</>MCP Agents Market
Plugin

claudex-loop

by chaseai-yt1.7kPythonUpdated 2026-09-06

Claude Code skill: four-phase plan hardening (recon, interrogate, Codex adversarial review, cross-model build & inspection) — two AI models harden your plan before a line of code exists, then swap jobs to build it. Whoever built it never grades it.

Claude CodeCodex

Claudex Loop is a Claude Code plugin that orchestrates collaborative AI workflows between Claude and OpenAI Codex across four phases: reconnaissance, requirements planning, adversarial plan review, and independent code inspection. The plugin ensures that one AI model hardens your implementation plan while another builds the code, then they swap roles for final inspection—guaranteeing whoever built the feature never grades their own work. It includes both a lightweight routing skill (claudex-route) for model recommendations and task handoffs, plus a full-loop workflow (claudex-loop) that enforces independent review before any code is written. Developers use it to reduce blind spots and improve code quality through structured cross-model verification.

Key Features

Four-phase workflow: reconnaissance, requirements settlement, adversarial plan review, and independent final inspection
Cross-provider collaboration between Claude Code and OpenAI Codex with automatic role swapping
Standalone claudex-route skill for model recommendations and bounded task delegation without full workflow overhead
Approval tracking with SHA256 fingerprinting of plans and code changes to invalidate stale reviews
Configurable builder and inspector roles with explicit model selection (Claude Fable 5.1, GPT-6 Astra)
Bounded review permissions: read-only sandbox for Codex reviews, file-only access for Claude reviews
Append-only decision log recording findings, dispositions, models used, and authorship for mixed-model edits
Configurable round budgets for plan review (default 5), build fixes (default 2), and inspections (default 2)

Use Cases

  • 01Review implementation plans with a different AI model before writing code to catch logic flaws and edge cases
  • 02Build features with one AI provider then inspect the final code with another for unbiased quality control
  • 03Route specific tasks (CSV imports, test diagnostics, fixture generation) to the most cost-effective or capable model
  • 04Harden greenfield project assumptions through independent adversarial review before implementation begins
  • 05Maintain independent audit trails when multiple AI models contribute to the same codebase
  • 06Enforce proof-of-work validation with automated acceptance checks before and after code inspection

Related Plugins

View more

claudex-loop — FAQ

What is the Claudex Loop plugin?+

Claudex Loop is a Claude Code plugin that coordinates multi-phase AI workflows between Claude and OpenAI Codex, ensuring one model reviews plans while the other implements them, then swapping roles for final inspection. It prevents any single AI from both building and grading the same code.

How do I install the Claudex Loop plugin in Claude Code?+

Run '/plugin marketplace add chaseai-yt/claudex-loop' followed by '/plugin install claudex-loop@claudex-loop' in a Claude Code session. For manual installation, copy the skills directories from the repository into ~/.claude/skills/ and restart.

Which AI clients does Claudex Loop work with?+

Claudex Loop works with Claude Code and OpenAI Codex. Both CLI tools must be installed and authenticated on your system for the full cross-provider workflow to function.

Do I need API keys or special prerequisites?+

You need authenticated installations of both the Claude CLI and Codex CLI, plus Python 3.10 or later. No separate API keys are required beyond your existing CLI authentication. Check with 'codex login status' and 'claude auth status'.

Is Claudex Loop free to use?+

The plugin itself is open-source under the MIT license and free to use. However, you'll incur costs from your Claude and Codex API usage according to each provider's pricing for the models you select.

Can I use only the routing skill without the full review workflow?+

Yes, the claudex-route skill works independently with no Python dependency and provides model recommendations or single task handoffs. It's self-contained and can be installed separately from the full claudex-loop workflow.

How do I install claudex-loop?+

Open the source repository on GitHub and follow its README. claudex-loop is a plugin — MCP Agents Market links you directly to the official repo.

Is claudex-loop free?+

claudex-loop is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches