Skip to content

Layer 6: Data Flow

DTO-to-storage chains and transformation steps.

Overview

Data flows through the system in well-defined stages:

  1. Input — User prompts (stdin), recipe YAML (files), hook events (JSON stdin)
  2. Parsing & Validation — Clap CLI parser, recipe::parser (serde_yaml), HookInput (serde_json), security::XpiaDefender (XPIA pattern scan)
  3. Routing & Classification — commands::dispatch (match on Command enum), workflows::WorkflowClassifier (Q&A / Investigation / Development), recipe::condition_eval
  4. Execution — agent_core::session (conversation state), agentic_loop::loop_core (tool-use cycle), recipe::executor (step runner)
  5. State Persistence — state::AtomicJsonFile (file lock + atomic write), memory::backend (store/retrieve), LadybugDB Kuzu (graph insert/query)
  6. Output — stdout (user response), file output (generated code), GitHub issues (gh API)

Key transformation: user prompt string -> Clap Command enum -> WorkflowType classification -> agent session -> tool-use cycle -> memory store + stdout response.

Diagram (Graphviz)

Data Flow — Graphviz

Diagram source