Codex Cli State Of Play

Codex CLI: State of Play — May 2026

Date: 2026-05-23

Where Codex CLI Stands Today

OpenAI’s Codex CLI has evolved from a simple terminal coding assistant into a full agentic platform. As of v0.133.0 (May 21, 2026), it includes:

  • Subagents for parallel, specialised work
  • Goal mode for persistent, long-running objectives
  • Plugins and marketplace for distributable workflows
  • MCP integration for connecting to external tools
  • Hooks for governance and audit logging
  • Python SDK (openai-codex) for programmatic access

The Agentic Stack

Codex’s agentic capabilities form a layered stack:

  1. Foundation: Rust-based CLI with bubblewrap sandboxing (Linux) and Docker devcontainer support
  2. Agent Layer: Built-in agents (default, worker, explorer) + custom TOML-defined agents
  3. Orchestration: Subagent spawning with configurable parallelism (max_threads) and depth limits
  4. Persistence: Goal mode for multi-session, long-running objectives
  5. Extension: Plugins bundling skills, apps, and MCP servers
  6. Integration: MCP client/server, Agents SDK compatibility, AGENTS.md cross-agent standard

Key Numbers

  • 150+ ecosystem tools catalogued in awesome-codex-cli
  • 136+ community subagents (VoltAgent collection)
  • 1,340+ community skills (antigravity-awesome-skills)
  • 20+ official launch plugins (Slack, Figma, Notion, Sentry)
  • 60,000+ projects using agents.md standard

Enterprise Angle

Thibault Sottiaux (OpenAI Head of Codex Product) positions Codex as “becoming the standard agent” for enterprise. Key enterprise features:

  • Permission profiles with inheritance
  • Hook-based audit logging and governance
  • Managed requirements.toml for reproducible environments
  • Plugin marketplace with enterprise controls
  • OAuth support for MCP servers

What’s Missing / Coming

  • Self-serve plugin publishing (announced “coming soon”)
  • Windows native support (expected late 2026)
  • PreToolUse hooks for non-Bash tools (current gap)
  • Deeper CI/CD integration patterns

For the Agentic Pod

The most relevant areas for Daniel’s agentic pod work:

  1. Subagent orchestration patterns — PR review pipelines, CSV batch processing
  2. Goal mode for long-running automation — migration tasks, documentation generation
  3. MCP server composition — chaining multiple MCP servers for complex workflows
  4. Cross-agent standards — agents.md for projects that use both Codex and Claude Code
  5. Codex as MCP server — letting other agents (Claude Code) call Codex agents