codex · prathamdby — Claude Code skill: run local Codex CLI, how to install
A skill for Claude Code and Codex that runs the local Codex CLI when the agent asks. Triggers when a task requires calling Codex CLI with specific flags and no manual approvals.
MIT 331 installs ★ 6 updated 2026-11-10
npx skills add prathamdby/skills --skill codex -a claude-codeWhat codex does
The default invocation is codex exec --yolo -C <abs-dir> "<prompt>" with a 15-minute parent wait. Since --yolo bypasses approvals and sandbox, calling this skill counts as consent. Plain codex opens the TUI without --yolo.
Before running, the skill resolves workspace, prompt, mode, session, worktree, sandbox and timeout. It checks the binary and codex login status, blocking on missing auth. --sandbox, --search, --ask-for-approval pass through only when the user names them. --full-auto gets rejected.
After exit, the skill inspects status, diff and runs tests for changed code. Terminal outcomes: SUCCESS, NO_CHANGES, BLOCKED. If the Codex binary is missing, the skill follows its install reference and blocks.
What it helps with
- run Codex CLI with approvals and sandbox bypassed
- call Codex in review mode for a specific directory
- run Codex in a named sandbox without
--yolo - check
codex loginstatus before execution - run Codex CLI in a worktree for parallel work
Example prompts for the agent
- Run Codex in /home/project with the prompt 'fix the tests'
- Run a Codex review in readonly sandbox
- Run Codex with model gpt-4o in a worktree
How to install codex
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill codex -a claude-code2Manually: put the files in place
~/.claude/skills/codex/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/codex/
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill codex -a codex2Manually: put the files in place
.agents/skills/codex/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill codex -a cursor2Manually: put the files in place
.cursor/skills/codex/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install prathamdby/skills skills/engineering/codex2Manually: put the files in place
.github/skills/codex/SKILL.md1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill codex -a gemini-cli2Platform CLI
gemini skills install https://github.com/prathamdby/skills --path skills/engineering/codexManually: put the files in place
.devin/skills/codex/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add prathamdby/skills --skill codex -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about codex
How to install codex in Claude Code?
Run npx skills add prathamdby/skills --skill codex -a claude-code — the skill file lands at ~/.claude/skills/codex/SKILL.md.
What does the codex skill do?
It runs the local Codex CLI on behalf of the agent: resolves parameters, builds the command with the right flags, executes it and verifies the result.
What does the codex skill need to work?
It needs the Codex binary installed and auth via codex login. Without the binary, the skill blocks and points to install instructions.
Source text
SKILL.md · MIT
--- name: codex description: > When the user wants the local Codex CLI run or its installation, login, or connected tools changed. --- # Codex ## Options and boundaries Default: `codex exec --yolo -C <abs-dir> "<prompt>"`; model omitted; parent wait 15m. Calling this skill authorizes that run: pass `--yolo` without asking. `--yolo` is the listed alias of `--dangerously-bypass-approvals-and-sandbox`; pass `--yolo`, not both. Bare `codex` is an asked-for TUI and omits `--yolo`. "Use <model>" adds `-m`. A named sandbox adds `--sandbox <mode>` and omits `--yolo`. "Worktree" adds boolean `--worktree`, never a name or invented path. "Review" selects `codex --yolo review` unless a sandbox is named. A named sandbox omits `--yolo` and uses `codex exec --sandbox <mode> review`. Missing values or conflicts are `BLOCKED`. `--full-auto` is rejected. `--search` and `--ask-for-approval` are global, before `exec`, and only when the user names them. Skip-git-repo-check, bypass-hook-trust, danger-full-access, approve-for-me, and ignore-rules are added only when the user names that control. For a named sandbox or a sandbox failure, follow `references/sandbox.md`. Do not ask, and do not initialize a repo to evade the git check. ## 1. Resolve Fix workspace, prompt, mode, session, worktree, sandbox, and wait. Check binary and `codex login status` without exposing secrets. Missing binary → follow `references/install.md`; missing auth → `BLOCKED` unless login was requested. Omit `--sandbox` unless the user named it. Record `cmd | workspace | mode | session | worktree | wait | evidence | terminal`. Done when target, binary, auth, and the yolo-or-named-sandbox choice are fixed. ## 2. Run Build argv, quote the prompt literally, and use `-C` for the target. Default text is quiet until exit; silence is not a hang. Wait until exit or the recorded deadline; kill only this run's PID at that deadline. Nonzero exit: capture stderr, inspect the tree, report `BLOCKED`. Do not ask the user and do not add a second bypass. For JSONL/last-message output, follow `references/streaming.md`. For resume/fork, follow `references/sessions.md`. For worktrees/parallel runs, follow `references/worktrees.md`. For review, TUI, login, MCP, or plugins, follow `references/management.md`. On a rejected flag, use scoped `codex --help` or `codex exec --help`. Done when the process exits or a named failure is reported. ## 3. Verify and report Inspect status and diff; CLI text is unverified. Run the narrowest covering tests for edited executable code. Check every child worktree; a named read-only sandbox is not a tree-unchanged guarantee. Remove only clean worktrees this run created and the user did not ask to keep; preserve login, MCP, and plugins. Report command, mode, session/worktree, wait, and outcome evidence. Done when every requested outcome is observed or a blocker is evidenced. Terminals: `SUCCESS`, `NO_CHANGES`, `BLOCKED`.
© prathamdby · MIT · prathamdby/skills
Similar skills
All skills for Agent setup: memory, context, orchestration →codex by other authors comes first.