The Codex CLI Ecosystem Map: Navigating 245+ Community Tools, Skills and Subagents

ecosystemcommunityskillssubagentspluginsmcpawesome-listcross-platformtoolingagent-skills

Sketchnote diagram for: The Codex CLI Ecosystem Map: Navigating 245+ Community Tools, Skills and Subagents

The Codex CLI Ecosystem Map: Navigating 245+ Community Tools, Skills and Subagents

Why This Matters

Twelve months ago, Codex CLI was a single binary with a handful of configuration options. Today, a curated list on the official OpenAI Codex GitHub Discussions page catalogues over 245 community-built tools across 20 categories — subagents, skills, plugins, MCP servers, IDE integrations, GUI wrappers, session managers and more1. The ecosystem has exploded, but discoverability remains the bottleneck. This article maps the landscape, highlights the most impactful tools in each category, and provides a decision framework for which layer to invest in first.

The Three Discovery Surfaces

Before diving into individual tools, it helps to understand where Codex CLI extensions live and how they are discovered.

graph TD
    A[Extension Discovery] --> B[Built-in]
    A --> C[Community Repos]
    A --> D[Cross-Platform Directories]

    B --> B1["openai/skills<br/>Official Skills Catalog"]
    B --> B2["/skills, /plugins<br/>TUI Commands"]
    B --> B3["developers.openai.com/codex/skills<br/>Docs"]

    C --> C1["RoggeOhta/awesome-codex-cli<br/>245+ tools"]
    C --> C2["VoltAgent/awesome-codex-subagents<br/>136+ agents"]
    C --> C3["ComposioHQ/awesome-codex-skills<br/>38 skills"]

    D --> D1["awesomeskills.dev<br/>Cross-Platform Atlas"]
    D --> D2["VoltAgent/awesome-agent-skills<br/>1,000+ skills"]
    D --> D3["skills.sh<br/>Vercel Package Manager"]

Built-in discovery uses the /skills and /plugins slash commands within the TUI, backed by OpenAI’s official openai/skills GitHub repository2. This is the curated, first-party surface — reliable but deliberately conservative.

Community repositories aggregate everything the official catalog does not include. The canonical hub is RoggeOhta’s awesome-codex-cli list, pinned in GitHub Discussion #16329 on the openai/codex repository1. It is actively maintained, with contributions from dozens of authors.

Cross-platform directories serve skills that work across Codex CLI, Claude Code, Gemini CLI, Cursor, GitHub Copilot, and other tools adopting the SKILL.md standard. The largest is VoltAgent’s awesome-agent-skills with over 1,000 entries3, alongside Vercel’s skills.sh CLI package manager which has accumulated over 26,000 installs since its January 2026 launch4.

Subagents: 136+ Specialist Workers

Subagents are TOML-defined worker agents that Codex spawns within a session. The community has produced an extraordinary number of pre-built subagent definitions.

VoltAgent/awesome-codex-subagents

The largest single collection, with 136+ agents organised into ten categories5:

Category Examples Use Case
Core Dev code-reviewer, test-writer, refactorer Daily coding workflows
Language Specialists rust-expert, go-expert, python-expert Language-idiomatic patterns
Infrastructure terraform-agent, k8s-deployer, docker-builder IaC and DevOps
Security vuln-scanner, secret-detector, compliance-checker Security auditing
Data/AI data-pipeline, ml-trainer, notebook-helper Data engineering and ML
DX docs-writer, changelog-generator, onboarding-guide Developer experience
Domains fintech-agent, healthcare-agent, edtech-agent Industry-specific patterns
Business proposal-writer, metrics-dashboard Non-coding business tasks
Meta agent-debugger, prompt-optimizer Agent self-improvement
Orchestration dispatcher, wave-controller, fan-out-manager Multi-agent coordination

To use any of these, drop the .toml file into .codex/agents/ in your project root and reference it in your session:

# .codex/agents/security-scanner.toml
[agent]
name = "security-scanner"
model = "gpt-5.4-mini"
instructions = "You are a security specialist. Scan for OWASP Top 10 vulnerabilities..."

[agent.sandbox]
mode = "read-only"

Enterprise and Specialised Packs

Beyond the VoltAgent mega-collection, several targeted packs stand out:

  • betterup/codex-cli-subagents — enterprise-focused agents for code review, migration and compliance auditing1
  • leonardsellem/codex-specialized-subagents — accessibility, internationalisation and performance profiling agents1
  • waltstephen/ArgusBot — a 24/7 supervisor agent for autonomous task completion with watchdog monitoring1
  • basilisk-labs/codex-swarm — swarm intelligence patterns for large-scale refactoring operations1

Multi-Agent Orchestrators

The orchestration layer above subagents has also matured:

  • ComposioHQ/agent-orchestrator — plans tasks, spawns agents, and handles CI failures autonomously1
  • mco-org/mco — a neutral orchestration layer supporting Codex CLI, Claude Code, Gemini CLI, OpenCode and Qwen Code simultaneously1
  • hcom (aannoo) — hierarchical communication framework with context preservation across agent boundaries1

Skills: From 38 to 1,370+

Skills are the most portable extension mechanism in Codex CLI — a SKILL.md file with YAML frontmatter and natural-language instructions2. Unlike subagents (which are Codex-specific TOML), skills follow the open Agent Skills standard adopted by over 25 tools under the Linux Foundation’s Agentic AI Foundation6.

