[BUG] [BUG] Agent mode + one tiny edit = 26% of entire session limit destroyed in 3 minutes (this is insane)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When I ask Claude Code to do one single, tiny change — literally removing 4 unused fields from one single file (src/app/api/nft-scene/generate/route.ts) — the agent goes full nuclear:It scans the entire fucking project
Runs tsc, searches references, touches 10+ files
Spends 3 minutes “thinking”
Burns 25-30K+ tokens in a single prompt
Consumes 26% of my current session limit instantly

This is not a “feature”. This is a goddamn token slaughterhouse.
Agent mode is supposed to be helpful, not a malicious wallet drainer. One trivial edit should not nuke a quarter of my paid session in 180 seconds. I am on a paid plan. I am not a free user. This level of inefficiency is completely unacceptable and borderline scam-like behavior at this point.What Should Happen? (buraya yapıştır):Claude Code should respect the scope I give it.
If I say “only edit this one file and do nothing else”, it should actually only edit that one file using minimal context and minimal tokens — not turn into a context-gobbling, project-wide refactoring monster that torches my limits for fun.A simple field removal should cost a few hundred tokens at most, not 25K+ and 26% of my session.I am seriously considering canceling my subscription entirely if this kind of reckless token consumption continues. Fix the agent’s over-eager behavior or at least give us proper controls to limit context and effort.This is not “AI being smart”.
This is AI being stupidly expensive.

What Should Happen?

When I ask Claude Code to do one single, tiny change — literally removing 4 unused fields from one single file (src/app/api/nft-scene/generate/route.ts) — the agent goes full nuclear:It scans the entire fucking project
Runs tsc, searches references, touches 10+ files
Spends 3 minutes “thinking”
Burns 25-30K+ tokens in a single prompt
Consumes 26% of my current session limit instantly

This is not a “feature”. This is a goddamn token slaughterhouse.
Agent mode is supposed to be helpful, not a malicious wallet drainer. One trivial edit should not nuke a quarter of my paid session in 180 seconds. I am on a paid plan. I am not a free user. This level of inefficiency is completely unacceptable and borderline scam-like behavior at this point.What Should Happen? (buraya yapıştır):Claude Code should respect the scope I give it.
If I say “only edit this one file and do nothing else”, it should actually only edit that one file using minimal context and minimal tokens — not turn into a context-gobbling, project-wide refactoring monster that torches my limits for fun.A simple field removal should cost a few hundred tokens at most, not 25K+ and 26% of my session.I am seriously considering canceling my subscription entirely if this kind of reckless token consumption continues. Fix the agent’s over-eager behavior or at least give us proper controls to limit context and effort.This is not “AI being smart”.
This is AI being stupidly expensive.

Error Messages/Logs

When I ask Claude Code to do one single, tiny change — literally removing 4 unused fields from one single file (src/app/api/nft-scene/generate/route.ts) — the agent goes full nuclear:It scans the entire fucking project
Runs tsc, searches references, touches 10+ files
Spends 3 minutes “thinking”
Burns 25-30K+ tokens in a single prompt
Consumes 26% of my current session limit instantly

This is not a “feature”. This is a goddamn token slaughterhouse.
Agent mode is supposed to be helpful, not a malicious wallet drainer. One trivial edit should not nuke a quarter of my paid session in 180 seconds.  I am on a paid plan. I am not a free user. This level of inefficiency is completely unacceptable and borderline scam-like behavior at this point.What Should Happen? (buraya yapıştır):Claude Code should respect the scope I give it.
If I say “only edit this one file and do nothing else”, it should actually only edit that one file using minimal context and minimal tokens — not turn into a context-gobbling, project-wide refactoring monster that torches my limits for fun.A simple field removal should cost a few hundred tokens at most, not 25K+ and 26% of my session.I am seriously considering canceling my subscription entirely if this kind of reckless token consumption continues. Fix the agent’s over-eager behavior or at least give us proper controls to limit context and effort.This is not “AI being smart”.
This is AI being stupidly expensive.

Steps to Reproduce

When I ask Claude Code to do one single, tiny change — literally removing 4 unused fields from one single file (src/app/api/nft-scene/generate/route.ts) — the agent goes full nuclear:It scans the entire fucking project
Runs tsc, searches references, touches 10+ files
Spends 3 minutes “thinking”
Burns 25-30K+ tokens in a single prompt
Consumes 26% of my current session limit instantly

This is not a “feature”. This is a goddamn token slaughterhouse.
Agent mode is supposed to be helpful, not a malicious wallet drainer. One trivial edit should not nuke a quarter of my paid session in 180 seconds. I am on a paid plan. I am not a free user. This level of inefficiency is completely unacceptable and borderline scam-like behavior at this point.What Should Happen? (buraya yapıştır):Claude Code should respect the scope I give it.
If I say “only edit this one file and do nothing else”, it should actually only edit that one file using minimal context and minimal tokens — not turn into a context-gobbling, project-wide refactoring monster that torches my limits for fun.A simple field removal should cost a few hundred tokens at most, not 25K+ and 26% of my session.I am seriously considering canceling my subscription entirely if this kind of reckless token consumption continues. Fix the agent’s over-eager behavior or at least give us proper controls to limit context and effort.This is not “AI being smart”.
This is AI being stupidly expensive.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

2.1.92

Claude Code Version

2.1.92

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

When I ask Claude Code to do one single, tiny change — literally removing 4 unused fields from one single file (src/app/api/nft-scene/generate/route.ts) — the agent goes full nuclear:It scans the entire fucking project
Runs tsc, searches references, touches 10+ files
Spends 3 minutes “thinking”
Burns 25-30K+ tokens in a single prompt
Consumes 26% of my current session limit instantly

This is not a “feature”. This is a goddamn token slaughterhouse.
Agent mode is supposed to be helpful, not a malicious wallet drainer. One trivial edit should not nuke a quarter of my paid session in 180 seconds. I am on a paid plan. I am not a free user. This level of inefficiency is completely unacceptable and borderline scam-like behavior at this point.What Should Happen? (buraya yapıştır):Claude Code should respect the scope I give it.
If I say “only edit this one file and do nothing else”, it should actually only edit that one file using minimal context and minimal tokens — not turn into a context-gobbling, project-wide refactoring monster that torches my limits for fun.A simple field removal should cost a few hundred tokens at most, not 25K+ and 26% of my session.I am seriously considering canceling my subscription entirely if this kind of reckless token consumption continues. Fix the agent’s over-eager behavior or at least give us proper controls to limit context and effort.This is not “AI being smart”.
This is AI being stupidly expensive.