dhpk-codex-bridgelisted

# Codex Bridge Use this skill only after routing has selected an explicit CLI handoff or second opinion. The dedicated `dhpk-codex-bridge` subagent hands a **self-contained** task to the GPT-5.6 family through the Codex CLI (`codex exec`) and relays its bounded, redacted output. The bundled `scripts/run-codex.sh` owns sandbox selection, approval policy, and output capture; this skill defines when to outsource, how to compose the prompt, and how to report the result. ## When to use - **Explicit opt-in** — the caller selects this bridge or a named second-opinion option. This bridge is not a default runtime path. - **Clear-spec bulk work** — a mechanical or well-specified implementation / data-analysis / transformation task that a cheaper capable model can do while Claude stays on higher-value work. - **Independent second opinion** — a review of a plan, root-cause diagnosis, or diff where you want a view that does **not** inherit Claude's reasoning (blind second perspective). ## When NOT to Use - The task needs our **conversation context** — Codex gets a fresh session and sees only the prompt. If you can't make the prompt self-contained, don't use this. - **Interactive / iterative** pairing — this is one-shot; there is no back-and-forth. - **In-session structured review** with a review-loop — use `change-verdict` (current-model or `scripts/review.sh --backend cli`) instead (see the retained paths below). - No explicit Codex opt-in — keep the work on the normal codex-free