codex

Invokes Codex CLI non-interactively for read-only consultation or delegated workspace-write tasks.

1|Updated Aug 24, 2026
One-click install
npx skills add https://github.com/mjun0812/skills --skill codex-mjun0812
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: codex
Source: https://github.com/mjun0812/skills/tree/main/skills/delegation/codex
Command: npx skills add https://github.com/mjun0812/skills --skill codex-mjun0812

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Getting a second opinion from Codex or offloading implementation work to it normally requires manual CLI interaction and context re-explanation. This Skill standardizes how the agent calls Codex CLI in non-interactive mode, packaging the prompt, model, reasoning effort, and sandbox level into a repeatable workflow. ## Core Features & Use Cases - Consultation Mode (default): Runs Codex in read-only sandbox to answer questions, review code, or provide a second opinion without touching files. - Delegation Mode: Runs Codex with workspace-write permissions to execute edits, but only when the user explicitly asks to delegate work. - Model & Effort Selection: Chooses among gpt-5.6-luna, gpt-5.6-terra, and gpt-5.6-sol with matching reasoning effort levels based on task complexity, with fallback from xhigh to high on failure. - Use Case: Ask "Codexにこの設計をレビューして" and the agent builds a self-contained prompt with file paths and context, runs codex exec in read-only mode, and summarizes agreements and disagreements with its own view. ## Quick Start Ask the agent to consult Codex about a design decision or delegate a specific implementation task to Codex, and it will run the CLI with the appropriate sandbox mode.

Frequently Asked Questions about codex

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

FAQPage Schema
How do I get a second opinion from Codex CLI inside an agent session?

Ask the agent to consult Codex about your question. It builds a self-contained prompt with background and file paths, runs codex exec in read-only sandbox mode, and summarizes the answer alongside its own agreements and disagreements.

How do I delegate a coding task to Codex CLI non-interactively?

Explicitly ask the agent to have Codex perform the work. It runs codex exec with workspace-write sandbox and -a never so no approvals are prompted, then reports the resulting changes for the calling agent to verify.

Which Codex model and reasoning effort should I use for a task?

Use gpt-5.6-luna with medium effort for light checks, luna with xhigh for normal work, gpt-5.6-terra with high for design decisions, and gpt-5.6-sol with xhigh for the hardest problems. When unsure, pick luna with xhigh.

Does the Codex delegation skill run without explicit user request?

No. The skill is only used when the user explicitly asks to consult or delegate to Codex, and model invocation is disabled. The agent must not invoke it on its own judgment or when the agent itself is Codex.

What happens if the codex CLI is not installed or authentication fails?

The skill checks for the CLI with which codex before running. If the CLI is missing, authentication fails, or a permission error occurs, it stops immediately and reports the error to the user instead of retrying.