Skip to content

ARCHITECTURE_OVERVIEW.md

Latest commit

 

History

History
24 lines (19 loc) · 3.21 KB

ARCHITECTURE_OVERVIEW.md

File metadata and controls

24 lines (19 loc) · 3.21 KB

Auto Claude Architecture Overview

High-level layout

  • Backend (apps/backend/): Python-based autonomous coding framework that runs multi-session builds via specs and orchestrates agents, QA, and project automation.
  • Frontend (apps/frontend/): Electron + React desktop UI that drives the framework and provides workspace management, terminals, and feature flows.

Backend

  • Entry point: apps/backend/run.py exposes the CLI for running specs, merging/reviewing builds, and managing workspaces, built around multi-session autonomous coding flows and Git worktrees for isolation.【F:apps/backend/run.py†L1-L76】
  • Project structure: agent execution, analysis, CLI, integrations, prompts, QA, and spec handling are organized into focused packages under apps/backend/.【F:apps/backend/README.md†L82-L117】
  • AI client layer: apps/backend/core/client.py centralizes Claude SDK configuration, auth, and tool selection (built-in file operations, optional MCP tools for Linear, Graphiti memory, Context7 docs, Puppeteer/Electron automation) to enforce sandboxing and per-agent tool scopes.【F:apps/backend/core/client.py†L1-L200】
  • Prompts: reusable prompt templates for planners, coders, QA reviewers/fixers, ideation, roadmap discovery, and specs live in apps/backend/prompts/, with additional MCP tool definitions in apps/backend/prompts/mcp_tools/ and packaged prompts under prompts_pkg. These files are loaded by agent flows to seed AI calls.

Frontend

  • Desktop stack: Electron main process, preload, and React renderer organized in a feature-based architecture under apps/frontend/src/, covering agent management, terminals, projects, settings, roadmap/ideation, insights, changelog, GitHub, worktrees, and onboarding flows.【F:apps/frontend/README.md†L47-L91】
  • Purpose: Provides the desktop UI that communicates with the backend, surfaces progress, and manages local projects/worktrees through feature modules and shared utilities.

AI workflows & execution flow

  • AI calls are funneled through the backend create_client helper, which injects OAuth credentials, selects Claude models, and attaches MCP tools based on agent type and project capabilities, ensuring all agent sessions share consistent security and context settings.【F:apps/backend/core/client.py†L136-L200】
  • Agents use prompt templates from apps/backend/prompts/ to run planning, coding, QA review, and follow-up steps. The CLI (apps/backend/run.py) triggers these sessions according to the selected spec and coordinates workspace isolation and recovery logic.【F:apps/backend/run.py†L6-L24】
  • Frontend feature modules invoke backend commands and display status, terminals, and analysis results, while Electron IPC handlers in src/main manage process communication and secure exposure of Node capabilities to the renderer.【F:apps/frontend/README.md†L52-L91】

Where to start

  • Run the backend CLI (python apps/backend/run.py --list then --spec <id>) to exercise the autonomous pipeline.【F:apps/backend/run.py†L15-L24】
  • Start the frontend (npm run dev in apps/frontend) to explore the UI and IPC-based controls that drive backend sessions.【F:apps/frontend/README.md†L25-L36】【F:apps/frontend/README.md†L93-L109】