Automatic codebase dependency mapping builds a living graph of modules, call chains, and runtime relationships using Tree-sitter and Neo4j, accessible to AI tools via MCP. It surfaces transitive impact, dead code, and risk clusters across repos, reducing duplicate logic and enabling safer refactorin
AI-driven development speeds delivery but hides debt—duplicate logic, dead code, and architectural drift—without a modern audit. A graph-based, auto-updating code map (Neo4j-backed) like Pharaoh reveals reachability, duplication, dead code, and risk across the whole repo. When connected to AI agents
This playbook argues for preventing orphaned AI code by prioritizing code connection and production reachability, using graph checks like Pharaoh to verify exports are reachable. It prescribes an entry-point–first design, upfront codebase mapping, full module context, and wiring as you go. It also e
Dead exports are unreferenced exports in JavaScript—clutter that slows AI-assisted development and increases risk. Detect them with project-wide static analysis, using Knip for monorepos and ts-prune for TypeScript, while watching for false positives from barrels and dynamic imports. Tree-shaking wo
Dead code in fast-moving TypeScript projects slows builds and confuses AI-driven development. It piles up from rapid changes, fear of breaking unseen consumers, AI duplicates, and hidden abstractions. Automated tools like Knip (and ts-prune, unimported) plus graph maps (ENRE-ts, Pharaoh) reveal
Before refactoring with AI agents, you must check blast radius to avoid silent breakages. The article promotes deterministic knowledge-graph blast-radius analysis (Pharaoh) that parses code with Tree-sitter and stores relationships in Neo4j, enabling multi-hop reachability for changes. Integrating t
AI coding accelerates regressions due to limited context, causing silent breakages and duplication in small teams. The guide promotes an architecture-first approach: model the codebase as a living knowledge graph that agents can query to guide changes. It adds guardrails—blast radius, reachability,
AI agents work best with true codebase context—a structured, graph-backed map of modules, dependencies, and endpoints—not file dumps. Flat context misses links and drifts while inflating token costs; deterministic parsing (Tree-sitter) keeps facts reliable. Pharaoh converts TS/Python repos into a Ne
Dead code detection MCP gives real-time, deterministic identification of unused or unreachable code, making its truth accessible to AI agents and tools like Claude Code and Cursor. It exceeds traditional tools by cross-repo reachability with Tree-sitter and graph databases and eliminates runtime LLM
The piece outlines eight strategies to search, validate, and reuse functions at scale in AI-augmented codebases, emphasizing speed, accuracy, and deterministic results. It centers on Pharaoh search_functions, a Neo4j knowledge graph with MCP that delivers import-path aware results and zero per-query
New posts on codebase intelligence, AI coding, and building Pharaoh.