datalad-status

name	datalad-status
description	Auto-invoke inside a DataLad dataset (.datalad/ present) when the user asks what has changed, what is modified, what files are untracked, or what the current dataset state is. Trigger on "what changed", "check state", "show modified", "what's untracked", "dataset status", or /datalad-status. Read-only — never modifies any files.
argument-hint	["paths..."]
user-invocable	true
disable-model-invocation	false
allowed-tools	Read, Bash, Glob

Skill: datalad-status

Inspect the current state of a DataLad dataset — what files are modified, untracked, added, or deleted — without making any changes.

Steps

1. Verify DataLad context — check for .datalad/ in the current directory or any parent:

   ls .datalad/ 2>/dev/null || git rev-parse --show-toplevel
   

   • Dataset found: continue.
   • No dataset found: inform the user and suggest /datalad-init or plain git status.

2. Run datalad status — optionally scoped to paths from $ARGUMENTS:

   datalad status [paths...]
   

   Present the full output to the user. Additional flags to mention if relevant:

   • --annex — shows verbose annex content state (number of copies, remote presence) for each annexed file; useful for understanding data availability
   • --untracked=no — suppresses untracked files from output (useful in large repos with many untracked directories that would otherwise clutter the output)

3. Interpret the status symbols — explain each symbol present in the output:

   Symbol	Meaning
   modified	Content changed since last save
   untracked	File exists on disk but is not recorded in the dataset
   deleted	File was tracked but is now missing from disk
   added	New file staged but not yet saved
   clean	No changes — working tree matches the last commit

   For annexed files, also explain:

   • content present: file pointer exists and content is available locally
   • content missing: file pointer exists but content was dropped (pointer only)

4. Suggest next actions — based on what was found:

   • Modified or untracked files → suggest /datalad-save
   • Missing annex content → suggest datalad get <path> to retrieve it
   • Output files that should have been produced by a command → suggest /datalad-run
   • Clean working tree → confirm no action needed

Constraints

• Read-only — never run any command that modifies the dataset, stages files, or commits.
• Always explain every status symbol that appears in the output — never show raw output without interpretation.
• If the user asks about a specific path, scope the status command to that path.
• Do not confuse datalad status (dataset state) with git status — they overlap but DataLad status correctly handles annexed files that git would misreport.