codex-first-executor

Routes tasks between Claude and Codex executors based on task characteristics.

6|2|Updated Jan 30, 2026
One-click install
npx skills add https://github.com/Jerrylalala/compound-engineering --skill codex-first-executor-jerrylalala
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: codex-first-executor
Source: https://github.com/Jerrylalala/compound-engineering/tree/main/plugins/compound-engineering/skills-custom/codex-first-executor
Command: npx skills add https://github.com/Jerrylalala/compound-engineering --skill codex-first-executor-jerrylalala

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When working with multiple AI coding executors, deciding which one should handle a given task is error-prone and inconsistent. This Skill provides a task-characteristic routing matrix that dispatches work to Claude or Codex based on risk level, context dependency, and task type, with built-in retry, timeout, and fallback handling. ## Core Features & Use Cases - Dispatch Policy Matrix: Routes high-risk changes (auth, payments, migrations) to Claude, mechanical batch patches to Codex, and cross-file refactors to Claude based on context dependency. - Executor Reliability Management: Implements retry on failure, 180-second timeout handling, rate-limit backoff, and automatic degradation from Codex to Claude. - Cross-Review Mode: Supports Codex and Gemini as independent review perspectives via [C] and [G] parameters with weighted multi-model arbitration. - Use Case: During a ce:work session, after the task list is created, the routing decision sends a bulk rename operation to Codex while keeping an authentication refactor on Claude, logging the decision to state.md. ## Quick Start Ask the assistant to decide whether the current task should be dispatched to Codex or executed by Claude using the codex-first routing policy.

Frequently Asked Questions about codex-first-executor

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I decide whether to use Codex or Claude for a coding task?▼

Use the dispatch policy matrix: route high-risk changes like auth, payments, and migrations to Claude, mechanical batch patches like renames and formatting to Codex, and context-heavy cross-file refactors to Claude. Independent algorithms and documentation can optionally go to Codex.

How does Codex cross-review work with the [C] parameter?▼

With ce:review [C] or ce:brainstorm [C], Claude executes the main task and produces results, then Codex independently reviews without modifying code. Conclusions are merged using weighted trust scores: Claude at 1.0 and Codex at 0.85.

What happens when Codex times out or hits a rate limit?▼

Codex calls use a 180-second timeout. On failure the call retries once after 30 seconds; on rate limit (429) it waits per the retry-after header. If retries fail, execution degrades to Claude and the reason is recorded in state.md.

Can Gemini be used as the main executor instead of Codex?▼

No. Gemini is currently only active as a cross-review perspective via the [G] parameter in ce:review and ce:brainstorm. The executor routing matrix only routes between Claude and Codex; Gemini as a primary executor is deferred.

What must run before the executor routing decision?▼

The Executor Capability Gate check (skills-custom/executor-capability-gate) must pass before any routing decision is executed. This verifies the target executor is available and not rate-limited before dispatching the task.