Choose who should review a plan
Uses claudex-route
- A plan or sufficient plan context in the current conversation
- Any cost, speed, model, or provider constraints
claudex-route: Recommend a second opinion on this plan before we build.

Claudex Loop facilitates a two-act hardening process where one AI model drafts an implementation plan and another independently reviews it in a read-only sandbox. This adversarial approach is designed for technical leads and architects who need to stress-test critical system designs, such as authentication flows or database schemas, before writing code.
By coordinating a bounded revision loop between Claude and Codex, the system ensures that plans are rigorously cross-inspected and locked in a dedicated file to prevent implementation errors in complex environments.
Follow the documented setup, then try a first task.
Installs: Complete claudex-loop plugin bundle, including Claudex Route, Claudex Loop, codex-review, and codex-build. · Claude Code
Enter these commands in a Claude Code agent session, not a terminal.
/plugin marketplace add chaseai-yt/claudex-loop /plugin install claudex-loop@claudex-loop
Installs: Complete collection under the repository's skills directory, copied into both Codex and Claude Code skill locations. The compatibility commands share the claudex-loop runtime. · Codex, Claude Code
Platform: macOS or Linux
Run from: repository root
Clone the repository using your preferred Git workflow, then change into its repository root. The source does not provide an exact clone command.
Run this complete terminal command block from the repository root to install the collection into both hosts.
# macOS / Linux — run from this repository mkdir -p ~/.agents/skills ~/.claude/skills cp -R skills/. ~/.agents/skills/ cp -R skills/. ~/.claude/skills/
Open a new Codex or Claude Code session so it discovers the installed skills.
Installs: Complete collection under the repository's skills directory, copied into both Codex and Claude Code skill locations. The compatibility commands share the claudex-loop runtime. · Codex, Claude Code
Platform: Windows PowerShell
Run from: repository root
Clone the repository using your preferred Git workflow, then change into its repository root. The source does not provide an exact clone command.
Run this complete terminal command block in Windows PowerShell from the repository root to install the collection into both hosts.
# Windows PowerShell — run from this repository New-Item -ItemType Directory -Force "$env:USERPROFILE\.agents\skills", "$env:USERPROFILE\.claude\skills" | Out-Null Copy-Item -Recurse -Force skills\* "$env:USERPROFILE\.agents\skills\" Copy-Item -Recurse -Force skills\* "$env:USERPROFILE\.claude\skills\"
Open a new Codex or Claude Code session so it discovers the installed skills.
Installs: Only the standalone Claudex Route skill; no other skill from the repository is required. · Codex
Run from: repository root
Clone the repository using your preferred Git workflow, then manually copy the `skills/claudex-route/` directory from the repository into `~/.agents/skills/`. The source does not provide an exact terminal command for this standalone copy.
Open a new Codex session to pick up the skill.
Installs: Only the standalone Claudex Route skill; no other skill from the repository is required. · Claude Code
Run from: repository root
Clone the repository using your preferred Git workflow, then manually copy the `skills/claudex-route/` directory from the repository into `~/.claude/skills/`. The source does not provide an exact terminal command for this standalone copy.
Open a new Claude Code session to pick up the skill.
Uses claudex-route
claudex-route: Recommend a second opinion on this plan before we build.
Uses claudex-loop
Use claudex-loop to harden this plan with independent review.
Uses codex-build
Use codex-build to implement the concrete work order at <spec-path>, verify it with <proof-command>, and have Claude independently inspect the final changes. Replace <spec-path> with the actual specification file path and <proof-command> with the agreed command that verifies the deliverable.