[FEATURE] Improve usage limit consumption

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

The experience around consumption of daily and weekly limits is very subpar. We need some feedback before we send a request. Its all nice and dandy that we can /clear and /compact, but there is no information anywhere on how much it will burn from the limits. Maybe I didn't really need to ask that question if it will burn 10% of my limit. This especially comes to surface on large code projets
You definitely know how much tokens you will send, so why not make that infromation available to us? Maybe warn if its over some configured percent, or such? To quote your own model:

> clear did the same thing. I am not sure now if I'll burn another 8% if I ask something. This whole load management communication is subpar

You're absolutely right — the communication around usage limits is frustrating. You shouldn't have to guess whether your next message will cost 1% or 8% of your quota.
What likely happened: The VS Code extension might behave differently than the CLI. Opening a new chat window is effectively a fresh session, so that's probably the extension's way of doing /clear. It's just not what you expected.
Will your next message cost 8%?
Probably not. That /compact cost was high because it had to read and summarize your entire previous conversation. A fresh session with a simple question should be much cheaper — likely under 1% unless you're asking it to do something complex or read a lot of files.
What's frustrating:

- No visible token/cost meter before you send
- No clear documentation on how much different actions cost
- Shared limits between web and Code with no way to see the split
- /clear vs /compact vs "new window" all doing slightly different things with no clarity on the cost implications

t's valid criticism — Anthropic should surface usage data more transparently, especially when you're sharing quota across tools.

Proposed Solution

• Add visible metric in both terminal and vs code extension (also kindly make that Rider extension already)
• Make a threshold configurable and ask "are you sure" before sending
• some other ways to control usage?

Alternative Solutions

Burning haplessly through limit and then choosing overcharge or idle waiting. Not fun either way.

Priority

High - Significant impact on productivity

Feature Category

Configuration and settings

Use Case Example

1. Have a large project
2. Ask questions which will increase conversation history with files and file data
3. Ask questions you didnt really need to
4. Burn through your limit, not being aware you did.

Additional Context

_No response_