Why curation is its own pack¶

catalogue-curation sits one meta-level above the rest of the catalogue. Where governance-extras governs this repo's decisions, catalogue-curation governs the catalogue's growth and reproduction — adding packs, adopting skills, forking the whole thing. A few design choices explain its shape.

It's a catalogue-operator pack, not an adopter pack¶

Most packs earn their place by being a habit a builder reaches for. This one's user is a catalogue operator — you, and anyone who forks the catalogue. That's a narrower audience, which is why the pack is opt-in, repo-scope, and off every default profile. It cleared the charter's bar as a forward bet: the manual cost of curation suppressed the practice, and lowering it (plus the ecosystem-growth angle of attributed forks) should make curation routine.

Single authoritative source, no merge-back¶

Forking follows the model that keeps confidential-and-public mirrors sane: one repo is authoritative, the derivative is read-only-derived. There is no "sync my fork's edits back upstream" flow — the only inbound path is a fresh, reviewed assimilation of a specific unit. This is deliberate: bidirectional sync between a catalogue and its forks is how divergence and leak accidents happen.

Fail-closed, because leaking is the worse error¶

Export scrubbing has two possible mistakes: omit a file that was actually safe, or ship one that leaked identity. The pack always prefers the first. The verify gate hard-fails on any surviving anchor rather than trusting the strip step — you'd rather re-add an over-omitted file than discover an org's name in a public fork after the fact.

Assimilation shapes, it doesn't paste¶

Bringing a skill in isn't copy-and-reformat. External work carries other projects' conventions and anti-patterns; assimilation judges the raw content for safety, then reshapes it to this repo's craft — activation, progressive disclosure, anti-pattern steering — or rejects it. The same craft discipline applies whether you're adopting a skill or authoring a new one from scratch, which is why the pack is framed as the operator's toolkit rather than an import-only tool.

No new engine¶

Every part of the pack is skills plus declarative manifests and one lint — no scheduler, no service, no daemon. That's consistent with the catalogue's other "loops," and it keeps the pack something a maintainer can read and trust rather than a black box.