Glossary

A directional dependency rule between code layers (e.g., tests → framework → utilities) that, when violated, indicates structural drift in the codebase.

The process of analyzing source code into an Abstract Syntax Tree to extract structural information like classes, functions, and import relationships without executing the code.

A feedback cycle of running graphify benchmark after graph configuration changes to measure whether modifications improve or degrade the token reduction ratio.

A directed graph representing which functions or methods invoke which other functions, used to trace execution paths through a codebase without reading source files.

A cluster of graph nodes that are more densely connected to each other than to the rest of the graph, automatically named by an LLM during the clustering phase to represent a logical subsystem.

The fixed token budget available to an AI model in a single session, which knowledge graphs help conserve by replacing bulk file reading with structured relationship data.

A directed relationship between two nodes in a knowledge graph, representing connections like "imports," "calls," "inherits from," or "defined in."

A unified knowledge graph at ~/.graphify/global-graph.json that merges node and edge data from multiple repositories to enable cross-repo structural queries.

The condition where a knowledge graph no longer reflects the current state of the codebase due to file additions, deletions, or refactors since the last graph build.

A prompt that references community names, node identifiers, or structural relationships from the knowledge graph to scope the AI agent's attention and reduce discovery overhead.

A recurring structural pattern in the graph where multiple file groups share the same shape (same imports, inheritance, and naming conventions), indicating a replicable convention.

A structured representation of a codebase's entities (nodes) and their relationships (edges) stored as JSON, enabling an AI agent to navigate code structure without reading every source file.

An entity extracted from source code and represented in the knowledge graph — typically a class, function, method, module, or file.

The process of identifying structural regularities (shared imports, inheritance chains, naming conventions) from graph data and articulating them as explicit constraints for AI code generation.

An opening prompt pattern that loads the graph context and immediately scopes the AI agent to a specific community and task, eliminating the discovery phase.

The ratio between total source corpus tokens and graph.json tokens, representing the theoretical maximum context savings when using a knowledge graph instead of reading all files.

A parser generator tool that produces concrete syntax trees from source code, used by Graphify to extract structural information (classes, functions, imports, call chains) locally without sending code to external services.

A minimal test (1-2 cases) that proves an operator correctly delegates to a shared foundational behavior, rather than exhaustively retesting that behavior.