OOXML Lookup

Query the ECMA-376 (Office Open XML) schema offline to find out what is legal in a .docx, .xlsx or .pptx file. Use when writing, reading or debugging OOXML markup by hand: what may go inside an element and in what order, what attributes it takes, what values those attributes accept, which namespace or prefix to write, and what the difference is between the Transitional and Strict profiles. Also resolves a schema validation error into "here is what would have been legal at that position". Covers wordprocessingml, spreadsheetml, presentationml, drawingml and vml. Does NOT validate files (that is ooxml-validate), does not generate documents, and has no specification prose, behaviour notes or semantic search — it answers from the XSD schema graph only.

The ECMA-376 schema as a local SQLite graph, queried through one CLI. No network, no account, no install — node scripts/ooxml.mjs from a bare checkout.

Every command prints JSON on stdout and nothing else.

The loop

Most questions are one of these three, in this order.

1. What is this thing?

bash

node scripts/ooxml.mjs element w:tblPr

Returns every declaration of the name. More than one is normal and not an error — OOXML declares element names locally, inside types, so w:tblPr is CT_TblPr inside w:CT_Tbl and CT_TblPrBase in three other places. The scope.declared_in field tells you which is which.

2. What is legal here?

bash

node scripts/ooxml.mjs children w:CT_PPr      # what may go inside, in order
node scripts/ooxml.mjs attributes w:CT_Ind    # what attributes it takes
node scripts/ooxml.mjs values s:ST_TwipsMeasure   # what values are legal

children returns two views of the same thing: tree, the nested sequence/choice/all structure with cardinalities, and order, a flat list in schema order. Read order for "what goes here"; read tree when the choice/sequence distinction matters.

min/max of -1 means unbounded.

3. What does this validation error mean?

bash

node scripts/ooxml.mjs explain '{"id":"Sch_UndeclaredAttribute",
  "description":"The '\''bogus'\'' attribute is not declared.",
  "xpath":"/w:document[1]/w:body[1]/w:p[1]/w:pPr[1]/w:ind[1]"}'

Takes a diagnostic from an ooxml-validate report — or the whole report, it will use the first diagnostic — and answers the question that always follows: then what would have been legal there. The xpath does real work: it disambiguates a name with several content models by looking at its ancestors.

An id it does not recognise is not a failure; it still tells you what the schema allows at that position.

Worked example

You are hand-writing a paragraph and want to indent it.

bash

$ node scripts/ooxml.mjs attributes w:CT_Ind --compact
{"query":"w:CT_Ind","profile":"transitional","found":true,"type":"w:CT_Ind",
 "count":12,"attributes":[{"name":"end","qualified":true,"use":"optional",
 "type":{"qname":"w:ST_SignedTwipsMeasure",...

firstLine is typed s:ST_TwipsMeasure. What can you write in it?

bash

$ node scripts/ooxml.mjs values s:ST_TwipsMeasure --compact
{"type":"s:ST_TwipsMeasure","one_of":[
  {"type":"s:ST_UnsignedDecimalNumber","base":"xsd:unsignedLong"},
  {"type":"s:ST_PositiveUniversalMeasure","base":"s:ST_UniversalMeasure",
   "facets":{"pattern":"[0-9]+(\\.[0-9]+)?(mm|cm|in|pt|pc|pi)"}}]}

So w:firstLine="720" (twips) or w:firstLine="0.5in", and now you know the unit suffixes are a closed set.

Things that will bite you

• Always check the profile. Answers default to Transitional, which is what Word, Excel and PowerPoint actually write. Strict uses different namespace URIs for the same vocabulary and drops VML entirely. If a document declares purl.oclc.org/ooxml/... namespaces, pass --profile strict. diff shows what changes for one symbol.
• A name is not an identity. ST_Percentage is a pattern-restricted string in shared-commonSimpleTypes and a union in dml-main; ST_Direction is ltr|rtl in wml, horz|vert in pml and norm|rev in dml-diagram. A bare name like that comes back as ambiguous:true with a variants array rather than one arbitrary answer, so check for variants before reading a result. Qualify names when you can: s:ST_Percentage, not ST_Percentage.
• An empty answer is often correct. attributes w:CT_Tbl returns nothing because wml carries table properties as child elements. found:true with count:0 means "none", not "lookup failed".
• search is substring, not semantic. It finds tblPr; it will not find "how do I make a table border". There are no embeddings here by design.

Names

Write them any way you like — w:tblPr, wml:tblPr, {http://schemas.openxmlformats.org/wordprocessingml/2006/main}tblPr, or bare tblPr. A bare name that matches several vocabularies returns all of them rather than guessing.

Prefixes are the conventional ones (w, a, p, s, m, r, v, x, c, o, xdr, wp, …). Run namespace <uri> to go the other way, or namespace <prefix> to see which vocabulary a prefix reaches.

Two of those — x for spreadsheetml and c for charts — are accepted because the ecosystem writes them, not because a schema binds them; namespace reports those under aliases with a citation, and leaves prefix null, which is what the standard actually says. Answers come back in the canonical spelling, so element x:worksheet replies sml:worksheet. That is deliberate: x is also VML's excel namespace, and printing both as x: would render two different namespaces identically. The lookup checks both and returns whichever actually has the name.

An ambiguous name returns ambiguous:true and a variants array, one entry per distinct meaning. The message says which kind of ambiguity it is: several vocabularies sharing a local name (qualify it, and you get one answer), or one name declared in several places within a vocabulary — w:tblPr — where the answer depends on the parent and each variant carries applies_when_declared_in.

Direct SQL, when the subcommands do not fit

The database ships with the skill and the connection is read-only:

bash

node scripts/ooxml.mjs sql "SELECT local_name FROM symbols
  WHERE kind='complexType' AND local_name LIKE 'CT_Tbl%'"

Tables: profiles, vocabularies, namespaces, prefix_aliases, symbols, symbol_profiles, compositors, child_edges, group_edges, attr_edges, inheritance_edges, enums, simple_type_facets, union_members.

Two things to know before writing a join. Symbols are keyed on the vocabulary, not the namespace URI, because the two profiles are the same vocabulary under different URIs — so symbols has no profile column and every edge table does. And symbols.parent_symbol_id is 0 for a global declaration, not NULL.

Prefer the subcommands where they fit: they resolve inheritance and expand group references, and a raw child_edges query will not.

What this does not do

• Validate a file. That is ooxml-validate, a separate tool. This one answers what the schema permits; it never opens your document.
• Explain what Word actually does. The schema is the standard, and implementations diverge from it. There are no behaviour notes here.
• Search the specification text. No prose, no PDFs, no embeddings. If you need to read the specification itself, https://ooxml.dev does that well and this deliberately does not compete with it.