Manage Product

Skill: manage-product

Turn kept comparisons into products. A comparison is a cmp/ branch; a product is a named deliverable that groups a chosen subset of them (plus their outputs) into something you will release and cite. This is the first place the harness represents multiple distinct outputs from one project — the ledger's products[] is the registry the later disseminate/* skills release and cross-link. You own the grouping judgment; you delegate the ledger save to the datalad doer.

> Ledger note: products[] is an upsert registry (edited in place), unlike the append-only > log:. Follow docs/project-ledger.md: find a product by id and update its fields, or add a > new entry — never drop or rewrite another product, and move status only forward. Every write is > datalad save-d so the grouping is provenanced.

When to use

• The user wants to bundle comparisons into a paper / dataset release / report, add a comparison to

an existing product, or see/adjust what products exist.

• Do NOT use to create or run a comparison (analyze/propose-comparison / run-comparison), to

release a product with a DOI (disseminate/dataset-release), or to cross-link products (disseminate/link-outputs) — this skill only defines the grouping.

Steps

1. Identify the product — get from the user:

• id — a stable kebab-case slug, unique within products[] (e.g. main-paper)
• kind — paper | dataset | report | article | agent-bundle | other
• title — a human label
• which comparisons belong to it (cmp/ branches). One comparison may appear in more than

one product (e.g. a shared figure) — that is allowed; call it out so it is intentional.

1. Verify the comparisons exist and are complete — delegate to the datalad doer:

> "list branches matching cmp/* and the recent log; confirm each of ` > exists and has a run-comparison entry." If a named comparison has no recorded run, warn that the product references incomplete work and ask whether to include it anyway or run it first (analyze/run-comparison`).

1. Upsert the product in project.yaml products[] (per docs/project-ledger.md): create the

entry if id is new, else update it. Set kind, title, status (planned → in-progress → released), comparisons (the branch list), and outputs (the dataset-relative result paths, e.g. derivatives/cmp-/). Leave dois: [] and relations: [] — those are owned by disseminate/dataset-release and link-outputs.

1. Log it — append one entry to project.yaml:

{ ts, op: manage-product, stage: analyze, note: "grouped into product ()", branch: }.

1. Save — delegate to the datalad doer:

> "save: datalad save -m 'manage-product: groups '."

1. Report — the product id, its comparisons/outputs/status, any incomplete comparisons

flagged, and the next step (add more comparisons, or disseminate/dataset-release to release it, then disseminate/link-outputs to link it to other products).

Constraints

• Keep products[] ids unique; do not duplicate a comparison within a single product's list.
• Products group and describe comparisons — they never move, rename, or modify the branches or

their data. This skill edits only the ledger.

• Do not set dois or relations here; releasing and cross-linking are separate skills.
• The log: stays append-only; products[] is upserted in place but never destructively (status

moves forward, entries are not deleted). Keep the ledger schema-valid (schemas/project.schema.json).

• Delegate every DataLad operation (branch listing, save) 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.