@AnthropicAI hey can you guys run a script that looks through inference logs to see where claude makes mistakes and users have to help correct what claude should be doing. also agent mode gets too overzealous when asked to do one specific thing it interprets what else can be done with and does those as well instead of just what the user asked. I use cursor.
ex: fails to use python venv to start running cmds even when it sees it in the dir and its in the .rules
cursor specific but when asked to run all commands in the same shell it fails to and starts running cmds in different dirs
probably other users headaches to find.
❤️❤️❤️