用于将 Claude Code 技能和代理迁移到 Codex 生态系统。它将可重用的工作流程转换为 Agent Skill 包， 可重用角色转换为自定义代理 TOML 文件，规范化 MCP 工具命名空间 Codex，并通过本地Codex子代理保持委派行为 适当时进行配器。

⚠ 实验性——接口、工件布局和验证规则可能 临时更换。

这是什么

该仓库是一个迁移系统：

Claude 源件存在于 .claude/skills/ 和 .claude/agents/。 迁移工作流程作为Codex技能包存储在本仓库中， Python 脚本。生成的输出落在 .codex/skills/ 和 .codex/agents/，并带有验证和跟踪状态以保持迁移 可审计。

关键的架构决策：移民并非盲式音译

目标是先保持工作流程意图，然后用以下方式表达该行为 原生Codex表面。实际上，这意味着SKILL.md包为 工作流，.codex/agents/*.toml 可重用角色的定制代理， 下划线归一化的MCP工具命名空间，以及spawn_agent(...)风格

ccode-to-codex

Experimental toolkit for migrating Claude Code skills and agents into the Codex ecosystem. It converts reusable workflows into Agent Skill packages, reusable roles into custom-agent TOML files, normalizes MCP tool namespaces for Codex, and preserves delegation behavior through native Codex subagent orchestration where appropriate.

⚠ Experimental — interfaces, artifact layouts, and validation rules may change without notice.

What it is

This repository is a migration system:

Claude source artifacts -> migration workflows -> Codex-native outputs

Claude source artifacts live under .claude/skills/ and .claude/agents/. The migration workflows live in this repository as Codex skill packages and Python scripts. The generated outputs land under .codex/skills/ and .codex/agents/, with validation and tracker state to keep the migration auditable.

Key architectural decision: migration is not blind transliteration

The goal is to preserve workflow intent first, then express that behavior using native Codex surfaces. In practice, this means SKILL.md packages for workflows, .codex/agents/*.toml custom agents for reusable roles, underscore-normalized MCP tool namespaces, and spawn_agent(...)-style orchestration only for source workflows that genuinely delegate work.

Current capabilities

• Migrate Claude skills into Codex Agent Skill packages.
• Migrate Claude agents into Codex custom-agent TOML outputs.
• Classify blockers, dependency gaps, and manual-review cases before writes.
• Normalize MCP namespaces from source tool references into Codex-callable form.
• Preview or dry-run migrations before touching live .codex/ outputs.
• Validate migrated artifacts, refresh migration-state.json, and inspect progress with the dashboard.
• Provide minimal reference examples for both Agent Skills and custom agents.

What is actually required today

• Python 3.11+
• PyYAML>=6.0.2,<7 (installed via requirements.txt)
• Claude source inputs under .claude/skills/ and/or .claude/agents/
• OpenAI Codex CLI when consuming or exercising migrated .codex/ outputs

To invoke these migration workflows as Codex skills in a target repository, install the operational packages under that repository’s .codex/skills/ directory. The .claude/skills/ and .claude/agents/ directories are source inputs only; the migration reads them and writes Codex-facing outputs under .codex/.

mkdir -p /path/to/target-repo/.codex/skills
cp -R .codex/skills/migrate-to-codex /path/to/target-repo/.codex/skills/
cp -R .codex/skills/migrate-agents-to-codex /path/to/target-repo/.codex/skills/
cp -R .codex/skills/verify-skill-migration /path/to/target-repo/.codex/skills/
cp -R .codex/skills/migration-dashboard /path/to/target-repo/.codex/skills/