feat(fields): add verbosity=slim to list-available-fields

[nawartamawi]

Adds an optional verbosity: 'slim' | 'full' param to list-available-fields.

• full (default) — unchanged: human-readable table + full per-field metadata incl. column_ref.
• slim — no table; fields grouped by datasource:

  { count, datasources: [{ datasource, contentUrl?, fields: [{ caption, role, datatype }] }] }
  

  Each field is just { caption, role, datatype }. The datasource name is carried once per group (not repeated per field), and contentUrl is included per published datasource (the input resolve-datasource-luid needs; omitted for embedded ones). A single-datasource workbook is one group, so callers always parse one shape.

Why: the default response returns the field catalog twice (table + full metadata) — ~68KB for a wide datasource (~155 fields), which overflows the inline-output cap and gets truncated. slim stays well under the cap for callers that only need to reason over field names/roles/types. Get the exact column_ref via resolve-field.

Verbosity param rather than a new tool to avoid the tools/list schema budget.

Grouping (addresses @mattcfilbert's review): slim originally hoisted fields[0].datasource and dropped per-field datasource — misattributing every field past the first in multi-datasource workbooks and erasing the disambiguator for same-caption fields. Slim now always groups by datasource, which is correct for 2+ datasources and keeps the payload small (datasource string once per group, not per field). Added multi-datasource + duplicate-caption + embedded (no contentUrl) tests; full path unchanged.

[salesforce-cla]

Thanks for the contribution! Unfortunately we can't verify the commit author(s): Nawar Tamawi <n***@s***.com>. One possible solution is to add that email to your GitHub account. Alternatively you can change your commits to another email and force push the change. After getting your commits associated with your GitHub account, refresh the status of this Pull Request.

[mattcfilbert]

Review — posted by MattGPT (automation on Matt Filbert's behalf, approved by Matt).

P2 — verbosity=slim misattributes fields in multi-datasource workbooks (src/tools/desktop/fields/listAvailableFields.ts, slim projection)

listAvailableFields() (src/desktop/metadata/field-builder.ts:265) iterates all datasources (skipping only Parameters) and returns one flat array where each field carries its own datasource. The slim path hoists fields[0].datasource to the top level and drops per-field datasource — so in a workbook with 2+ datasources, every field from the second datasource onward is silently attributed to the first. That's not just lossy; for same-caption fields across datasources it's actively wrong, and it removes the only disambiguator a client has.

Full mode is unaffected: its header does the same hoist for display but still returns the complete per-field array — so the earlier verification of the hoist as "not lossy" was correct for that path; this only bites slim.

Suggested fix: keep datasource per-field in slim whenever the workbook has more than one datasource (or group slim output by datasource), and add a multi-datasource / duplicate-caption test — current slim tests only cover the single-datasource case.

[mattcfilbert]

Two things before the next round, both verified against current feature/desktop (ca668602):

1. Rebase needed — the branch predates the live-session contract. Current feature/desktop makes workbookFile optional ("Omit workbookFile to read the live session workbook") so the lean/default profiles can call this tool with only a session while the XML/cache tools are hidden; this branch's head still has workbookFile: z.string() required and a cache-path callback. Please rebase and keep the live-session path working for both verbosity=full and slim — and add a slim test for session without workbookFile.

2. Run the byte-ratchet tests before merge. src/server.desktop.test.ts gates the 46k full-surface cliff and a 1,200-byte per-tool budget (exceptions are enumerated explicitly, e.g. bind-template). The new verbosity enum + description lands on a tool that's in the default profiles; if it trips the ratchet, trim the param text rather than raising a cap.

The multi-datasource slim grouping itself looks right, and dropping column_ref from slim is fine given resolve-field exists for the follow-up. Nit (no action needed to merge): the contentUrl extraction is only exercised through mocks — a real published-datasource XML fixture at the metadata level would make it trustworthy.

(Posted by MattGPT, Matt's review automation; Matt reviewed the findings.)

[tonythebest]

Non-blocking follow-up: the slim projection tests cover mocked contentUrl values, but not extraction from real workbook XML. It would be useful to add a field-builder fixture with one published datasource (repository-location/@_id) and one embedded datasource, including an empty-ID case. I don't think this should block the PR.

[tonythebest]

The multi-datasource issue is fixed, the grouped slim contract is sound, and CI is green. Approved; please rebase on feature/desktop and rerun CI before merge.

[mattcfilbert]

approved!

[nawartamawi]

good points @tonythebest both concerns addressed.

[mattcfilbert]

lgtm