Codex Handoff for Openclaw

Codex Handoff transforms Claude into a technical supervisor that orchestrates the Codex CLI to execute complex coding plans with automated review loops.

philipbankier
v1.1.1
Mar 21, 2026
0
713
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install codex-handoff-skill

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install codex-handoff-skill using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).

Prefer to download?

Get the raw skill files in a ZIP archive.

What is Codex Handoff?

Codex Handoff is a sophisticated supervisor loop designed for developer productivity. It establishes a clear separation of concerns where Claude Code acts as the supervisor and judge, while the Codex CLI serves as the primary executor. This architecture allows developers to offload finalized coding plans to an automated system that handles the heavy lifting of implementation while maintaining high quality through rigorous oversight.

By leveraging this specific entry in the Openclaw Skills library, users can automate the transition from planning to execution. The skill manages the entire lifecycle of a coding task—from locating plan files in your repository to performing multi-pass iterations and phased executions—ensuring that the final output aligns perfectly with the initial technical specifications.

Codex Handoff Use Cases

  • Offloading finalized architectural plans to an automated coding agent for rapid implementation.
  • Managing complex, multi-phase refactors where a supervisor is needed to audit individual stages.
  • Executing routine coding tasks in a headless, full-auto mode while maintaining a git-based audit trail.
  • Implementing features across large codebases where a review-and-correct loop is necessary for accuracy.

How Codex Handoff Works

  1. The skill locates a valid plan file within the project directory, specifically searching in docs/plans/ or .claude/plans/.
  2. It analyzes the plan to detect distinct phases or stages, determining if a sequential execution strategy is required.
  3. A structured Codex prompt is dynamically generated, incorporating the plan details, project context, and coding standards.
  4. The Codex CLI is triggered in full-auto mode to execute the coding tasks identified in the current phase.
  5. The supervisor performs a technical review of the git diff and runs test suites to audit the completion against a scorecard.
  6. If tasks remain incomplete, the skill initiates a correction loop; otherwise, it advances to the next phase or finalizes the task.

Codex Handoff Setup

To utilize this skill within the Openclaw Skills ecosystem, ensure you have the Codex CLI installed and your project is initialized as a git repository.

# Verify Codex CLI installation
codex --version

# Ensure a plan exists in one of the expected directories
mkdir -p docs/plans/
touch docs/plans/2024-01-01-feature-plan.md

# Run the handoff command via your AI agent
/codex-handoff

Codex Handoff Data Schema & Taxonomy

Codex Handoff organizes its operations based on the following file structure and metadata:

Data Type Location / Format Description
Plans docs/plans/*.md or .claude/plans/ Markdown files containing H2 phase headings and task lists.
Prompts /tmp/codex-handoff-{ts}.md Temporary structured instructions sent to the Codex executor.
Review Docs references/review-process.md Contains the scorecard and decision matrix for task auditing.
Standards references/prompt-templates.md Templates for initial execution and phase-scoped prompts.

Codex Handoff Advanced Features

  • Phased Execution Flow: Automatically detects and sequentially processes complex plans broken down into numbered stages.
  • Automated Correction Loops: Supports a configurable max-iterations parameter (defaulting to 5) to refine code until it passes all checks.
  • Strict Supervisor Mode: Ensures the AI agent never modifies code directly, preserving the integrity of the supervisor-executor relationship.
  • Multi-Model Support: Allows passing specific model flags to the underlying Codex CLI via the --model argument.
  • Full-Auto Execution: Leverages the workspace-write permission for hands-free implementation of entire feature sets.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*