# Claude To Openai Converter > Analyze Claude Code artifacts and convert them into the appropriate OpenAI target: Codex/OpenAI skills, AGENTS.md repository instructions, or API/Agents SDK tool-based agents. Use when a repo contains CLAUDE.md files, .claude/agents, .claude/skills, or .claude/commands and Codex should first inventory the source, produce a migration plan, identify non-equivalent features, and then implement a careful conversion. - Skill: `artsmc/claude-to-openai-converter` (Agent Skill, multi-file: 5 files) - Install (CLI): `npx skillmds@latest add artsmc/claude-to-openai-converter` - Raw SKILL.md: https://api.skillmd.com/api/skills/artsmc/claude-to-openai-converter/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Integrations & APIs - Author: artsmc (https://skillmd.com/u/artsmc) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/artsmc/claude-to-openai-converter --- # Claude To OpenAI Converter Convert Claude Code memory files, skills, slash commands, and subagents into the closest OpenAI/Codex form without doing a blind syntax rewrite. ## Workflow 1. Inventory the Claude source before proposing edits. Use `rg --files` to look for: - `CLAUDE.md` - `CLAUDE.local.md` - `.claude/agents/*.md` - `.claude/skills/**/SKILL.md` - `.claude/commands/*.md` - `.claude/settings.json` 2. Classify each artifact by behavior, not filename. - Project guidance and working conventions usually map to `AGENTS.md`. - Reusable prompt workflows usually map to a Codex/OpenAI skill. - Tool-gated specialist behavior may map to a skill, to repo instructions, or to an Agents SDK handoff design. - Claude-only features such as subagent memory, worktree isolation, background execution, or hooks may require approximation rather than direct conversion. 3. Produce a conversion plan before editing. Include: - Source inventory - Proposed target for each source file - Gaps or non-equivalent features - Files to create or modify - Validation steps 4. Convert incrementally. Preserve intent, constraints, and trigger conditions. Rewrite structure for the target platform instead of copying Claude frontmatter fields verbatim. 5. Validate the result. - For Codex skills, run the local skill validator. - For repo instructions, check that the guidance is concise, actionable, and placed where Codex will discover it. - For API-oriented output, ensure tools and handoffs are described as explicit tool-calling flows. ## Target Selection Default to the smallest target that preserves behavior. ### Convert to `AGENTS.md` Use `AGENTS.md` when the Claude source primarily describes: - repository conventions - architecture notes - test commands - review/checklist expectations - coding style or workflow constraints ### Convert to a Codex/OpenAI skill Use a skill when the Claude source is a reusable workflow with a clear trigger, such as: - a custom slash command - a Claude skill directory - a specialized migration or analysis playbook - a focused expert behavior that is mostly prompt-driven When building a skill, keep `SKILL.md` concise and move detail into `references/`, `scripts/`, or `assets/` only when needed. ### Convert to an API or Agents SDK design Use this path when the Claude source depends on runtime orchestration rather than static prompt instructions, such as: - explicit agent handoffs - multiple specialists with separate contexts - structured tool-calling requirements - application-managed state or tool outputs In that case, produce: - agent role definitions - tool schemas - handoff rules - a short implementation plan for Responses API or Agents SDK integration ## Mapping Rules Read [claude-formats.md](references/claude-formats.md) and [openai-targets.md](references/openai-targets.md) before converting unfamiliar constructs. Use this default mapping: - `CLAUDE.md` -> `AGENTS.md` - `.claude/commands/*.md` -> a Codex skill, unless it is just general project guidance - `.claude/skills/*/SKILL.md` -> a Codex skill with rewritten frontmatter and resource layout - `.claude/agents/*.md` -> either a Codex skill plus `AGENTS.md` guidance, or an Agents SDK design if true agent delegation is required Do not assume a Claude subagent always becomes an OpenAI skill. Separate-context delegation is often a runtime concern, not a static skill concern. ## Output Contract Before making edits, present a compact plan with these headings: - `Inventory` - `Target mapping` - `Conversion plan` - `Open questions` After the plan, implement the conversion unless a major target ambiguity makes that risky. When reporting non-equivalences, be explicit: - what Claude feature exists - whether OpenAI/Codex has a direct equivalent - what approximation you are using - what behavior is lost or moved into application code ## Implementation Notes - Prefer preserving semantics over preserving filenames. - Keep descriptions trigger-oriented. They are the main activation mechanism for skills and subagents. - If converting to a Codex skill, include `agents/openai.yaml` metadata that matches the skill. - If converting to `AGENTS.md`, compress guidance and remove Claude-specific UI references like `/memory` or `/agents`. - If converting to API tools, express Claude capabilities as explicit tool schemas and tool-call loops. ## References - Claude source formats: [claude-formats.md](references/claude-formats.md) - OpenAI/Codex targets: [openai-targets.md](references/openai-targets.md) - Conversion checklist: [conversion-checklist.md](references/conversion-checklist.md)