several limitations that make it unsuitable for real-time development scenarios:

## Problem
1. Code Size Handling

Fails to handle code files larger than ~400–500 lines.

This is impractical for production where thousands of lines must be processed.

2. Stability Issues During Fixes

While addressing a specific bug, it often introduces regressions in surrounding code.

Causes frequent failures even with minor bug fixes.

3. History & Context Loss

Code analysis quickly consumes the session history.

When a fix is attempted, the history is cleared, leading to incomplete or repeated cycles of analysis without resolution.

4. Lack of Control (Auto Apply)

Auto-apply changes break core business logic without user validation.

A “stop auto apply” option is essential to allow review before committing changes.

5. File & Dependency Awareness

Current approach focuses only on the immediate code segment instead of the full file.

This narrow focus increases error cascades and rework.

Agent credibility is reduced, as fixing its own introduced errors restarts the cycle and clears history.

Summary:
Amazon Q shows potential but is currently failing to meet enterprise requirements for large-scale, stable development. Key improvements needed are: better large-file handling, preserving session history, dependency-aware analysis, and manual control over applying fixes.

## Expected behavior
should not break the code while working on issue fixing.
code analysis should be more robust.
should increase the memory limits.

## System details (run `AWS: About` and/or `Amazon Q: About`)

- OS: Windows
- Visual Studio Code version: 1.103.2
- AWS Toolkit version:
- Amazon Q version: 1.91.0

The report provides no source files, tests, or entry points; begin by reproducing the large-file, regression, history, auto-apply, and dependency-awareness behaviors in Windows with VS Code 1.103.2 and Amazon Q 1.91.0. Done would require the requested robustness, preserved context, dependency-aware analysis, and user-controlled application of fixes, but the scope needs to be split into actionable issues first.

Written by the indexing model from the issue text.