# Codex > Delegate a task to OpenAI Codex for an independent analysis. Use when asked to get a second opinion, run Codex, use Codex for review, or when a task would benefit from a separate model's perspective — code review, security audit, plan review, architecture analysis, etc. Use when this capability is needed. - Skill: `tomevault-io/codex-2` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/codex-2` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/codex-2/raw - Safety review: pending (external: skill-scanner PASS, skillspector CAUTION) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/codex-2 --- # Codex Run a task through OpenAI's Codex CLI in read-only mode and report the results back. ## Step 1: Determine the prompt - If the user provided a specific task (e.g., `/codex review this PR for security issues`), use that as the Codex prompt. - If the user invoked `/codex` without arguments, ask what they want Codex to analyze. - Prepend context to the prompt as needed — current branch, relevant file paths, or a summary of recent changes — so Codex has enough information to produce useful output. ## Step 2: Run Codex Execute the following command via Bash, substituting `` with the full prompt from Step 1: ```bash codex exec --sandbox read-only -m gpt-5.3-codex -c model_reasoning_effort=high "" 2>/dev/null ``` **Notes:** - Stderr is redirected to `/dev/null` because Codex streams thinking tokens there, which would consume excessive context. - If the command fails (auth error, model unavailable), tell the user and suggest checking that `CODEX_API_KEY` or `OPENAI_API_KEY` is set. - For large codebases, set a generous timeout (up to 10 minutes). ## Step 3: Capture and save output 1. Capture the stdout from the Codex command — this is the final analysis. 2. Write it to a markdown file in the project root: `codex-report-YYYY-MM-DD-HHMMSS.md`, with a brief header noting the prompt used and model. 3. Summarize the key findings inline in the conversation so the user gets an immediate overview without opening the file. ## Output format The inline summary should be concise — highlight the top findings, flag anything critical, and note where the full report is saved. Example: > **Codex analysis complete** — saved to `codex-report-2026-02-07-143022.md` > > Key findings: > - [finding 1] > - [finding 2] > - [finding 3] > > See the full report for details. --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/cthacker) — claim your Tome and manage your conversions.