EternallLight/codex-pro-cli-wrappers

Fast Codex Spark wrappers for common CLI tools

★ 0Forks 0ShellGitHub ↗Compare

README

Codex Pro CLI Wrappers

Fast pro-* commands for asking Codex Spark focused questions about local CLI tools.

The wrappers run codex exec with a small prompt, gpt-5.3-codex-spark, low reasoning effort, and most ambient context disabled. The result is a lightweight command such as:

pro-git "Summarize my local changes"
pro-rg "Find the auth middleware"
pro-gh "What was I working on last week?"

What Gets Installed

Two real executables are installed:

  • pro-cli: generic wrapper used by most pro-* symlinks.
  • pro-gh: GitHub-specific wrapper with a faster activity-summary prefetch path.

The installer creates symlinks only when the matching local CLI exists:

pro-cmux
pro-docker
pro-gcloud
pro-git
pro-jq
pro-kubectl
pro-npm
pro-packx
pro-psql
pro-rg
pro-terraform

Missing tools are skipped. For example, pro-aws is not created unless you add it yourself and have aws installed.

Requirements

  • macOS or Linux with zsh
  • Codex CLI on PATH
  • A Codex account/model setup that can use gpt-5.3-codex-spark
  • ~/.local/bin on PATH, or set BIN_DIR during install
  • Target CLIs installed locally for the wrappers you want
  • gh and jq for the optimized pro-gh activity summaries

Install

Clone the repo and run:

git clone https://github.com/EternallLight/codex-pro-cli-wrappers.git
cd codex-pro-cli-wrappers
./install.sh

If ~/.local/bin is not already on PATH, add this to ~/.zshrc:

export PATH="$HOME/.local/bin:$PATH"

Then restart your shell or run:

source ~/.zshrc

Custom Install Location

Use PREFIX to install somewhere other than ~/.local:

PREFIX="$HOME/.local" ./install.sh

Or set the binary and prompt locations separately:

BIN_DIR="$HOME/bin" CODEX_PRO_HOME="$HOME/.config/codex-pro" ./install.sh

Usage

Ask a focused question:

pro-docker "What containers are running and which ports are exposed?"
pro-kubectl "What context and namespace am I using?"
pro-terraform "Is this directory formatted and valid?"

Pipe a prompt:

printf 'Turn this JSON into a jq filter that lists names\n' | pro-jq

Use the generic runner directly:

pro-cli git "Show the last five commits in this repo"

GitHub Activity Summaries

pro-gh has a fast path for activity questions:

pro-gh "What was I working on last week?"

For last week, it uses the previous ISO calendar week, Monday through Sunday. For past week or last 7 days, it uses a rolling seven-day range.

The wrapper prefetches compact GitHub evidence with parallel gh search calls, then asks Spark to summarize it. This is much faster than making the model discover each gh query on its own.

Verification

Run the smoke tests after installation:

zsh ~/.local/share/codex-pro/tests/pro-cli-smoke.zsh
zsh ~/.local/share/codex-pro/tests/pro-gh-smoke.zsh

The tests stub codex; they verify wrapper wiring and fast-mode flags without spending a model call.

How It Works

The wrappers call Codex like this:

codex --dangerously-bypass-approvals-and-sandbox \
  --disable plugins \
  -c 'project_doc_max_bytes=0' \
  -c 'skills.include_instructions=false' \
  -c 'include_apps_instructions=false' \
  -c 'memories.use_memories=false' \
  -m gpt-5.3-codex-spark \
  -c 'model_reasoning_effort="low"' \
  exec \
  --skip-git-repo-check \
  --ignore-rules

They keep your real HOME and CODEX_HOME, so existing Codex auth and tool auth such as gh, gcloud, or kubectl still work.

Customize Prompts

Prompts live in:

~/.local/share/codex-pro/prompts

The generic wrapper always loads default-cli.md. If prompts/pro-<cli>.md exists, it appends that CLI-specific guidance.

For example, edit:

~/.local/share/codex-pro/prompts/pro-kubectl.md

to adjust how pro-kubectl behaves.

Add Another Wrapper

Create a prompt:

cat > ~/.local/share/codex-pro/prompts/pro-aws.md

Create a symlink:

ln -sf ~/.local/bin/pro-cli ~/.local/bin/pro-aws

The target CLI name is inferred from the wrapper name. pro-aws will look for aws on PATH.

Safety Notes

These wrappers are optimized for speed, not deep planning. They suppress most Codex context and run with bypassed approvals, so use them for quick CLI help, read-only inspection, and small focused questions. The prompts tell the model to avoid destructive commands unless explicitly requested, but the shell invocation itself is permissive.

Contributors

EternallLight

Issues