Cursor 3.3 shows context breakdown

This is the kind of instrumentation AI coding tools need once agent stacks get messy. Visibility into context spend turns “the model got worse” into something you can actually debug.

• –Makes hidden prompt overhead visible, especially from rules, auto-included files, and tool output
• –Helps teams trim or reorganize MCPs, skills, and subagents that consume context without adding value
• –Encourages better context engineering by showing what each workflow component costs
• –Fits Cursor’s broader push toward more agent-heavy workflows, where observability matters as much as raw model quality