Symbol Graph

The symbol graph is the most important structure in the Codebase surface. It is what allows an agent to state, before writing a single character, exactly what a proposed change will affect.

Nodes#

A node is a named, addressable entity in the source: a function, method, class, interface, type alias, constant, module or configuration key. Each carries a stable identifier derived from its qualified path and signature rather than from its file offset, so reformatting, moving a file, or reordering declarations does not invalidate references to it.

Edges#

Blast radius#

Blast radius is the transitive closure of reverse edges from the symbols an edit touches, bounded by a depth limit and by interface boundaries. It is computed before the edit and attached to the change plan, which converts an unbounded risk into a number the planner and a reviewer can both reason about.

target      resolveSessionToken/1
direct      14 references across 6 modules
transitive  63 symbols (depth 3, stopped at 2 package boundaries)
crosses     1 published interface  →  requires interface review
covered by  9 tests  (3 unit, 5 integration, 1 contract)
uncovered   2 references in modules with no attributed tests
            · session/legacy-bridge
            · admin/impersonation

Cross-language edges#

Many real breakages happen at language boundaries: a schema definition consumed by three services in two languages, an interface description generating clients, a query string embedded in application code. Where these boundaries are declared in a machine-readable form, the graph links them, so renaming a field in a schema surfaces the consuming code in every language rather than only in the one being edited.