claudex-loop

Skills for working across Claude Code and Codex.

This repository contains separate skills for choosing a model, making a one-off handoff, and running a complete development workflow. They share a repository and plugin distribution; Claudex Route is independent of the Claudex Loop workflow.

Skill	Use it for	Dependencies
claudex-route	A model recommendation or one scoped handoff	Self-contained; selected CLI needed only for delegation
claudex-loop	Requirements, plan review, implementation, and final inspection	Both CLIs and Python 3.10+
codex-review	Explicit Codex plan-review compatibility command	Shared claudex-loop skill
codex-build	Explicit Codex builder compatibility command	Shared claudex-loop skill

Claudex Loop

Claudex Loop gives a plan an independent review before implementation, then gives the code an independent inspection. Your current conversation handles requirements and coordination; the other provider challenges the plan with concrete evidence. The host arbitrates findings, records decisions, and keeps the loop bounded.

Start here	Requirements and plan	Plan review	Default build	Final inspection
Claude Code	Current Claude session	Codex	Claude	Fresh Codex session
Codex	Current Codex session	Claude	Codex	Fresh Claude session

Choose either builder with builder=claude or builder=codex. The inspector follows the builder choice and always uses the other provider. If the coordinator takes over fixes, those new edits need another independent inspection. With mixed authorship, the log records who wrote and reviewed each part.

Model choices remain configurable. Use Claude Fable 5.1 and GPT-6 Astra when selected and available on your accounts, or retain each CLI's configured model. The host UI selection does not automatically change the other CLI's configuration. Requested and observed model information is recorded separately, and there is no silent model/provider fallback.

Claudex Route: standalone task routing

Use claudex-route when you want help choosing who should handle a task. It recommends a model and a role with a short reason: stay with the current agent, get a second opinion on a plan or implementation, investigate a blocker, or delegate a bounded task. It can perform one handoff when requested. Recommendations alone do not launch another model or authorize edits.

claudex-route: Who should handle this CSV import feature? Prioritize cost.
claudex-route: Recommend a second opinion on this plan before we build.
claudex-route: Pick a suitable model and have it diagnose this failing test read-only.

For example, Luna may suit a focused fixture-generation task, Sol a bounded implementation, and Astra or Fable a difficult review. These are task-fit recommendations, not a fixed ranking; available models, context, verification, and current pricing matter. Staying with your current model is a valid result. Route is a self-contained instruction skill with no Python dependency; a delegated run requires the selected CLI and account access. It does not use the full loop's approval-binding runner.

Claudex Route runs independently. Use Claudex Loop when you want repeated plan review, implementation, and independent inspection.

The Claudex Loop workflow

flowchart LR
    R[Host scouts code and context] --> Q[Settle material decisions]
    Q --> P[Write plan and acceptance checks]
    P --> V[Other provider reviews]
    V -- Revise --> A[Host arbitrates and updates]
    A --> V
    V -- Approved --> B[Authorized builder implements]
    B --> T[Run proof checks]
    T --> I[Other provider inspects final code]
    I --> H[Present diff and remaining findings]

1. Recon: inspect existing code and relevant docs, or research greenfield assumptions. Present an assumptions ledger with sources.
2. Requirements: resolve decisions that change the outcome. Batch independent questions, preserve user intent, and write a plan with observable acceptance criteria and proof commands.
3. Plan review: the other provider reads the plan and relevant code, returns evidence-backed findings, and revisits revisions in the same session. Stop at the round budget or an explicit verdict: APPROVED, REVISE, or BLOCKED.
4. Build and inspect: once implementation is authorized, the selected builder works from the plan. Independently run proof checks and inspect the final changes with the other provider in a fresh session.

The user controls consequential decisions and authorization. A request to review a plan does not authorize implementation. A request to plan and implement does not need redundant build approval. Commits, pushes and publication follow the user's existing instructions.

PLAN.md records what to build; PLAN-REVIEW-LOG.md records the findings, dispositions, models, proof and remaining uncertainty. Both paths are configurable. Detailed CLI diagnostics live in a unique directory outside the target checkout.

Install

Both CLIs must be installed and authenticated for the full cross-provider workflow. Python 3.10+ runs the shared adapter; no runtime pip packages or separate API keys are required. Check codex --version, codex login status, claude --version and claude auth status. See the runtime reference for tested CLI versions and permission boundaries.

Claude Code plugin

/plugin marketplace add chaseai-yt/claudex-loop
/plugin install claudex-loop@claudex-loop

This is the beginning of the README. Read the rest at the original. Read the full README ↗