Datalad Configuration

Skill: datalad-configuration

Read and write dataset-scoped, clone-local, or global DataLad/git configuration using datalad configuration. This is the DataLad-aware wrapper around git config that handles DataLad-specific keys and scopes correctly.

Scopes

| Scope | Where stored | Effect | |-------|-------------|--------| | dataset | .datalad/config (committed) | Applies to all clones of this dataset | | local | .git/config (not committed) | Applies only to this clone | | global | ~/.gitconfig | Applies to all repos for this user |

Use dataset scope for settings that should travel with the dataset (e.g., annex backend choice). Use local for clone-specific overrides (e.g., credential paths).

Steps

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

``bash ls .datalad/ 2>/dev/null `` Configuration can still be inspected/set globally even outside a dataset, but warn the user if no dataset is found.

1. Determine action — from $ARGUMENTS or conversation context:

• get: read the current value of a config key
• set: write a new value
• unset: remove a key

1. Execute:

### get ``bash datalad configuration get # or list all keys in a section: datalad configuration get --scope dataset datalad. ``

For common inspection use cases, also accept bare git config: ``bash git config --list --show-origin ``

### set Gather: key (in section.key form), value, and scope. Construct and show: ``bash datalad configuration set --scope = ``

Multiple keys can be set in one call: ``bash datalad configuration set \ --scope dataset \ "annex.backend=SHA256E" \ "datalad.ui.progressbar=none" ``

For dataset scope: the change writes to .datalad/config, which must be committed. Suggest: datalad save -m "configure " after setting.

### unset ``bash datalad configuration unset --scope ``

1. Show command and execute — always display the full command before running.

Common configuration keys

| Key | Scope | Purpose | |-----|-------|---------| | annex.backend | dataset | Hash backend for new files (SHA256E recommended) | | annex.largefiles | dataset | Pattern deciding what goes into annex vs. git | | datalad.ui.progressbar | local/global | Progress output style | | datalad.locations.cache | local/global | Where DataLad caches data | | remote..annex-ignore | local | Tell annex to ignore a remote | | user.name / user.email | local/global | Git author identity |

Constraints

• Always show the full command before executing.
• For --scope dataset changes: always remind the user to datalad save after setting —

.datalad/config is a tracked file and the change is not recorded until committed.

• Never use git config directly for DataLad-specific keys (datalad.*, annex.*) —

use datalad configuration to ensure DataLad's validation and defaults are applied.

• If the user asks to configure annex largefiles patterns, explain the gitattributes

alternative (echo "*.csv annex.largefiles=nothing" >> .gitattributes) as it is more portable and does not require a dataset commit to take effect.

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.