@ljh0107-l/codex-pilot
v1.1.0
Published
An unofficial OpenAI Codex CLI fork with PromptPilot prompt enhancement.
Readme
Codex Pilot
An unofficial OpenAI Codex CLI fork with PromptPilot prompt enhancement.
This project is not affiliated with OpenAI.
Codex Pilot is based on OpenAI Codex CLI and intends to keep the core Codex CLI behavior intact.
The product change is a minimal prompt-enhancement layer in the TUI.
What is Codex Pilot?
Codex Pilot is an experimental fork of OpenAI Codex CLI.
The goal is to provide a lightweight PromptPilot experience in the Codex terminal UI:
- show how Codex is likely to understand your prompt
- rewrite rough instructions into clearer coding-agent prompts
- preview the enhanced prompt before execution
- reduce vague or overly broad agent runs
Codex Pilot is not a replacement for OpenAI Codex CLI. It is a small fork focused on prompt enhancement UX.
Installation
npm install -g @ljh0107-l/codex-pilotRun the CLI with:
codex-pilotPromptPilot features
Ctrl+Xprompt enhancement- interpreted intent preview
- enhanced prompt preview before execution
- apply/cancel flow that edits the composer draft without submitting
- language-aware output that keeps interpreted intent and enhanced prompts in the original prompt's primary language
- optional ACE context-aware enhancement that can use lightweight project context
Custom PromptPilot model
By default, PromptPilot uses the active Codex session model. To use a separate OpenAI-compatible chat completions model for prompt enhancement, add:
[prompt_pilot]
model = "your-optimizer-model"
base_url = "https://your-openai-compatible-endpoint/v1"
api_key_env = "YOUR_OPTIMIZER_API_KEY"
ace_default_enabled = false
ace_max_iterations = 5base_url defaults to https://api.openai.com/v1 when omitted. api_key_env defaults to
OPENAI_API_KEY; set it to an empty string for local proxies that do not require Authorization.
ace_default_enabled controls whether new TUI sessions start with ACE on. It defaults to false;
/pilot context on and /pilot context off still only affect the current session.
ace_max_iterations controls how many read-only context refinement passes ACE can use before it
must produce a prompt draft. It defaults to 5 and is clamped to 1-20.
ACE context
PromptPilot ACE is a context-aware prompt enhancement layer. It is off by default unless
ace_default_enabled = true is set, and command changes only affect the current TUI session.
/pilot context on
/pilot context offWhen ACE is on, Ctrl+X enhances the current composer draft with lightweight project context such
as project docs, AGENTS.md, manifests, git status, and relevant file signals. Apply only replaces
the composer draft; it does not submit the prompt. ACE does not modify files or run commands while
enhancing the draft.
What this fork does not change
Codex Pilot intends not to modify:
- authentication
- sandboxing
- tool execution
- file editing logic
- approval logic
- model provider logic
- core Codex agent behavior
PromptPilot changes should stay limited to the TUI prompt input and preview experience.
