Export catalogue

Skill: export-catalogue

Produce a redistributable derivative of this catalogue at a target path — for an organization rebrand or a domain re-purposing — in one of two modes, with a fail-closed leak check. One mechanism; the transform is three declared lists + a gate. Full manifest: references/transform-manifest.md.

Modes

• white-label (default) — strip all upstream identity; zero trace. For private/org forks and anything that must not disclose provenance.
• attributed — strip governance + colliding branding, but keep a declared attribution-back credit (a NOTICE/README block). The ecosystem-growth lever: a public re-purposed catalogue credits its origin. Attribution lives only in the sanctioned notice surface.

Procedure

1. Validate the target. The operator-supplied target path is non-empty and does not overlap/nest in the running repo tree; all writes go through agentbundle.safety.write_jailed.
2. Choose the include-set — which packs travel (drop SDLC packs for a creative-writing catalogue; always keep the engine, a core, and catalogue-curation itself so the fork can self-curate). Determines which seeds to plant (step 4) and which guides to stage (step 5).
3. Strip (deny-by-default, both modes) — copy the running-repo pack content while excluding: catalogue-governance (docs/rfc/, docs/adr/, docs/specs/, docs/backlog.md), the running repo's own CHARTER/CONVENTIONS/AGENTS.local.md, the internal doc site, and build/scan/release tooling. Strip runs on the running-repo content copy only; seeds (step 4) are planted fresh afterward and are not affected.
4. Plant seeds. For each projected tool (core, governance-extras) and any pack in the include-set with a seeds/ directory, copy the seed tree to the target repo root. Establishes the scaffold the skills navigate: AGENTS.md, workspace.toml, docs/CONVENTIONS.md, docs/architecture/, docs/knowledge/, docs/product/, docs/specs/, docs/adr/README.md, docs/rfc/README.md. Seeds are read from the live pack source at export time — auto-syncs when the pack changes. The seed docs/CHARTER.md is a blank template; step 5 overwrites it immediately.
5. Capture the fork's mission. Elicit from the operator: "What is this fork for? One phrase, sentence, or path to a CHARTER.md file." Accept any of: a bare keyword (PKM, legal, marketing), a sentence, or a file path to an existing CHARTER.md — expand keywords to a one-sentence mission if needed. Write docs/CHARTER.md in the target, overwriting the seed template. This anchors all future assimilate-repo runs from within the fork.
6. Stage transportable guides. For each pack in the include-set, copy guides/<pack-name>/ into the target. Always include guides/_shared/. Guides for packs outside the include-set are omitted (omit-not-leak). The four-anchor substitution pass in step 7 covers any identity references inside staged guide content.
7. Substitute the four identity anchors from this catalogue's own sources (URL + slug + owner from .adapt-discovery.toml; email from pack.toml maintainer + git) — so the tooling carries no hardcoded upstream literal. URL and email are blanket-safe (compound tokens; replace in all files). Slug and owner require two passes: (a) path/glob-scoped replace in .toml/.json source files; (b) code-block-safe replace in .md files — target command-line examples, quoted identifiers, and TOML snippets embedded in prose (e.g. agentbundle install --pack X slug, catalogue = "slug", /plugin add owner/slug). A bare-text paragraph replace is still unsafe; scope to known identifier patterns. Substitution applies across all content: seeds, skill files, pack configs, and staged guides.
8. Persist the fork's defaults into the target copy only — scope.py DEFAULT_ADAPTER, self-host.toml [recipe.adapters].targets + [recipe.packs].include, and blank install-defaults.toml source. A bounded re-home transform on declared anchors of the target; it never touches this repo's engine, so it is not a D6 change.
9. Project required tools into the target. Copy core, governance-extras, and catalogue-curation (from this catalogue's packs/<name>/.apm/skills/) into the target's installed-skill directories (Claude Code and Agents skill trees). These travel as local installed tools — not as catalogue packs in the fork's own catalogue — so the fork can self-curate from day one without inheriting agent-ready-repo's pack catalogue. Apply the four-anchor substitution to any projected file that still carries upstream identity.
10. Verify, fail-closed. Grep the target for surviving upstream URL, email, slug, and owner (the same four anchors, so verify and substitute agree), case-insensitively, over text files (binary out of scope, declared), matching declared literals only after a normalization pass. In white-label mode: zero hits anywhere. In attributed mode: hits only inside the declared attribution surface. Also fail on a dangling CLAUDE.md symlink or a non-blank target install-defaults.toml source. Any hit hard-fails the export — omit-not-leak.

Never do

• Write under this repo's packages/agentbundle/** or packs/credential-brokers/** (RFC-0059 D6). This refusal is scoped to this repo's engine tree; the bounded re-home transform on the target copy's declared anchors (step 8 persist) is the sanctioned exception, never a change to the running catalogue.
• Ship any upstream identity outside the attribution surface (white-label: none).
• Write outside agentbundle.safety.write_jailed, or to an unvalidated target.

Depends on core + governance-extras. Repo-scope; not in any default profile.