost_samples` and `self._corrupted_samples`?
I enabled `CLAUDE_BASH_MAINTAIN_PROJECT_WORKING_DIR` and so far had 0 issues of Claude being confused by the directory *not* changing - so maybe that should just be the default? Perhaps with an...
The issue may not be as bad more recently, but there is definitely still some tendency to over-optimize how little is read (at least when not using the Explore agents).
I believe this is still relevant
Now, with the Sonnet 4.5 being much more "I know better than you" than previous model, this is even worse! I just had (in a fresh session): ``` > Review...
I think this might be correlated with the file size - perhaps something like @-mentioned files longer than 1000 lines are not actually included in the context, so Claude has...
Earlier today I realized that this is more important than I originally thought - Claude Code was stopped waiting for me to approve/disapprove it plan, and I killed and restated...
And the exact same issue for the new clarification Q&A UI! The new clarification UI is extremely cool and very convenient, but suffers from the same exact issue - whereas...