Data Flow

Analysis Pipeline

graph TD
    REPO[Repository] --> PARSE[1. Parse<br/>tree-sitter]
    PARSE --> STORE[2. Store<br/>SQLite]
    STORE --> GRAPH[3. Graph<br/>petgraph + Louvain]

    GRAPH --> METRICS[4a. Metrics<br/>SLOC, complexity]
    GRAPH --> ANALYZE[4b. Analyze<br/>anti-patterns]
    GRAPH --> DIAGRAM[4c. Diagram<br/>Mermaid]

    METRICS --> NARRATE[5. Narrate<br/>Claude Code]
    ANALYZE --> NARRATE
    DIAGRAM --> NARRATE

    NARRATE --> LEARN[6. Learn<br/>curriculum]
    LEARN --> DB[(SQLite DB)]

Step Details

Step	Input	Output	Parallel?
Parse	File system	ParsedFile[] (symbols, imports, calls, heritage)	Yes (rayon)
Store	ParsedFile[]	SQLite rows (files, symbols)	No
Graph	Stored data	KnowledgeGraph (petgraph + communities + entry points)	No
Metrics	Parsed files + graph	Complexity, fan-in/out, heatmap scores	Yes (rayon)
Analyze	Parsed files + graph	Pattern detections (god class, circular dep, etc.)	Yes
Diagram	Graph	Mermaid strings	No
Narrate	Graph + context	8 narrative types stored in DB	Sequential (LLM calls)
Learn	Graph + metrics	Chapters, sections, quizzes	No

Interactive Server Flow

Browser <---- SvelteKit 5 (embedded) ----> Axum API ----> SQLite
   |                                          |
   +--- WebSocket <-- EventBus (broadcast) ---+
                          |
                    Claude Code CLI
                    (subprocess, stream-json)

• REST API serves all read endpoints (files, symbols, graph, communities, chapters, etc.)
• SSE streams Q&A responses from Claude Code
• WebSocket pushes real-time events during analysis

Harvest Pipeline

GitHub Trending --> codeilus-harvest --> shallow clone queue
                                              |
                                    codeilus analyze (full pipeline)
                                              |
                                    codeilus-export --> single HTML per repo
                                              |
                                    codeilus deploy --> Cloudflare Pages