Token Usage Transparency, Context Controls & Approval Gates for Composer/Auto Mode

Ideas Feature Requests
, ,
1

Feature request for product/service

Cursor IDE

Describe the request

Problem

There are cases where a simple request unexpectedly results in massive token consumption (millions of tokens).

While prompting style plays a role, users currently lack:

  • Clear visibility into what context is being sent
  • Fine-grained control over how much context is included
  • Approval mechanisms before large token calls execute
  • Tooling to intelligently optimize context for longevity and cost efficiency

For teams working in large codebases, this creates:

  • Cost unpredictability
  • Fear of using Auto / Composer freely
  • Reduced trust in background agent behavior

Requested Features

1. Pre-Execution Token Estimate + Approval Gate

Before Composer / Auto executes a large call:

  • Show:

    • Estimated input tokens
    • Estimated output tokens
    • Estimated cost

  • If over user-defined threshold → require confirmation

Example:

:warning: This action will use ~480,000 tokens (~$X).
Continue? [Approve] [Cancel] [Optimize Context]

User-configurable thresholds:

  • Hard cap (never exceed)
  • Soft cap (ask before exceeding)
2. Context Visualization Panel

A live “Context Inspector” showing:

  • Files included
  • Snippets included
  • Conversation history length
  • System rules (.cursorrules)
  • Memory injections
  • External tool outputs

With the ability to:

  • Toggle items on/off
  • Collapse large files
  • Replace full file with summarized version
  • Pin essential files

This makes context intentional instead of opaque.

3. Smart Context Optimizer

Before sending context, allow:

  • “Optimize Context” button that:

    • Deduplicates repeated code
    • Removes irrelevant history
    • Summarizes large files
    • Converts full files to embeddings when possible
    • Switches to structural diffs instead of raw file dumps

Goal: same intelligence, fewer tokens.

4. Persistent Context Strategy Modes

Allow users to choose per-project:

  • :locked: Strict Mode – minimal context, manual inclusion only
  • :balance_scale: Balanced Mode – auto include but capped
  • :rocket: Full Auto Mode – current behavior

This gives professionals control without disabling advanced features.

5. Composer Subagent Transparency

If Composer 1.5 spawns subagents or expands context automatically:

  • Show when it happens
  • Show why
  • Show how many tokens that branch used

This is not about restricting capability — it’s about transparency.

Why This Matters

Large codebases require longevity of token budgets.

Better context controls would:

  • Increase trust
  • Reduce billing shock
  • Encourage heavier usage (because users feel safe)
  • Make Cursor viable for long-term professional workflows

Right now context feels “magic.”
It needs to feel “intentional.”

If anyone else has experienced unexpected token spikes, I’d like to hear how you handled it.

1 Like
3

Soft Cap would be a great start. The unexpected million token usage is brutal.