[BUG] Claude Code repeatedly generates invalid tool invocations
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?
I often see Claude Code generate an invalid tool invocation, fix it, and then sometime later make the same mistake. For example, CC has generated the following tool invocation twice this morning:
Bash(flutter analyze --no-pub 2>&1 | grep -E "error|warning" | grep -v "test\\")
It fails with this error message: grep: Trailing backslash
And then CC fixes the error with this:
Bash(flutter analyze --no-pub 2>&1 | grep -E "^error")
This is one obvious example but I've seen others. I'll try to document them here too.
What Should Happen?
Claude should not generate invalid tool invocations. And if it does, it should not does so multiple times.
Error Messages/Logs
Steps to Reproduce
I have no idea how to reproduce this other than having CC make a change to a Flutter project.
Claude Model
Sonnet (default)
Is this a regression?
No, this never worked
Last Working Version
_No response_
Claude Code Version
v2.1.126
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
Windows Terminal
Additional Information
_No response_
This issue has 5 comments on GitHub. Read the full discussion on GitHub ↗