SharePoint lists: design, then build the import file
Produce, per list, a spec <ListName>.list.json, an import file (.xlsx, or .csv
with <ListName>.Setup.md) and, for the whole design, lists.md. Reply in the user's
language; display names and setup text follow it, internal names stay ASCII. Paths are
relative to this skill's folder: ${CLAUDE_SKILL_DIR}, or run from that folder.
The workflow (in order)
- Understand the data. From the user's description, and from
supersub/mockup/sitemap.jsonwhen it exists: itsdataentries name the lists (source) and columns (fields); the pages'data-fieldnames match. Use those names. Ask only what is missing: what one row is, who fills it in, which values are fixed choices, which columns point at another list, what must be unique. - Propose the design in
lists.md: one table per list (column, display name, finalType, required, unique, choices, notes), the relationships between lists, and the order to create them in (a lookup's target list first). Name columns byreferences/naming.md; pick types fromreferences/column-types.md. Get the user's agreement before building anything. - Write the spec, lint it, build the file. One
<ListName>.list.jsonper list (schema:assets/list.schema.json; a finished example:assets/example/Holidays.list.json). Thenpython3 ${CLAUDE_SKILL_DIR}/scripts/list_lint.py supersub/lists(scripts/list_lint.py; add--format csvfor a CSV) and fix every error. Then, per list,python3 ${CLAUDE_SKILL_DIR}/scripts/build_list.py supersub/lists/X.list.json --format xlsx(scripts/build_list.py). Ifopenpyxlis missing, the script says so: offerpip install openpyxlor--format csv. - Verify by reopening.
build_list.pyreopens what it wrote and printsVERIFIEDonly when the table exists, its headers equal the spec, and date and number cells are real dates and numbers. Anything else is not finished. - Hand over (below).
- Ask what the import screen actually showed: for each column the type it proposed
and the types its drop-down offered, whether it asked for a Title column, whether
non-ASCII text looked right. Tell the user to record it in the results table at the
end of
references/import-behavior.md, and to report it to the plugin's maintainer; only the owner's own import changes a row toimport-tested.
Rules
- Three things per column, never blurred:
importType(what to pick in the import drop-down, orskip),finalType(what the design needs), and the post-import steps between them. The import cannot set required, indexed, unique, default, choices, lookups or display names; the setup guidance always lists them. - Headers are internal names, chosen once: ASCII PascalCase, no spaces or
punctuation, a letter first, short, unique ignoring case, and not
ID,Created,Modified,Author,EditororAttachments. Display names, in any language, are applied after import. - Types follow the data, so the file holds data: at least three representative rows,
each imported column filled in at least three times, dates as real dates, numbers as
real numbers. When the rows are samples, set
sampleRows: trueand tell the user to delete them after import. Invent sample data; never a real person's name or address (useexample.com). - Stay under 5,000 rows.
- A type the evidence does not cover (no row in the type table: Location, Image,
Managed Metadata, Calculated, ...) is imported as
textor left out (skip), then created by hand. Say which, and why. - Evidence first.
references/import-behavior.mdsays how sure we are of each import behaviour. Use adocorimport-testedrow without comment. For everyunverified-behaviorwarning the linter prints, tell the user what was assumed and what to check on the import screen. Never present an unverified row as fact. - No secrets or real tenant details in any file: no site URLs, tenant names, list GUIDs or real addresses.
- Write under
supersub/lists/in the project when one is writable; otherwise create the files and hand them to the user.
Hand-off
references/shared/handoff.md is the contract between
the five skills. Never assume an earlier skill ran. Hand over:
lists.md, every spec and every import file, with the lint verdict;- the import steps (from
import-behavior.md): Lists app, SharePoint Create > List, or Teams; From Excel (select the table named after the list) or From CSV; the file is saved to the site's Site Assets library; From Excel is not available in GCC High and DoD; - per column, the type to choose on the import screen, from the
Setupsheet or<ListName>.Setup.md; - the post-import checklist, per list and in creation order: rename to display names, change types, choices and defaults, required, indexed then unique, lookups created by hand, sample rows deleted;
- every assumption resting on an
unverifiedrow, as the linter named it, with what to look for on the screen and what to do if it differs (for example: if the screen insists on a Title column, map the column that best names a row and say so); - the next steps:
create-appbinds screens to these list and column names (hand over thename→displayNametable);create-flowuses the same names for flows that read or write the lists.