[BUG] Claude Code doesn't verify its own changes end-to-end, causing repeated runtime bugs

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 implementing a multi-step feature spanning several files, Claude consistently:

1. Modifies components in isolation without checking how they connect to each other
2. Misses fields from data structures — copies only part of the data, ignoring the rest
3. Doesn't update all switch/match expressions that need to handle new cases
4. Doesn't verify infrastructure config exists for newly created components
5. Sets incorrect state values without checking the full state machine

Every single one of these bugs was discoverable by reading the existing code. No external context needed — just follow the data flow from input to output.

The result: I spend hours debugging issues in production that should have been caught before the code was written. What should be done correctly in one pass takes 3-5 iterations over multiple days, each time with me finding the next bug, explaining it,
and waiting for a fix.

Expected behavior: When changing a pipeline or data flow, Claude should trace the entire chain and verify all touch points before considering the work complete. Not just the file being edited — everything connected to it.

What Should Happen?

1. Ask Claude Code to implement a feature that spans multiple files in a data processing pipeline (e.g., parser → transformer → message → serializer → handler)
2. Let Claude complete the implementation
3. Run the code
4. Observe runtime errors: missing fields in data objects, unhandled switch/match cases, missing config declarations, incorrect state values
5. Report each bug to Claude, get a fix, run again
6. Find the next bug that should have been caught in step 2
7. Repeat steps 5-6 multiple times

Error Messages/Logs

Steps to Reproduce

The issue is not about complex or ambiguous requirements. Every bug was discoverable by simply reading the code that already existed in the project. Claude edits one file at a time without verifying how changes affect connected components. It treats
each file as independent, when in reality they form a chain where a mistake in one breaks everything downstream.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.63

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

iTerm2

Additional Information

_No response_