Preflight Checklist

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

What's Wrong?

Claude Code repeatedly ignored my explicit instructions to work on disk D: (D:/projects/m@kerlab/) and instead kept opening, reading, building, and running files on C: (C:/Users/sound/boxcraft/) — a
dead/abandoned copy of the same project.

This happened across multiple sessions over ~10 days despite:

• CLAUDE.md explicitly stating "Disco di lavoro: D: — tutti i progetti vanno su D:, mai creare progetti su C:"
• Memory files saved with the same instruction
• The project having a full GSD workflow (.planning/) on D:/projects/m@kerlab/
• Me verbally correcting this multiple times

What happened

• Claude Code launched builds, dev servers, wrote .env.local files, ran Docker checks — all on C:/Users/sound/boxcraft/ instead of D:/projects/m@kerlab/
• Even after being corrected, it continued reading C: files (e.g. trying to check disk usage of C:/Users/sound/boxcraft/)
• It also guessed port numbers (8500) instead of reading the actual package.json config (port 7500)
• Hundreds of thousands of tokens were wasted on operations targeting the wrong directory

Impact

• Significant token waste across multiple sessions (Max plan, Opus model with 1M context)
• Hours of user time lost correcting and re-explaining
• Operations performed on wrong codebase (wrong .env.local written, wrong dev server started)

Expected behavior

Claude Code should:

1. Respect CLAUDE.md instructions about working directory
2. Check GSD .planning/ state to find the active project
3. NEVER take initiative on paths/ports/configs — ask the user when unsure
4. Once corrected, NEVER repeat the same mistake

Environment

• Windows 11
• Claude Code CLI (Max plan, Opus 4.6 1M context)
• Project: M@ker Lab (Next.js 16 + React 19 + Supabase)

Request

I am requesting a token/billing credit for the wasted resources caused by this persistent bug in instruction following. The AI repeatedly ignored both static config (CLAUDE.md) and dynamic corrections from
the user.

What Should Happen?

WORKING IN GSD!!!!!! BUT HE WORKS OUT OF IT!

Error Messages/Logs

Steps to Reproduce

Claude Code repeatedly ignored my explicit instructions to work on disk D: (D:/projects/m@kerlab/) and instead kept opening, reading, building, and running files on C: (C:/Users/sound/boxcraft/) — a
dead/abandoned copy of the same project.

This happened across multiple sessions over ~10 days despite:

• CLAUDE.md explicitly stating "Disco di lavoro: D: — tutti i progetti vanno su D:, mai creare progetti su C:"
• Memory files saved with the same instruction
• The project having a full GSD workflow (.planning/) on D:/projects/m@kerlab/
• Me verbally correcting this multiple times

What happened

• Claude Code launched builds, dev servers, wrote .env.local files, ran Docker checks — all on C:/Users/sound/boxcraft/ instead of D:/projects/m@kerlab/
• Even after being corrected, it continued reading C: files (e.g. trying to check disk usage of C:/Users/sound/boxcraft/)
• It also guessed port numbers (8500) instead of reading the actual package.json config (port 7500)
• Hundreds of thousands of tokens were wasted on operations targeting the wrong directory

Impact

• Significant token waste across multiple sessions (Max plan, Opus model with 1M context)
• Hours of user time lost correcting and re-explaining
• Operations performed on wrong codebase (wrong .env.local written, wrong dev server started)

Expected behavior

Claude Code should:

1. Respect CLAUDE.md instructions about working directory
2. Check GSD .planning/ state to find the active project
3. NEVER take initiative on paths/ports/configs — ask the user when unsure
4. Once corrected, NEVER repeat the same mistake

Environment

• Windows 11
• Claude Code CLI (Max plan, Opus 4.6 1M context)
• Project: M@ker Lab (Next.js 16 + React 19 + Supabase)

Request

I am requesting a token/billing credit for the wasted resources caused by this persistent bug in instruction following. The AI repeatedly ignored both static config (CLAUDE.md) and dynamic corrections from
the user.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

No response

Claude Code Version

OPUS 4.6 CLAUDE CODE MAX 1M

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

LAST CHANCE