arbor CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Commands

# Build
cargo build --workspace

# Test all crates
cargo test --workspace

# Test single crate
cargo test -p arbor-graph
cargo test -p arbor-core

# Test single test by name
cargo test -p arbor-graph -- ranking::tests::test_pagerank_basic

# Lint
cargo clippy --workspace --all-targets --all-features

# Format check
cargo fmt --all -- --check

# Format fix
cargo fmt --all

# Benchmarks (criterion; CI regression gate in .github/workflows/benchmarks.yml)
cargo bench -p arbor-graph

# Release build (CLI binary)
cargo build --locked --release -p arbor-graph-cli

# Run CLI locally
cargo run -p arbor-graph-cli -- <command>

Architecture

Arbor is a semantic code graph engine — it parses codebases into a dependency graph and exposes that graph to CLIs, GUIs, WebSocket clients, and AI agents (via MCP).

Crate Dependency Order

arbor-core  →  arbor-graph  →  arbor-watcher
                     │                │
                     └───── arbor-server ─────┐
                                              │
                     arbor-mcp ───────────────┤
                     arbor-cli ───────────────┘
                     arbor-gui ──────────────────→ arbor-{core,graph,watcher}

Crate Roles

arbor-core — Tree-sitter AST parsing. Extracts functions, classes, structs, imports, and call edges for 9 production languages (Rust, TS/JS, Python, Go, Java, C/C++, C#, Dart) plus 5 fallback parsers. Each language lives in crates/arbor-core/src/languages/. parser_v2.rs is the active parser; parser.rs is legacy.

arbor-graph — In-memory petgraph + sled persistence. Key modules:

• builder.rs — converts parsed nodes/edges into the graph, builds per-file import maps for cross-module edge filtering
• ranking.rs — PageRank with 10% weight for test-file callers
• heuristics.rs — entry point detection (main, HTTP routes, webhooks, jobs, CLI commands)
• impact.rs — blast radius, shortest path (A*)
• slice.rs — context trimming (token-aware, tiktoken)
• symbol_table.rs — cross-file FQN resolution
• confidence.rs — edge confidence scoring
• store.rs — sled-backed persistence