Add Skills tab sourced from probabl-ai/skills

[yanndebray]

Summary

Add a new tab to scikit-learn-central that surfaces the skills hosted in probabl-ai/skills.

Motivation

scikit-learn-central currently centralizes scikit-learn ecosystem resources (packages, notebooks, etc.). The probabl-ai/skills repo hosts reusable skills that would benefit from being discoverable in the same place — giving users a single entry point to browse and use them alongside the rest of the ecosystem content.

Proposed change

• Add a new top-level tab (e.g. "Skills") to the navigation.
• Fetch / list the skills from github.com/probabl-ai/skills (similar to how other tabs source content).
• Display each skill with its name, description, and a link to the source.

Open questions

• Should the skills list be fetched dynamically from the GitHub API or generated at build time?
• What metadata do we want to surface per skill (description only, tags, examples)?
• Any auth / rate-limit considerations if the source repo is private?

[yanndebray]

Implementation plan

Source of truth

• Skills live in probabl-ai/skills under skills/<skill-name>/SKILL.md.
• Each SKILL.md has YAML frontmatter (name, description, version) followed by markdown body. There are currently 14 skills (e.g. sklearn-api, skore-api, skrub-api, build-ml-pipeline, …).
• The local repo already contains one skill at skills/sklearn-expert/SKILL.md — same format, so the parser can be shared.

Data pipeline (build-time, mirrors packages/use-cases)

1. Add scripts/fetch-skills.{ts,mjs} — pulls the skills/ tree from probabl-ai/skills (default branch), parses each SKILL.md frontmatter + first paragraph of body, and writes:
   • data/skills.json — index of skill slugs (allowlist, like data/use-cases.json)
   • data/skills/<slug>.json — one file per skill: { slug, name, description, version, source_url, body_preview, tags? }
2. Wire the script into pixi.toml / package.json (alongside the existing stat/catalog scripts) and document a pixi run fetch-skills command.
3. CI: refresh data/skills*.json on a schedule (same cadence as chore: update package stats commits).

Frontend (Vue)

• src/types/skill.ts — Skill, SkillsIndex types.
• src/composables/useSkills.ts — eager-glob @data/skills/*.json, filter through allowlist in @data/skills.json, sort alphabetically. Mirror useUseCases.ts.
• src/components/SkillCard.vue — card with name, description, version chip, "Source" link to the GitHub path, copy-install snippet (e.g. claude plugin install probabl-ai/skills/<slug>).
• src/views/SkillsView.vue — search input + grid of cards. No filter dropdowns in v1; add tag filters in a follow-up if skills gain tags.
• src/router/index.ts — register /skills route, tabKey: 'skills'.
• src/components/ViewTabs.vue — append { key: 'skills', to: '/skills', icon: 'fa-wand-magic-sparkles', label: 'Skills' }. Include in the count badge logic.
• src/App.vue — add skills: skills.value.length || null to tabCounts; for the header submit button, point Skills tab to https://github.com/probabl-ai/skills issues (no in-app submit modal in v1).

Open questions to resolve before coding

• Fetch vs. vendor: do we bundle the JSON in this repo (build-time fetch, current proposal) or hit GitHub at runtime? Build-time keeps the site working offline and avoids rate limits — recommended.
• Detail view: clicking a card → modal with rendered markdown body, or just deep-link to GitHub? Start with deep-link, add a modal if engagement warrants.
• Local skills/sklearn-expert/: keep as-is, include in the indexed list, or migrate it to probabl-ai/skills? Migrating gives a single source of truth.
• Tag/category taxonomy: skills currently have no tags. Decide whether to add them upstream (in each SKILL.md frontmatter) before designing filter UI.

Scope of v1 PR

1. fetch script + first generated data/skills*.json snapshot
2. types + composable + view + card + route + tab entry
3. README note on refreshing skills data

Filters, detail modal, and submit flow land in follow-ups.

[ylechelle]

ping @glemaitre

[glemaitre]

Yep. I want first to have the repo of the skills to have the install fixed and test and then see how to better leverage the external repo (and think a bit how it could scale to add other third-party skill as we do for packages).