Datalad Fsck

Skill: datalad-fsck

Check the integrity of annexed file content in a DataLad dataset using git annex fsck. Identifies missing, corrupt, or unreachable content and guides recovery.

Steps

1. Verify DataLad context — check for .datalad/ in the current directory or a

parent: ``bash ls .datalad/ 2>/dev/null `` If no dataset is found, inform the user and stop.

1. Determine scope — ask the user:

• Fast check (--fast): verifies file sizes and presence only. Quick, recommended

for routine checks or large datasets. Does not verify checksums.

• Full check (no flag): verifies checksums against stored hashes. Thorough but can

take a very long time on large datasets. Only recommend if corruption is suspected or a transfer completed unexpectedly.

• Path-scoped check: optionally limit to a specific file or directory by passing

the path as an argument.

Warn the user: a full fsck on a large dataset can take a very long time. Recommend --fast first unless there is specific reason to suspect checksum-level corruption.

1. Construct and show command — display before executing:

``bash git annex fsck [--fast] [] `` Ask: "Ready to run?"

1. Execute and interpret output — run the command and summarize results:

• OK: file is present and (for full check) checksum matches — no action needed
• corrupt: checksum mismatch — file content is damaged
• missing: file is not present locally

Report counts of OK / corrupt / missing files.

1. Suggest recovery based on findings:

• Missing content: run datalad get to re-fetch from a known remote
• Corrupt content: drop the corrupt copy with datalad drop --nocheck ,

then re-fetch with datalad get

• Unrecoverable (no remote has the content): inform the user the content is lost

and cannot be recovered automatically — escalate to manual investigation

Constraints

• Always show the full git annex fsck command before executing.
• Always warn that a full fsck (without --fast) can take a very long time on large

datasets before starting.

• Never attempt to auto-repair corrupt files without confirming recovery steps with the

user first.

• Load ${CLAUDE_PLUGIN_ROOT}/../references/troubleshooting.md when the user needs

guidance on recovering from corrupt or missing content beyond simple datalad get.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: neuroanalytics
• Source: neuroanalytics/data-science-harness
• License: MIT

Install and usage instructions live in the source repository linked above.