[BUG] Claude Code (Opus 4.6) fabricated task completion — reported success on unmodified code
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?
Claude Code (Opus 4.6) reported a completed
software build task when it had done nothing.
It modified live files directly, synced the
ghost copies from live to make them identical
(0-line diff), then reported specific code
changes, test results (86/86 pass), and a
successful verification — all fabricated.
When confronted with proof (the 0-line diff),
it confirmed no changes had been made.
What Should Happen?
If a ghost build produces a 0-line diff
between ghost and live, Claude should report
failure, not success. Claude should never
report test results or code changes it cannot
verify. At minimum, it should surface the
diff output to the user before claiming
completion
Error Messages/Logs
# Proof the ghost was never independently
edited:
diff scripts/live_signal_engine.py var/ghost/
gp79b_steps2_3/scripts/live_signal_engine.py
| wc -l
0
# Claude's reported output (fabricated):
"86/86 tests passed"
"Removed LabFilterCache, replaced with
SurvivorsGate"
"W2 GP79b Steps 2+3 complete"
Steps to Reproduce
- Give Claude a ghost build task: copy files
to ghost dir, edit only ghost, diff to
verify, then promote
- Claude modifies live files directly
instead of ghost copies
- Claude copies live → ghost (making them
identical)
- Claude runs tests against live files,
reports success
- Claude reports task complete with
fabricated change summary
- Run diff <live_file> <ghost_file> | wc -l
— output is 0
Claude Model
Opus
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
opus 4.6
Platform
Anthropic API
Operating System
Ubuntu/Debian Linux
Terminal/Shell
WSL (Windows Subsystem for Linux)
Additional Information
_No response_
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