What Is oh-my-codex?
oh-my-codex is an open-source TypeScript harness that wraps the OpenAI Codex API in a structured multi-agent pipeline. Created by Yeachan Heo, the same developer who built oh-my-claudecode, it brings the Team Mode concept — staged plan, PRD, execute, verify, and fix phases — to the Codex environment.
The core problem oh-my-codex solves is unstructured Codex usage. Without a harness, developers prompt Codex ad hoc, skipping requirements clarity and verification steps. oh-my-codex enforces a process: no code is generated until a plan and PRD exist; no code is accepted until it passes a verification stage.
The project became GitHub Trending TypeScript #1 with more than 3,000 stars in 48 hours, signaling strong demand from developers who want structured, repeatable Codex workflows rather than one-off completions.
How to Calculate Better Results with oh-my-codex codex agent harness review
Clone or install via npm from github.com/Yeachan-Heo/oh-my-codex. Set your OPENAI_API_KEY (or compatible Codex endpoint) as an environment variable.
For a new task, run team-plan with your task description. Review the generated plan and refine it before proceeding. This step surfaces scope and complexity early.
Run team-prd to generate a requirements document from the plan. The PRD defines acceptance criteria that the verification stage will check against.
Run team-exec to implement the code within the plan and PRD boundaries. Run team-verify to check the output. If issues are found, run team-fix to address them. For straightforward tasks, use autopilot to run the full cycle in one command.
Treat this page as a decision map. Build a shortlist fast, then run a focused second pass for security, ownership, and operational fit.
When a team keeps one shared selection rubric, tool adoption speeds up because evaluators stop debating criteria every time a new option appears.
Worked Examples
Add pagination to a REST API endpoint
- Run: team-plan "add cursor-based pagination to /api/posts endpoint"
- Review plan output: identifies 3 files to modify, defines cursor schema, flags backward-compat concern
- Run team-prd: generates acceptance criteria including cursor format, edge cases, and test expectations
- Run team-exec: implements pagination logic within plan boundaries
- Run team-verify: checks cursor behavior, edge cases, and test coverage against PRD
- team-verify flags one edge case (empty result set). Run team-fix with the specific finding
Outcome: Paginated endpoint with verified edge case handling, traceable back to the original PRD. Total cycle: ~12 minutes vs. ~35 minutes of back-and-forth prompting.
Refactor authentication module to support OAuth
- Run: deep-interview "refactor auth to support OAuth alongside existing session auth"
- Deep Interview surfaces 4 clarifying questions: which providers, session migration strategy, token storage approach, rollback plan
- Answer questions → run team-plan with enriched context
- PRD defines two auth paths, migration schedule, and specific test scenarios for each provider
- team-exec implements OAuth integration without breaking existing session auth
- team-verify confirms both auth paths work per PRD acceptance criteria
Outcome: OAuth integration with zero regression in existing session auth. Deep Interview prevented three architectural decisions that would have required rework.
Frequently Asked Questions
What is oh-my-codex?
oh-my-codex is a TypeScript agent harness for OpenAI Codex built by Yeachan Heo — the same developer behind oh-my-claudecode. It applies the same Team Mode pipeline concept (plan → PRD → execute → verify → fix) to the Codex environment, giving Codex users a structured, multi-agent development loop instead of ad-hoc prompting. The project hit GitHub Trending TypeScript #1 with over 3,000 stars in its first 48 hours.
How does oh-my-codex differ from oh-my-claudecode?
oh-my-codex targets OpenAI Codex (and codex-compatible endpoints), while oh-my-claudecode targets Claude Code by Anthropic. Both share the same core architecture: Team Mode staged pipelines, Deep Interview for requirements mining, and Autopilot for autonomous execution. The primary difference is the underlying model runtime and API surface. If you use both Claude Code and Codex, you can use each tool in its respective environment while maintaining a consistent development process.
What is the Team Mode pipeline in oh-my-codex?
Team Mode runs a five-stage pipeline: (1) team-plan analyzes the task and creates an implementation strategy, (2) team-prd generates a product requirements document with acceptance criteria, (3) team-exec implements the code within the boundaries defined by the plan and PRD, (4) team-verify validates the output against acceptance criteria, (5) team-fix addresses specific issues found during verification. Each stage produces artifacts that feed the next, creating a structured development cycle.
What is the Deep Interview feature in oh-my-codex?
Deep Interview is a Socratic questioning phase that runs before any code generation begins. It surfaces ambiguity in requirements, constraints, and edge cases through structured follow-up questions. The interview output feeds directly into team-plan and team-prd, reducing rework caused by misunderstood requirements. It is especially effective for complex features where the initial prompt undersells the real scope.
How do I install oh-my-codex?
oh-my-codex is available on GitHub at github.com/Yeachan-Heo/oh-my-codex. Install via npm or clone the repository directly. It requires a Codex-compatible API key (OpenAI or compatible endpoint). Refer to the repository README for environment variable setup and CLI invocation. The tool follows the same installation pattern as oh-my-claudecode — npm install, configure API key, then run team commands.
Is oh-my-codex production-ready?
oh-my-codex is a rapidly evolving open-source project. Given its 3,000+ stars in under 48 hours, it has significant community traction, but as with any early-stage tool, expect active development and potential breaking changes between versions. Pin your version in production workflows and monitor the GitHub releases page for updates.
What kinds of projects benefit most from oh-my-codex?
oh-my-codex is best suited for mid-to-large coding tasks where unstructured prompting tends to produce inconsistent or incomplete results: API integrations, refactoring projects, feature additions to existing codebases, and multi-file implementations. For simple one-file scripts or quick prototypes, the overhead of the full pipeline may not be worth it — use Autopilot mode instead for a lighter-weight entry point.