Codex + Gemini + Claude Quick Launch

Cross-platform interactive launchers for Codex, Gemini, and Claude Code.

This package provides lightweight wrapper commands that help you:

• choose a working directory,
• pick a model,
• choose a permission/sandbox mode,
• and launch your selected CLI with consistent prompts.

It supports:

• Windows (PowerShell-based launcher flow)
• macOS and Linux (Node-based launcher flow)

Commands installed

Primary short commands:

• codexq
• geminiq
• claudeq

Legacy aliases:

• codex-quick-launch
• gemini-quick-launch
• claude-quick-launch

Your original upstream CLIs remain available directly:

• codex
• gemini
• claude

Install

npm install -g @openai/codex @google/gemini-cli @anthropic-ai/claude-code https://codeload.github.com/aGamingGod1234/codex-gemini-quick-launch/tar.gz/refs/heads/main

Basic usage

codexq
geminiq
claudeq

Optional dry-run (show command without launching):

codexq --no-launch
geminiq --no-launch
claudeq --no-launch

Overall launch flow

When you run any launcher (codexq, geminiq, claudeq), it follows this high-level flow:

1. Check for updates for this quick-launch package.
2. If an update is available, it attempts to auto-update by running the configured global npm install command.
3. Continue to launcher UI (even if update fails).
4. Ask for target directory.
5. Ask for model (unless passed by arg where supported).
6. Ask for permission/sandbox mode.
7. Show summary.
8. Launch selected CLI (or print command if --no-launch).

Platform behavior

Windows

• Uses scripts/agent_launcher.ps1.
• Keeps the existing PowerShell interactive behavior.
• Supports providers: codex, gemini, claude.

macOS/Linux

• Uses scripts/agent_launcher_unix.js.
• Provides matching interactive flow in Node.js.
• Supports providers: codex, gemini, claude.

Provider behavior

Codex

• Model flag: -m <model>
• Adds reasoning config: -c reasoning_effort=<level> (Windows flow uses interactive reasoning selection)
• “Allow all permissions” mode adds:
  • --dangerously-bypass-approvals-and-sandbox

Gemini

• Model flag: -m <model>
• “Allow all permissions” mode adds:
  • -s false --approval-mode yolo

Claude

• Model flag: --model <alias-or-model>
• Supported aliases in launcher:
  • default, sonnet, opus, haiku, sonnet[1m], opusplan
• “Allow all permissions” mode adds:
  • --dangerously-skip-permissions

Auto Prompt Refinement (global install behavior)

On postinstall, this package configures native prompt hooks for the upstream CLIs so refinement happens even when users run plain codex, claude, or gemini directly (not only codexq/claudeq/geminiq).

Installed artifacts:

• ~/.claude/hooks/cgl-refine.js (+ ~/.claude/hooks/common.js)
• ~/.gemini/hooks/cgl-refine.js (+ ~/.gemini/hooks/common.js)
• ~/.claude/settings.json merge: adds/updates hooks.UserPromptSubmit.hooks[] entry named cgl-auto-refine
• ~/.gemini/settings.json merge: adds/updates hooks.BeforeAgent.hooks[] entry named cgl-auto-refine
• ~/AGENTS.md merge: adds/updates a marked Codex instructions block (CGL_AUTO_REFINE_START/CGL_AUTO_REFINE_END)

Behavior:

• The hook/snippet instructs agents to always output a short Refined Prompt/Refined Spec section first, then continue implementation in the same response.
• Settings merges are non-destructive: existing keys and other hooks are preserved.

Opt-out:

• Set CGL_DISABLE_REFINE=1 and rerun postinstall (npm install -g ...) to skip install/update of refinement config.
• If you previously installed refinement hooks/snippets, remove the cgl-auto-refine hook entries from ~/.claude/settings.json and ~/.gemini/settings.json, and remove the marked block from ~/AGENTS.md.

Manual verification:

1. Launch claude, submit any prompt, confirm the first section is Refined Prompt (or Refined Spec) before coding output.
2. Launch gemini, submit any prompt, confirm the first section is Refined Prompt (or Refined Spec) before coding output.
3. Launch codex in a directory under your home folder, submit any prompt, confirm the first section is Refined Prompt (or Refined Spec) before coding output.

Automatic updates

The launcher performs periodic update checks. If a newer version is detected, it attempts to run:

npm install -g @openai/codex @google/gemini-cli @anthropic-ai/claude-code https://codeload.github.com/aGamingGod1234/codex-gemini-quick-launch/tar.gz/refs/heads/main

Behavior:

• If update succeeds, launcher proceeds using updated installation.
• If update fails (network, permissions, npm issues), it prints a warning and continues with current installed version.

Filesystem differences handled

The project accounts for core platform filesystem differences:

• Separators: Windows commonly \, macOS/Linux /
• Case sensitivity:
  • Windows: typically case-insensitive
  • Linux: case-sensitive
  • macOS: commonly case-insensitive by default (can vary)
• User/app data locations:
  • Windows: %USERPROFILE%, %APPDATA%, %LOCALAPPDATA%
  • macOS: $HOME, ~/Library/Application Support
  • Linux: $HOME, often XDG dirs such as ~/.local/state
• Path roots:
  • Windows: drive letters (C:\)
  • macOS/Linux: single root (/)

Repository structure (important files)

• bin/codex-quick-launch.js – codex command wrapper
• bin/gemini-quick-launch.js – gemini command wrapper
• bin/claude-quick-launch.js – claude command wrapper
• scripts/agent_launcher.ps1 – Windows interactive launcher
• scripts/agent_launcher_unix.js – macOS/Linux interactive launcher
• scripts/update-check.js – update detection + auto-update logic
• scripts/postinstall.js – install metadata + Windows legacy shim cleanup

Notes

• Requires Node.js 18+.
• Quick launchers do not replace upstream CLIs; they orchestrate launch configuration and then call those CLIs.