curate-next

name	curate-next
description	Use when asked to curate the next N KG-Registry stub resources. Selects uncatalogued stub pages not already listed in reports/curation_problems.tsv and dispatches single-resource curation in parallel.

curate-next

Curate the next N stub resources in KG-Registry.

Prerequisite

• Run this skill only from a local clone of the KG-Registry repository.
• The workflow depends on scanning local files under resource/, reading reports/curation_problems.tsv, editing checked-out pages, and running local validation commands.
• If the repository is not cloned locally, stop and clone it before attempting batch curation.

When to use

• "Curate the next 3 stub resources"
• "/curate-next 5"
• "Work through a batch of KG-Registry stubs"

Skip when:

• The user names a specific resource. Use the single-resource curation workflow instead.
• The task is to generate new stub pages. That is a separate maintenance flow.
• The task is purely about reviewing reports/curation_problems.tsv without curating pages.

Inputs

• Required positional argument N: number of resources to curate.
• N must be a positive integer between 1 and 5 inclusive.
• If N is missing, non-integer, <= 0, or > 5, ask the user to clarify rather than guessing.

The cap keeps the batch small enough to curate, validate, and summarize cleanly.

Workflow

1. Validate N.

   • Never default to a value the user did not provide.

2. Build the candidate stub list.

   • Find stub pages from resource/ by looking for stub in the domains list.
   • Prefer a command such as:

     rg -l '^\s*-\s*stub\s*$' resource
     

   • Sort the results for deterministic selection.

3. Exclude resources already logged as blocked.

   • Read reports/curation_problems.tsv and collect the resource_id values from the first column.
   • Drop any candidate whose directory or file stem matches an existing blocked resource_id.

4. Select the next N candidates.

   • Use the first N survivors after sorting.
   • If fewer than N candidates remain, do not pad with blocked or non-stub files. Report the shortfall.

5. Dispatch single-resource curation.

   • For each selected resource, invoke the kg-registry-curation workflow.
   • If the environment supports parallel sub-agents safely, dispatch the selected resources in parallel.
   • If parallel execution is used, isolate each worker from the others before editing. Separate git worktrees are preferred for true parallel edits.
   • Keep each worker focused on one resource file and its direct validation.
   • Make it explicit in each dispatched prompt that web search will likely be necessary for stub expansion and provenance resolution.
   • Make it explicit in each dispatched prompt that original_source and secondary_source must use KG-Registry identifiers, minting a new unique KG-Registry ID when an upstream resource is referenced but not yet present in the registry.

6. Validate each curated file.

   • Each resource must pass uv run make validate-file FILE=<target-file> before being reported complete.
   • If prettification changes the file, validate again afterward.
   • Use the kg-registry-validation skill for the post-edit validation workflow when additional checking is needed.

7. Report outcomes.

   • List curated resources and whether each completed successfully.
   • List blocked resources and whether they were logged to reports/curation_problems.tsv.
   • State any shortfall if fewer than N resources were actually curated.

Quick reference

Task	Command
List stub files	`rg -l '^\s*-\sstub\s$' resource
View blocked resource IDs	cut -f1 reports/curation_problems.tsv
Validate one curated file	uv run make validate-file FILE=resource/<id>/<id>.md

Common mistakes

• Treating every incomplete page as a stub. This workflow is driven specifically by domains containing stub.
• Re-curating resources already documented in reports/curation_problems.tsv.
• Padding the batch with arbitrary resources when fewer than N valid stub candidates remain.
• Dispatching multiple curations into one shared working tree without isolation.
• Dispatching workers without telling them to use web search when the stub itself lacks enough evidence.
• Letting workers write provenance with raw labels or URLs instead of KG-Registry IDs.
• Reporting success before each file has passed validation.

Relationship to other skills

• kg-registry-curation performs the actual per-resource edit and validation steps.
• kg-registry-validation performs the focused validation and quality checks for curated files.
• curate-next is the batch selector and dispatcher.