When vibe-coding, a lot of tokens are being wasted because the agent decides to create an inline script. It assumes the shell (*nix most of the time) or eventually forgets in what shell it's running after you've made it noticed, ending these requests in syntax errors all around.
Even if you ask the agent to write the script into a file, tokens have already been wasted. Vibe-coding agents often go to use inline scripts even when they have already the tool available in their workspace.
As a consequence, a lot of your daily and weekly allowance is wasted because the vibe-coding agent fails to remember in what shell they're sandboxed in.