codex

An agent skill by garrytan, from garrytan/gstack. Tags: cli, development, openai.

What it does

OpenAI Codex CLI wrapper — three modes. (gstack)

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add garrytan/gstack --skill codex

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/garrytan/gstack
cp -r gstack/codex ~/.claude/skills/codex

Safety box score

Grade F, high risk: 74 of 100 risk points, rated 2026-10-16. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill is a wrapper around the Codex CLI that runs shell commands, probes auth/model state, and sends repo content or plan text to an external AI service for review, challenge, or consult. The main risks are broad command execution, outbound data transfer, and the skill’s own instructions to read and write files under the user’s home dot-directories and repo metadata.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	3 (high)	It runs multiple shell commands and native helpers, including `~/.claude/skills/gstack/bin/gstack-skill-start`, `gstack-codex-probe`, `codex review`, and `codex exec`, so installing it grants a general command-execution workflow.
Network access	3 (high)	It explicitly invokes Codex against the repo diff or embedded plan text and enables `-c 'web_search="cached"'`, so code, prompts, and possibly plan contents are sent to remote Codex/OpenAI services.
Filesystem reach	3 (high)	It reads and writes outside the project, including `~/.claude/`, `~/.codex/config.toml`, `.context/codex-session-id`, `.git/info/exclude`, and `.gstack/tmp/`, and the consult/review paths require writing temp prompt files and saving session state.
Secrets & credentials	2 (moderate)	It probes and accepts credential sources like `$CODEX_API_KEY`, `$OPENAI_API_KEY`, and `~/.codex/auth.json`, and consult mode persists a session id in `.context/codex-session-id`, so it touches authentication state even though it does not directly exfiltrate secrets.
Privilege & persistence	1 (low)	It does not request sudo or install services, but it does recommend `npm install -g @openai/codex` if missing and edits git exclude settings via `git rev-parse --git-path info/exclude`, so there is some user-level system/config interaction.
Instruction hijack surface	3 (high)	The instructions tell the agent to treat the skill file as executable workflow, follow one-time blocks from `gstack-skill-start`, and obey prompts embedded in tool output, which is high-risk prompt-injection surface by design.
Irreversible actions	2 (moderate)	It can independently launch reviews/challenges, write plan-file status reports, and save session state, but it does not directly deploy, pay, delete, or post to external systems without the user’s explicit invocation.
Transparency	1 (low)	The files are plain text and readable; there are no binaries or decoded blobs, though behavior depends on external helpers like `gstack-codex-probe` and the Codex CLI.

Before installing

Read `SKILL.md` and `sections/*.md` and confirm you want a tool that sends repo content to Codex/OpenAI.
Check whether you are okay with writes to `~/.claude/`, `~/.codex/`, `.context/`, and `.git/info/exclude`.
Run it once in a throwaway repo or scratch branch before using it on sensitive source.
Verify the helper scripts it calls, especially `gstack-codex-probe` and `outside-review-result.ts`, since they are outside this bundle.

Rated 2026-10-16 by an automated review of 5 files from the skill's repository.

Source

Repository: garrytan/gstack (all skills from this repository)
Path: codex/SKILL.md
Branch: main
Updated: 2026-10-05

sega-master-system-devkitsms — Program, build, debug, and port Sega Master System (and Game Gear / SG-1000) homebrew games in C with devkitSMS (SMSlib + PSGlib) and the SDCC Z80 compiler.
unity-cli — Use when interacting with Unity CLI from the terminal, or to control a running/connected Unity Editor from the command line — create or modify GameObjects.
address-sanitizer — Builds and runs code under AddressSanitizer to catch buffer overflows, use-after-free, and other memory errors during fuzzing or tests.
aflpp — Sets up and runs AFL++ for multi-core fuzzing of C/C++ projects built with afl-clang-fast or afl-gcc-fast.
agents-debug — Use when your agent or environment is broken — wrong answers, errors, timeouts, tool failures, or CLI issues. Reads traces and logs to diagnose root causes.
agents-get-started — Use when a developer wants to create a new agent project or get started with AgentCore.