← ClaudeAtlas

hey-codexlisted

Use when the user wants to reach OpenAI Codex CLI — ask it a question, hand it a task by spec so it writes the code in the project, get a second opinion from another model, read an image or a screenshot, generate a picture; also updating Codex CLI itself and diagnosing its errors. Triggers: "ask codex", "call codex", "give codex a task", "let codex write it", "send this to codex", "ask codex to write/review/look at/draw", "what does codex say", "ask codex to clarify", "have codex generate an image", "hey codex", "update codex", "which codex version", "codex crashed", "codex is not responding", "codex error", hey-codex, codex, codex exec. Russian: «вызови кодекс», «спроси кодекса», «дай кодексу задачу», «отправь в кодекс», «попроси кодекса написать/проверить/посмотреть/нарисовать», «что скажет кодекс», «уточни у кодекса», «пусть кодекс сгенерирует картинку», «эй кодекс», «обнови кодекс», «какая версия кодекса», «кодекс упал», «кодекс не отвечает», «ошибка codex».
warodan/hey-codex · ★ 1 · AI & Automation · score 74
Install: claude install-skill warodan/hey-codex
# hey-codex — calling the OpenAI Codex CLI Codex runs locally (`codex`, npm `@openai/codex`, sign-in through ChatGPT). This file is how to call it. Rare flags, diagnostics and pitfalls live in `REFERENCE.md` next to it. **What the machine needs:** Node.js and npm, `codex` installed (`npm install -g @openai/codex`) with `codex login` done, a POSIX shell for `sh` and for stdin redirection (on Windows — Git Bash; §2 has a separate PowerShell form), git for §3, network access. If `codex` is missing or not logged in — tell the user and do not try to install it yourself. ## 0. Update check — first thing `check_update.sh` sits next to this file. Run it at the start of a call: ```bash for p in "${CLAUDE_SKILL_DIR}" ~/.claude/skills/hey-codex ~/.agents/skills/hey-codex \ ./.claude/skills/hey-codex ./.agents/skills/hey-codex; do [ -n "$p" ] && [ -f "$p/check_update.sh" ] && { sh "$p/check_update.sh"; break; } done; true ``` The line finds the script on its own. `CLAUDE_SKILL_DIR` is not an environment variable — Claude Code substitutes the real skill directory into this file as it loads it, so the first candidate points at wherever the skill was actually installed. An agent that does not substitute it leaves it empty, `[ -n "$p" ]` skips that candidate, and the standard install locations are tried instead. **If the script is not found, just skip the step and carry on** — do not fix paths or hunt for it manually: without it the skill works fully, only the automatic vers