Claude Code deleted large model files unnecessarily during file permission troubleshooting

Documentation Type

Other

Documentation Location

_No response_

Section/Topic

File and directory operations involving large model files / data preservation before deletion

Current Documentation

No specific documentation exists for this scenario. Claude proceeded to instruct deletion
of a directory containing large downloaded model files (~5GB) without first moving or
preserving the checkpoint files, when the actual fix only required re-cloning the
repository itself.

What's Wrong or Missing?

While helping set up a ComfyUI custom node (ComfyUI-LatentSyncWrapper), Claude identified
that the node was cloned via WSL into /mnt/f/ and Windows processes couldn't read it due to
a permission error (\\wsl.localhost\... path access denied).

The correct fix was:

1. Move the checkpoints folder (containing ~5GB of downloaded model files) to a safe

location

1. Delete and re-clone the repository from the Windows side
2. Move the checkpoints folder back

Instead, Claude instructed the user to delete the entire folder first, then re-clone, then
re-download all model files — resulting in unnecessary re-downloading of ~5GB of data.

Impact: User had to re-download ~5GB of model files that could have been preserved with a
simple move operation. If the models had been larger (e.g. 20GB+), this mistake would have
wasted significantly more time and bandwidth.

Expected behavior: Before instructing users to delete any directory containing large files,
Claude should check for and preserve valuable data (model weights, checkpoints, etc.) by
moving them first.

Suggested Improvement

● Before instructing users to delete any directory, Claude should always check if it contains
large or hard-to-recover files (model weights, checkpoints, datasets, etc.). If it does,
Claude must instruct the user to move those files to a safe location first, then proceed
with the delete/re-clone operation, and finally move the files back. This is especially
critical in AI/ML workflows where model files can be 5GB to 100GB+, and re-downloading them
wastes significant time and bandwidth.

Suggested rule: "Never delete a directory that contains large binary files without first
explicitly moving those files to safety. Always ask yourself: can this data be recovered
easily? If not, preserve it first."

Impact

High - Prevents users from using a feature

Additional Context

● The user is requesting compensation for the wasted time and bandwidth caused by this
mistake. The unnecessary re-download of ~5GB of model files took approximately 30-40
minutes of the user's time. If this had occurred with larger models (20GB+), the impact
would have been even more significant. The user requests that Anthropic consider crediting
additional Claude Code usage quota as compensation for the disruption caused by this AI
error.