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 is a harness engineering layer designed to prevent AI coding sessions from drifting. It solves the problem of "vibe coding," where scope changes silently, context is lost, and tasks are marked as "done" simply because the AI assistant claimed so, without rigorous verification.

How it works

Instead of a single long conversation, Arcgentic transforms the coding process into a gated engineering workflow. It introduces a set of fixed roles—Orchestrator, Planner, Developer, Test, and Auditor—each with specific responsibilities and stop conditions.

The process typically follows this sequence:

  1. Orchestrator: Manages routing and role dispatch.
  2. Planner: Brainstorms and defines the project plan and round structure.
  3. Developer: Implements the code, performs local verification, and provides a self-audit.
  4. Test (Optional): Conducts realistic user/session testing if required by the plan.
  5. Auditor: Performs an independent review to decide if the work passes, needs fixes, or is incomplete.

It integrates as a plugin for Codex and Claude Code, managing the creation and handoff between these role-specific threads or agents to ensure a disciplined sequence of work.

Who it’s for

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

Highlights

  • Gated Workflow: Replaces ad-hoc prompting with a repeatable sequence of planning, development, and independent auditing.
  • Role-Based Dispatch: Automates the handoff between specialized roles (Planner, Developer, Auditor) to reduce context drift.
  • Customizable Topology: Allows projects to override the default role-routing graph via a state configuration file.
  • MCP-UI Status Panel: Provides an optional interactive status panel via the Model Context Protocol (MCP) to track round progress and audit verdicts.
  • Codex & Claude Code Integration: Verified automation for Codex and experimental support for Claude Code's sub-agent modes.

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
    Chachamaru127/claude-code-harness

    A disciplined delivery loop and safety layer for AI coding agents that enforces a structured Plan-Work-Review-Release workflow to prevent agent drift.

  • 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
    affaan-m/ECC

    An agent harness operating system that provides a coordinated engineering system and toolbox for AI agents to plan, test, and verify code changes.

  • Project
    SethGammon/Citadel

    An operating layer for Claude Code and OpenAI Codex that preserves repository state between sessions and coordinates multi-step AI coding workflows with explicit verification.