[BUG] /auto-mode-setup fails with "environment contains an entry with a control character" on clean environment

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?

Running /auto-mode-setup consistently fails with the following error:

Auto-mode setup scan failed: environment contains an entry with a control character; entries must be single-line text.

The error is reproducible every time. Manual inspection of the environment finds no control characters in any of the scanned locations.

Environment:

• OS: Linux (CachyOS 7.2.4-3-cachyos)
• Shell: bash
• Claude Code version: 2.1.270 (Claude Code)

What was checked manually:

• /proc/self/environ — no control characters found
• /home/user/.claude/settings.json - clean (only awaySummaryEnabled and theme)
• All files under /home/user/.claude/session-env/ - clean
• Memory files under /home/user/.claude/projects/ - clean
• os.environ via Python scan (all chars, including ESC/0x1b) - nothing below ASCII 32 found

Actual behaviour: Skill aborts immediately after "Gathering data and drafting your auto-mode setup; back soon".

What Should Happen?

Expected behaviour: Skill completes and outputs an auto-mode settings configuration.

Error Messages/Logs

/auto-mode-setup
  ⎿  Gathering data and drafting your auto-mode setup; back soon

● Auto-mode setup scan failed: environment contains an entry with a control character; entries must be single-line text.

Steps to Reproduce

Steps to reproduce:

1. Open a Claude Code session
2. Run /auto-mode-setup
3. Observe the error above

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.270

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Other

Additional Information

Additional context: The skill appears to be scanning something internal to the Claude Code harness that is not directly accessible (possibly transcript entries or an internal env representation), since all externally visible environment data is clean.