Arch1eSUN/Arcgentic

Mechanical plan/dev/self-audit/external-audit gates for AI coding agents, with a configurable role-routing topology engine, an MCP-UI live status panel, and native-tooling Claude Code V2 dispatch. Claude Code, Codex, OpenClaw, and PyPI.

What it solves

Arcgentic provides a structured engineering harness for AI coding agents like Codex and Claude Code. It prevents "vibe coding"—where sessions drift, scope changes silently, and tasks are marked as "done" without verification—by replacing ad-hoc prompting with a gated, repeatable engineering workflow.

How it works

It acts as a layer around the coding agent, assigning specific roles and stop states to ensure work is disciplined. The workflow follows a sequence: an Orchestrator manages routing and dispatch; a Planner brainstorms and structures the project; a Developer builds and performs a self-audit; a Test role (optional) handles realistic user testing; and an Auditor provides an independent review to decide if the work passes or needs fixes.

Users can choose between two modes:

  • Single session, multiple agents: Faster execution where roles run within the same session.
  • Multiple sessions, multiple threads: Stronger isolation and audit discipline using separate project threads for each role.

Who it’s for

  • Heavy users of Codex and Claude Code performing complex, multi-round engineering work.
  • AI-native engineering teams and agent builders who require clear role boundaries and handoffs.
  • Developers who need a verifiable audit trail of planning, testing, and auditing for AI-generated code.

Highlights

  • Gated Workflow: Implements a strict sequence of planning, development, self-audit, and external audit.
  • Role-Based Dispatch: Uses fixed roles (Orchestrator, Planner, Developer, Test, Auditor) to maintain consistency across rounds.
  • Custom Topologies: Allows projects to override the default role/state routing via a topology configuration.
  • MCP-UI Integration: Includes an optional MCP server that renders a live round-status panel for supported graphical clients.
  • Platform Support: Verified automation for Codex and experimental support for Claude Code.

Related

  • Project
    appautomaton/latex-arxiv-SKILL

    An agentic harness for writing verified AI/ML review papers in LaTeX, featuring a gated, issue-driven workflow that prevents citation hallucinations.

  • Project
    awslabs/aidlc-workflows

    AI‑DLC is an AWS Labs framework that adds structured, auditable workflows to AI coding assistants (Claude, Kiro, Codex, Cursor, Opencode, Copilot). It provides a harness‑neutral core, multiple workflow profiles, human approval gates, and a detailed audit trail, enabling repeatable AI‑driven software delivery across many popular AI development environments.

  • Project
    maxritter/pilot-shell

    A professional engineering harness for Claude Code and Codex that adds durable context, quality gates, and runtime verification to AI-assisted coding.

  • Dispatch
    Scaling AI Coding: How Claude Code Navigates Large Codebases

    An exploration of agentic search and the 'harness' architecture used by Claude Code to manage multi-million line monorepos and legacy systems without relying on traditional RAG indexing.

  • Project
    a5c-ai/babysitter

    A deterministic orchestration framework that enforces strict, code-defined workflows for AI agents to prevent hallucinations and ensure reliability.