Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    codex

    1

    Hand one scoped task to OpenAI Codex on your worktree and review its unstaged edits before you commit.

    Free

    Security scanned
    codex

    codex

    Example session with this skill installed

    Ask codex for a second opinion on my uncommitted auth middleware change.

    • Read your context and instructions
    • Compiled the codex

    Codex runs read-only from the worktree root; the session reports Codex's final message from the last-message file, the git status, and the Codex session id to resume the thread.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Codex as a second pair of hands — it edits, your session decides what lands.

    What you get

    The skill runs codex exec non-interactively from your worktree root:

    • Unstaged edits — Codex changes files; it cannot stage or commit. Your session reviews git diff and commits.
    • A last-message file — Codex's final answer written to the scratchpad, checked after every run along with git status.
    • A session id — kept from the run's output, so later rounds resume the same Codex thread with its context.
    • Steering — codex queue drops a message into the thread; the next resume obeys it.

    Long runs go to the background. A failed run without a final message reports Codex's stderr error line instead of guessing.

    The sandbox has a ceiling: read-only for reviews, workspace-write for edits, never wider — no extra dirs into .git, no bypass flag.

    How you use it

    AskDoes
    /codex <brief> / "ask codex"fresh run with workspace-write
    /codex sandbox:read-only <brief> / "codex second opinion"review run that cannot edit
    /codex resume:<session id> <follow-up>continues that Codex session: answers, fixes, re-reviews

    The package ships /codex-implement-review-demo beside it (codex-implement-review-demo/SKILL.md): a scripted walk-through of review, resume and steering in a scratch repo.

    Setup

    Claude Code, a git repository, and the OpenAI Codex CLI installed and logged in — codex login status must answer logged in.

    What leaves your machine

    The brief and whatever Codex reads from your worktree go to OpenAI through your own Codex login. The skill adds no other network path.


    Tested per version — smoke-checked before every release.

    Codex never stages or commits; your session reviews the diff and commits. The sandbox stops at workspace-write, with no extra dirs into .git and no bypass flag. Nothing runs until codex login status answers logged in.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 18 days ago

    • Free with an Agensi account

    Requires

    git
    OpenAI Codex CLI (logged in)

    Permissions

    Read Files
    Write Files
    Terminal / Shell
    Network Access

    Allowed Hosts

    OpenAI (through the Codex CLI)

    File Scopes

    repo
    /**
    scratchpad
    /codex-last.md

    Grants only Read, codex exec, codex login status and codex queue; Codex edits land unstaged in the worktree.

    Listed18 days ago

    What's inside

    Frequently Asked Questions

    No. Codex edits land unstaged; a linked worktree's git index sits outside the Codex sandbox, so your session reviews `git diff` and commits.

    No. workspace-write is the ceiling: no --add-dir into any .git dir and no --dangerously-bypass-approvals-and-sandbox.

    Pass `sandbox:read-only` with your brief; fresh runs default to workspace-write.

    Pass `resume:<session id>` with the follow-up; the session id comes from the first run's output and the thread keeps its context and sandbox.

    `codex queue` drops a message into the thread; the thread consumes it on its next turn, so the next resume obeys it.

    A run that exits nonzero without a final message is reported with Codex's stderr error line instead of a guess.

    The brief and whatever Codex reads from your worktree go to OpenAI through your own Codex login. The skill adds no other network path.