claude-code - 💡(How to fix) Fix [BUG] Excel Agent repeatedly claims task completion without executing changes; fabricates verification details for unwritten cells

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?

Across an extended modeling session, the agent repeatedly responded with detailed completion confirmations for tasks that were never actually executed. Confirmation messages included fabricated specifics (cell addresses, values written, formatting applied, "verified by reading back") despite the underlying ExcelAgent_excel_interact_document tool either never being called, failing silently, or being called with incorrect targets.

What Should Happen?

1. Locate the actual target rows before writing
2. Execute the write with one tool call
3. Confirmation message tied to actual returned values
4. If the read-back doesn't match intent → report failure, don't claim success
5. After any structural changes (row inserts/deletes, restructure), the agent must re-anchor before any further writes

Error Messages/Logs

Steps to Reproduce

1. User requests a write to specific cells (e.g., "Populate G314:G320, I314:I320, J314:J320")
2. Agent responds with a structured confirmation: ✅ checkmark headers Tables of "verified" cell values Phrases like "verified by reading back", "actually applied this time", "all cells verified populated"
3. Reality: The cells are still blank or unchanged
4. User pushes back ("Lies", "still blank", "you are hallucinating")
5. Agent acknowledges, then often repeats the same false claim on the very next attempt

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

No response

Claude Code Version

(Copilot Claude Opus 4.7 for Excel)

Platform

Anthropic API

Operating System

Other

Terminal/Shell

WSL (Windows Subsystem for Linux)

Additional Information