catalogue-curation¶

The catalogue-operator's toolkit — the skills a maintainer reaches for to grow and maintain an agent-skill catalogue, one meta-level above governance-extras. It is domain-agnostic: the same toolkit serves this software-delivery catalogue, a creative-writing catalogue, or an investment-research one.

Growing a catalogue means several recurring jobs — standing up a new pack, shaping a new skill to convention, bringing in good work from outside, and producing a redistributable copy for another org or domain. Each was hand-run, un-resumable, and craft-inconsistent. This pack makes them gated, resumable, and shaped to the repo's craft — and it's the home for catalogue operations generally, so new ones land here as they come up (see Roadmap).

The toolkit today¶

Skill	The catalogue job it serves
`propose-catalogue-pack`	Stand up a new pack — justify it's additive and fits the catalogue's charter, scaffold the shell, and emit an RFC (or reject it).
`assimilate-primitive`	Bring one external skill / subagent / hook (or a small bundle) in from a local path or URL — safely, then reshaped to our craft (activation, progressive disclosure, anti-pattern steering), or rejected.
`assimilate-repo`	Survey a whole external repo/catalogue into a reviewable RFC of per-candidate verdicts, resumable across sessions and worktrees via a ledger.
`export-catalogue`	Produce a redistributable derivative (organization rebrand or domain re-purposing) in `white-label` or `attributed` mode, with a fail-closed leak check.

The craft-shaping engine that assimilate-primitive uses to bring an external skill up to convention — activation-optimized description, collision check, progressive disclosure, anti-pattern steering — is the same discipline a maintainer applies when authoring a new skill from scratch; it's written to be reused for both.

Guardrails¶

Never mutates the agentbundle engine or credential-brokers through any skill — a path-gate blocks protected-tree changes absent a deliberate, human-authored RFC.
Ingested code runs the repo's own gates (lints + CodeQL/Snyk) before it lands, and known anti-patterns (a script that triggers a skill/agent, a misused agent, a flooding "skill") are steered to our shape or rejected — never laundered in.
Fail-closed export — a run hard-fails if any upstream identity would leak.
No new engine, no new dependency — skills plus declarative manifests only.

Roadmap¶

The pack is the deliberate home for catalogue operations as they arise — e.g. retiring a primitive, deprecating a pack, or a dedicated greenfield skill-authoring flow. Deferred items live in workspace.toml [backlog].

Learn it¶

Full walkthroughs, references, and the "why" live in the per-pack guide: docs/guides/catalogue-curation/.

Requires core and governance-extras. Repo-scope, opt-in; not in any default profile.