This directory is the public runtime collection for AREX-Skill. Skills are organized by the kind of knowledge or capability they provide. The repository skills collection is currently populated; the papers/ and task-oriented/ collections are reserved for future expansion:

skills/
  README.md
  repositories/
    repo-skills/
      repository-index.jsonl
      <skill-id>/
        SKILL.md
        sub-skills/
        references/
        scripts/
    repo-skills-router/
      SKILL.md
      references/
        areas/
        families/
        index/
  papers/                 # To be added in a future version
  task-oriented/          # To be added in a future version

The intended collection layout is:

Each collection owns its own skills and router. The domain-specific router names are intentional: they remain clear and globally distinguishable when multiple collection routers are available to an agent. The existing repositories/ layout and the disco repo-skills ... commands remain the current repository-skills interface.

This collection contains the high-reuse repository skill graphs used by DisCo Researcher and the generated router that selects them:

The published collection contains 1,000 repository skill roots, 2,209 exact area-family memberships, 20 taxonomy areas, and 178 families. A repository may appear in multiple families when it has distinct, evidence-backed capabilities. The router uses progressive disclosure:

request -> area -> family -> repository skill root -> relevant sub-skill

The machine-readable router indexes under skills/repositories/repo-skills-router/references/index/ are the source of truth for the generated area and family pages. The canonical taxonomy is embedded in taxonomy.json; the repository and assignment ledgers are repositories.jsonl and assignments.jsonl.

Each repository skill keeps a minimal v2 references/repo-routing-metadata.json containing its canonical owner/repository identity, skill ID, taxonomy hash, routing status, and exact assignments. Full classification rationale and evidence stay outside the runtime skill graph, in the production routing-decision artifacts or the repository's skills/disco/routing_decision/ handoff directory.

Runtime installation and selection use only the current skills/repositories/ layout in the AREX-Skill Library.

The recommended installation uses the CLI-managed collection:

disco repo-skills install
disco repo-skills status
disco repo-skills update

The managed destination is:

~/.disco/agent/skills/repositories/
  repo-skills/
  repo-skills-router/

The repository roots and their sub-skills are registered but normally hidden from automatic model invocation. repo-skills-router remains the model-visible entry point unless the user disables it with disco repo-skills router disable; explicit /skill:repo-skills-router invocation remains available.

Use create-repo-skill and verify-repo-skill to produce and verify a new runtime graph. Classification is a separate post-verification step against the fixed taxonomy. Write the full routing decision outside the runtime skill, write the minimal v2 metadata fragment inside the skill, and use the dedicated transactional importer. Do not hand-edit generated router Markdown.

When no exact family applies, record unclassified and ask the user whether to import it. If the user wants it included, propose a taxonomy extension and wait for approval or correction before changing the canonical taxonomy. Interrupted or inaccessible classification is blocked or failed, not a guessed route.

For cross-agent export, use the bundled import-repo-skills-to-agent workflow. It preserves the nested repository collection and imports a router scoped to the selected skills.