KingGyuSuh

Ultracodex — AI skill for Claude Code

AI community

Blend Codex headless into Claude Code's Workflow (ultracode) orchestration — cross-model adversarial verification, judge panels with a Codex juror, and second opinions.

How to install Ultracodex

This entry records only its repository, not the path inside it, so there is no exact command to give. Open KingGyuSuh/ultracodex and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Ultracodex does

Blend Codex headless into Claude Code's Workflow (ultracode) orchestration — cross-model adversarial verification, judge panels with a Codex juror, and second opinions.

Alternatives in AI

  • Codex — openai-codex Adversarial code review, Codex CLI integration, cross-model analysis 9.6k ★
  • Open Multi Agent — TypeScript multi-agent orchestration engine — one runTeam() call from goal to result 5.8k ★
  • Rb Setup — First-time setup 1.3k ★

README

ultracodex

**English** · [한국어](./docs/README.ko.md) · [日本語](./docs/README.ja.md) · [简体中文](./docs/README.zh-CN.md)

**Claude orchestrates. Codex cross-checks.** Blend Codex headless (`codex exec`) nodes into Claude Code's own Workflow (*ultracode*) orchestration — so a genuinely different model family verifies, judges, and second-guesses at the points where same-model agreement is least trustworthy.

`ultracodex` is an open-source Claude Code plugin — a single skill, `codex-workflow` — that teaches Claude how to mix [Codex](https://github.com/openai/codex) headless nodes into a **Workflow**: Claude Code's `agent()` / `pipeline()` / `parallel()` orchestration tool, a.k.a. *ultracode*. The orchestration logic stays in Claude's JavaScript; only *who does the work* changes at chosen nodes. This is **"Pattern A"** — and it is additive, not a model-switcher. Both families run in one Workflow.

Workflow (Claude JS orchestration)
 ├─ find / generate / synthesize .. Claude agent()  ← broad, fast, cache-warm
 └─ verify / judge / 2nd-opinion ... codex node     ← GPT, independent failure modes

**GPT-5.6 ready.** Model-agnostic by design: the GPT-5.6 generation (**Sol / Terra / Luna**, Codex CLI ≥ 0.144.0) worked with ultracodex on day one, no changes needed. The docs now add per-node **tier & effort** guidance — cheap tiers for wide verify fan-outs, the flagship (up to `max`, or `ultra` on Sol/Terra) for load-bearing verdicts — with the choice left to the orchestrating model.

**Requires ultracode mode.** Because the skill *authors and runs* Workflows, the **Workflow** orchestration tool must be available — which in Claude Code means **ultracode mode** (enable it with `/effort` → ultracode; it turns on dynamic workflow orchestration). Without it the plugin still loads and the skill is found, but there is no Workflow tool to drive. See [Prerequisites](#prerequisites).

Why it exists — the philosophy

More verifiers from the *same* mo