Harden Claude Code plans with claudex-loop
Use claudex-loop when a Claude Code operator should lock intent, stress-test a plan with Codex, and cross-inspect the final build before approving risky work.
npx skills add agentskillexchange/skills --skill harden-claude-code-plans-with-claudex-loop
Use claudex-loop when a Claude Code operator is about to build something expensive to get wrong, such as auth, migrations, payment flows, concurrency, schema design, or greenfield architecture. The operator invokes the claudex-loop skill to run Claude-led recon, an assumptions ledger, load-bearing interrogation, a locked PLAN.md, read-only Codex adversarial review rounds, and optional cross-model build inspection where the model that built the change does not grade it. Invoke this instead of using Claude Code normally when the workflow needs documented alignment, a bounded rival-model review loop, and final diff inspection before any high-stakes implementation proceeds. The scope boundary is Claude Code skill orchestration for plan hardening and cross-model review/build inspection; it is not a generic Claude plugin listing, Codex wrapper, planning template, or broad AI coding framework.
What this skill actually does
Inputs and prerequisites: Claude Code with skills or plugin marketplace support, OpenAI Codex CLI 0.130 or newer, authenticated Codex CLI, local repository for brownfield work, optional web or deep research capability for greenfield planning.
Setup notes: Install as a Claude Code plugin with /plugin marketplace add chaseai-yt/claudex-loop and /plugin install claudex-loop@claudex-loop. Alternatively, copy the repository skills directory into ~/.claude/skills/ for bare skill names. Verify Codex CLI is installed and authenticated before invoking /claudex-loop, /codex-review, or /codex-build.
Source and verification boundary: use https://github.com/chaseai-yt/claudex-loop as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Claude Code workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.