Codex Migration Map

This page maps the existing Claude-oriented skills, agents, and commands in this repo to a Codex-native structure.

The goal is seamless operator behavior in Codex without changing any current Claude functionality.

Translation Model

Translate each Claude asset by role, not by filename:

• stable repo rules -> AGENTS.md
• reusable operating workflow -> Codex skill
• supporting runbook -> skill reference playbook
• delegated execution persona -> subagent pattern only if needed
• capability surface -> existing MCP server

Current Sources

There are two main Claude-oriented layers in this repo:

• project-level assets in .claude/
• plugin-level assets in packages/mcp-server/plugins/automaker/

Recommended Codex Mapping

Existing Asset Type	Current Location	Codex-Native Target	Notes
Repo rules	CLAUDE.md, .claude/settings.json	AGENTS.md	Keep stable instructions here
Ava operator workflow	plugins/.../commands/ava.md	.codex/skills/ava/SKILL.md	Main orchestration skill
Headsdown workflow	plugins/.../commands/headsdown.md	.codex/skills/headsdown/SKILL.md	Continuous deep work skill
Lightweight operational specialists	.claude/skills/board-janitor.md, .claude/skills/pr-maintainer.md	Ava playbooks first	Promote to full skills only if needed
Team role specialists	.claude/skills/matt.md, kai.md, frank.md, jon.md, cindi.md	future Codex skills or subagent conventions	Create only as demand appears
Analysis agents	.claude/agents/deepdive.md, deepcode.md, security-vulnerability-scanner.md	future subagent patterns	Better as delegation patterns than static prose clones
Plugin commands	plugins/.../commands/*.md	Codex skills selectively	Only port the workflows that matter
Capability tools	packages/mcp-server/	unchanged MCP server	Reuse directly from Codex

Immediate Migration Set

These are the highest-value Codex-native equivalents to maintain operator continuity:

• ava
• headsdown
• deep-research
• deep-dive
• due-diligence
• plan-project
• AGENTS.md
• MCP connection to the existing protoLabs server

Near-Term Candidates

These should likely come next if the Codex workflow sticks:

• setuplab
• role-specialist skills such as matt, kai, frank, jon, cindi

Team Skill Interpretation

The project-level Claude skills split into two groups:

Specialist Implementers

• matt
• kai
• frank
• jon
• cindi

These map well to future Codex skills if you want explicit role activation.

Operational Utilities

• pr-maintainer
• board-janitor

These may not need standalone skills immediately. Many of their behaviors can live as Ava playbooks until the workflow volume justifies separate Codex skills.

Agent Interpretation

The Claude agents are best treated as patterns for delegation rather than direct one-to-one ports.

Examples:

• deepdive -> Codex-native deep-dive skill for investigation
• security-vulnerability-scanner -> security review delegated subagent pattern
• feature-planner -> planning delegated subagent pattern
• plugin deep-research command -> Codex-native deep-research skill
• plugin due-diligence command -> Codex-native due-diligence skill

In Codex, this usually means:

• keep the orchestration logic in the main skill
• delegate bounded subtasks only when necessary
• avoid creating a static subagent for every historical Claude persona

Migration Principle

Do not clone every Claude command into Codex.

Port only the workflows that:

• are used frequently
• encode real operating discipline
• benefit from Codex-native repetition

Everything else should either stay in docs, become a small playbook, or wait until usage proves it deserves a full skill.

Related

• Codex CLI Integration
• Claude Code Plugin
• Plugin Deep Dive