Navigating the graph

The bundled navigation skill teaches the agent to answer structural questions with graph tools instead of reading files or grepping. The core strategy:

1. Locate the symbol — by distinctive name via search_symbols, or deterministically via get_file_structure(path) when you know the file. Either way you get a node ID.
2. Narrow with graph traversal — once you have a node ID, use get_callers / get_callees / find_references rather than reading source files to understand relationships.
3. Read source only for implementation detail — get_node_info returns the source snippet for a node; read whole files only when you genuinely need surrounding context.

Question → tool

Question	Tool
Where is create_order defined?	search_symbols("create_order") (distinctive name)
Where is Location defined? (common noun)	get_file_structure(path) or search_symbols("models.Location")
What does create_order call?	get_callees(id, max_depth=2)
Who calls create_order?	get_callers(id, max_depth=3)
What references OrderService?	find_references(id)
What symbols are in order_service.py?	get_file_structure("order_service.py")
Show source + signature of a symbol	get_node_info(id)
How does this Python service talk to the TS client?	get_cross_language_calls(id)
What's around this class in the graph?	get_neighbors(id, depth=2)

Do not

• Read entire source files to find callers — use get_callers.
• Search a bare common noun and trust the result — it gets drowned by file/import nodes; use a distinctive/qualified name, get_file_structure, or grep as a fallback.
• Assume an edge list is complete when resolver_status != ok.