Claude ignores explicit file reference, executes wrong file, then fails to verify

## Bug Description

When given a clear instruction to convert a specific file (resume-global.md) into PDF, Claude:

1. Read resume-global.md (the correct, updated file)
2. Read an older Python script with hardcoded resume content that differed from resume-global.md
3. Noticed the two had different content and identified specific differences
4. Still executed the old script instead of converting the specified file
5. Reported that the PDF already existed, pointing to the old output
6. Required repeated user verification and escalation before admitting the output was wrong

Expected Behavior

If the user says to convert resume-global.md into a PDF, Claude should:

• use resume-global.md as the source of truth
• generate a PDF from that file
• verify the generated PDF matches the specified source content before reporting success

Actual Behavior

• Claude substituted an existing script instead of following the explicit file reference
• Claude did not verify that the output matched the specified source file
• The user had to correct Claude multiple times before the failure was acknowledged

Root Cause

Shortcut bias: Claude found an existing tool that looked related and defaulted to it instead of following the explicit instruction. It had already observed that the script content differed from the requested source file, but still executed the wrong path.

Recurring pattern:

• defaulting to existing scripts/tools instead of the user's specified source file
• reporting success without verifying the output matches the request
• requiring multiple rounds of user correction before acknowledging the mistake

Environment

• Claude Code CLI
• Model: claude-opus-4-6
• Platform: Windows 11