| A common frustration when working with code generation tools in Python projects is their tendency to use system Python. I've tried to add as a commonly added workspace rule to instruct the agent to always create a virtual environment using UV for package management. However, even when that's instructed in Claude.md, the agent often fails to adhere to the instruction and tries to install packages into system Python. <br><br>So I frequently find myself having to prompt, "please activate the virtual environment, please use the virtual environment." What I've noticed is that even when it does this, it sends the activation, VendActivation command in each bash completion, which is inefficient. A human developer would activate the virtual environment once and then run commands regularly within the activated virtual environment. <br><br>A more elegant solution would be for Claude code to have the ability to detect, to not only create virtual environments for instructions, but to detect them based on their folder pattern, activate them where they exist, and use them in a more efficient manner by activating them once and then holding that as a persistent bash environment for the session. |