Liab Deploy

Skill: liab-deploy

Give the project a data-sovereign distribution channel: self-hosted infrastructure that serves the DataLad dataset over git-annex remotes, so the lab controls where its data lives while keeping the same Distributability a cloud sibling provides. The infra is declarative (pyinfra), so the deployment itself is reproducible. You delegate sibling registration and the save to the datalad doer.

Load plugins/disseminate/references/liab-deployments.md for the deployment layout and mapping before generating.

When to use

• The project wants to self-host its dataset (data sovereignty, institutional policy), alongside or

instead of a cloud archive.

• Do NOT use to push to an existing sibling (disseminate/publish) or to mint a DOI

(dataset-release) — this scaffolds the destination infrastructure.

Steps

1. Gather targets — the host(s) that will run the deployment and whether Forgejo (git host) +

git-annex serving are both wanted. Confirm the operator has access to the target hosts.

1. Scaffold the deployment (per the reference) at liab/: inventory.py (hosts), deploy.py

(pyinfra operations for Forgejo + git-annex special remote), and config/. Do not run the deployment for the user — scaffold it and show the pyinfra inventory.py deploy.py command.

1. Register the sibling (datalad doer) — once the store is up, delegate:

> "siblings: register the self-hosted Forgejo/git-annex store as a sibling (create-sibling / > the annex special remote), with a storage --publish-depends so annexed content is served." Then the user pushes with disseminate/publish.

1. Register + log — record the deployment path (and, once live, the sibling name) under a product

(kind other) outputs[]; append { ts, op: liab-deploy, stage: disseminate, note: "Lab-in-a-Box deploy scaffold; sibling ", branch: }.

1. Save — delegate to the datalad doer: "save: datalad save -m 'liab-deploy: scaffold self-hosted serving'."
2. Report — the deployment path, how to run it, the registered sibling (once live), and the next

step: disseminate/publish to push to the self-hosted store, then link-outputs to relate the mirror to the dataset (IsVariantFormOf).

Constraints

• Scaffold, do not deploy: never run pyinfra or provision remote hosts on the user's behalf —

present the command and let the operator run it against their own infrastructure.

• The self-hosted store is a destination — it reuses the standard sibling/publish/provenance flow

(register via the datalad doer, push via publish), it does not bypass it.

• Ensure annexed content is actually served (a storage --publish-depends) so a clone can

datalad get from the self-hosted store — otherwise it distributes history without data.

• Record under a product's outputs[]; keep log: append-only and the ledger schema-valid.

Delegate sibling/save operations to the datalad doer.

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.