Token-cheap structural summary rooted at a file or folder path (workspace-relative). Structural facts only (names, signatures, line ranges) — no docstrings, use get_details for those. Increase depth to drill from folder -> file -> function/class detail.

Every function that calls fn (a node id from get_context/get_details), directly or transitively.

Every file (workspace-relative path) that file transitively depends on via its import edges.

Transitive ancestors and descendants of a class (a node id from get_context/get_details).

Full structural facts for one node (a node id from get_context), plus its full stored docstring if the source actually has one.

Prefix: code: http://example.org/code-ontology# Node IRIs: urn:code:<repo/relative/path.py> (module), urn:code:<path.py>::<Qualified.Name> (class/function).

Classes: code:Module, code:Class, code:Function Datatype properties: code:name, code:startLine, code:endLine, code:signature, code:returnType, code:description, code:contentHash Object properties: code:defines (parent->child), code:belongsTo (child->parent, inverse of defines), code:reads, code:writes, code:raises, code:decoratedBy, code:imports, code:calls, code:inherits (imports/calls/inherits objects may be a string Literal if unresolved, or a NamedNode if Pyright resolved the target — both may be present)

Named graphs: urn:code:graph:<path.py> holds one file's direct/structural triples; urn:code:graph:inferred holds the fully materialized transitive closure (R+, includes direct edges) of calls/imports/inherits across the whole repo.

Every query runs with use_default_graph_as_union=True, so a plain WHERE clause with no GRAPH block already searches across every named graph — add an explicit GRAPH urn:code:graph:inferred { ... } block only when you specifically want the transitive/materialized edges rather than a single file's direct ones.