Model Local Data Agent Skill

Model local data → WordPress-driven data

Some hand-authored sites render their main content from JavaScript data: an empty container like <div id="newestGrid"></div> is filled at runtime by a mount call (mountGrid('#newestGrid', newestObjets(4))) that maps over an in-file array (const OBJETS = [ … ]). Others render repeated content cards directly in static HTML, such as a blog/archive index of post-preview cards. On a straight carry these grids are empty or non-native in the block editor — the data lives in JS or fixed HTML, not WordPress records.

This skill reads the source and produces a data-model.json describing how to make the data WordPress-driven: records + taxonomy + per-item meta, native core/query loops in place of the mounts/cards, and a faithful card render — while the source's styling, animation, filtering, and detail-modal JS keep running (reading WordPress data from per-card DOM islands).

The deterministic converter (liberate_convert_local_site) consumes the file; this skill only authors the model by reading the source.

When it applies

Run this BEFORE liberate_convert_local_site when the source has any of:

• empty mount containers (<div id="…"> / <ul id="…"> with no children) filled by JS,
• an in-file data array of records (each with a stable id),
• a list/grid/catalog/gallery rendered by mapping that array to card markup,
• optionally a detail modal that looks an item up by id (ARR.find(x => x.id === id)).

It also applies when the source renders repeated content cards directly in static HTML, such as:

• a blog/archive index made of post-preview cards,
• repeated card links to article/detail pages,
• static category/date/excerpt/image card metadata that should become native WordPress posts and terms.

If the scaffold reports discovered.source === 'none', no JS data array or repeated static-HTML content-card grid was found — skip straight to convert.

How to produce it (scaffold-first)

• Run the deterministic scaffold. Call liberate_data_model_scaffold({ dir: "<source-dir>", outputDir: "<output-dir>" }). It writes <output-dir>/data-model.draft.json (a partial DataModel) and returns { model, skillTodos, discovered, validation }. The scaffold reports discovered.source as js-array, html-cards, or none. It has already extracted every record verbatim (items[]), enumerated taxonomy.terms and fields[], linked the mounts[], and set sourceArrays when JS arrays were found. The discovered summary lists which arrays were found/rejected and records dropped low-confidence/orphan containers under discovered.unmatchedContainers — if your data array or content-card grid is missing there, see the manual fallback.

For html-cards, the scaffold derives card.template deterministically from the source card markup, so there is no card.template todo to resolve. Confirm only the flagged role/order todos. Bodies come from following each card's link; when multiple cards share a target, the linked page is the single template and the card bodies stay as excerpts. Static HTML cards register as core post records with native category terms, not a custom CPT, and convert neutralizes the card container into a native query loop.

If discovered.source === 'none', stop here: do not resolve todos or write data-model.json; fall through to the pure carry convert path.

• Resolve ONLY the skillTodos. Each has a path, instruction, and source evidence. Do not re-author filled slots. Typical todos:
• card.template (JS array path only) — rewrite the per-item card function (in evidence) into a single-root skeleton with data-dla-* bindings preserving the source classes. Grammar: data-dla-text/attr/class/if; <expr> = 'lit' | id | title | content | cat.slug | cat.label | meta.<key> | gallery.<n>.caption | map.<name>.<expr>; <cond> = <expr> | <expr>=='lit' | <expr>!='lit'. Add value-keyed lookups (e.g. CAT_TONE) to card.maps.
• mounts[i].query.order — confirm/adjust ordering + perPage (scaffold defaults to date/DESC). A full/all/complete catalog grid usually renders in source array order, so set order: 'ASC'; use DESC only for newest/recent/latest grids.
• items[].id / items[].title / taxonomy — low-confidence role guesses; confirm or correct. If a value belongs in meta, move it (never drop it).
• discovered.unmatchedContainers — check these low-confidence/orphan containers for any dropped grid before converting; add missing mounts only when the source proves they are content-driven.

• Validate and write. Re-check that every map.<name>/meta.<key> the template references exists, item count == source record count, terms == the source's category set. Write the resolved model to <output-dir>/data-model.json. liberate_convert_local_site auto-activates the data path when that file is present.

Manual fallback (if the scaffold tool is unavailable)

If liberate_data_model_scaffold is not registered (e.g. the MCP server hasn't restarted to pick it up) or errors, author data-model.json by hand: read the source JS or static card HTML, copy every record into items[] (id/title/content → as appropriate, category → terms, other fields → meta, image lists → gallery), enumerate taxonomy.terms, list fields[], map each empty mount or repeated card container to a query, author card.template with the data-dla-* grammar above, and set sourceArrays when JS arrays were used. Never drop or summarize a record or field.

Faithfulness rules

• Never drop or summarize a record or field. The card render runs server-side (frontend AND the editor query-loop preview), so the binding must reproduce the source card markup exactly. Surface honest gaps rather than inventing data.