Key Skill Collections

Collection Scale Differentiator
openai/skills Official catalog Curated, first-party quality
ComposioHQ/awesome-codex-skills 38 skills SaaS integration via Composio (Slack, GitHub, Notion, 1,000+ apps)7
VoltAgent/awesome-agent-skills 1,000+ skills Largest cross-platform directory3
sickn33/antigravity-awesome-skills 1,370+ skills Includes CLI installer and bundle system1
huggingface/upskill ML/AI focus Official Hugging Face workflow pack1
affaan-m/everything-claude-code 147 skills + 36 agents Performance-optimised system1

Noteworthy Specialist Skills

Several specialist skills deserve attention for their depth:

  • Understand-Anything (Lum1104) — converts entire codebases into interactive knowledge graphs, useful for onboarding to unfamiliar projects1
  • ARIS (wanshuiyin) — autonomous ML research that runs while you sleep, with cross-model review built in1
  • bug-hunter (codexstar69) — adversarial AI security scanner with automatic fix generation via multi-agent pipeline1
  • ctf-skills (ljagiello) — capture-the-flag challenge skills covering web exploitation, binary analysis, cryptography, reverse engineering and OSINT1
  • Skywork-Skills (SkyworkAI) — AI office suite covering presentations, documents, spreadsheets, image generation and music1

Installing Skills

Skills can be installed globally or per-project:

# Global installation (available in all sessions)
mkdir -p ~/.codex/skills/my-skill
cp SKILL.md ~/.codex/skills/my-skill/

# Project-local installation (committed to repo)
mkdir -p .agents/skills/my-skill
cp SKILL.md .agents/skills/my-skill/

# Via Vercel's skills.sh package manager
npx skills add composiohq/awesome-codex-skills

Plugins: Bundling Skills, MCP and Apps

Plugins are the highest-level extension unit — a bundle of skills, MCP server configurations and app connectors packaged for distribution8. The plugin system became first-class in v0.117.0 (March 2026), and the ecosystem is still young compared to skills and subagents.

Key plugin resources:

  • codex-plugin-cc — cross-model code review delegation between Codex and Claude Code1
  • claude-review-loop — automated Stop Hook plugin that triggers cross-model review after every Codex session1
  • $plugin-creator — the built-in scaffolding skill for generating new plugin directory structures8

MCP Servers: Both Client and Server

Codex CLI acts as both an MCP client (consuming external tool servers) and an MCP server (exposing its agent capabilities to other tools)9. The community has built MCP servers for virtually every major developer service.

graph LR
    subgraph "Codex as MCP Client"
        C[Codex CLI] --> M1[Figma MCP]
        C --> M2[Jira/Atlassian MCP]
        C --> M3[Datadog MCP]
        C --> M4[GitHub MCP]
        C --> M5[Notion MCP]
        C --> M6[Memory MCP Servers]
    end

    subgraph "Codex as MCP Server"
        O1[Claude Code] --> S[codex mcp-server]
        O2[Agents SDK] --> S
        O3[Custom Harness] --> S
    end

The MCP server landscape is extensively documented in the official docs9, but community contributions worth noting include:

  • AgentMemory — 41 MCP tools with triple-stream retrieval (BM25 + vector + knowledge graph)1
  • codebase-memory-mcp — Tree-Sitter AST-based code understanding across 66 languages1
  • codex-mcp-code-review (Szpadel) — clean-context review subagent pattern via MCP1

Cross-Platform Bridges

One of the most active ecosystem niches is tools that bridge between Codex CLI and competing agents:

  • shinpr/sub-agents-skills — route tasks to Codex, Claude Code, Cursor or Gemini from any compatible tool10
  • mco — neutral orchestration layer across five CLI agents1
  • CliGate — local gateway proxy unifying authentication and billing across Codex CLI, Claude Code and Gemini CLI1

The cross-platform story is strengthened by the AGENTS.md open standard, now adopted by over 60,000 projects under the Agentic AI Foundation (AAIF)6. A single AGENTS.md file works across Codex CLI, Claude Code (via symlink workaround), Gemini CLI, Cursor, Copilot, Amp and dozens more.

IDE and GUI Integrations

Beyond the terminal, the community has built graphical interfaces:

  • Nimbalyst — macOS GUI for managing parallel Codex sessions with git worktree isolation1
  • CC Pocket — iOS app for remote session management via WebSocket bridge1
  • cmux — Ghostty-based terminal with agent notification system and PR-aware tabs1
  • codex-replay — HTML visualiser for JSONL session rollout files1

The Decision Framework: Where to Invest First

With 245+ tools available, the natural question is: where do you start? Here is a decision framework based on team maturity with Codex CLI.

flowchart TD
    START[New to Codex CLI?] -->|Yes| L1[Level 1: AGENTS.md + Official Skills]
    START -->|No| Q1{Pain point?}

    Q1 -->|Repetitive tasks| L2[Level 2: Custom Skills]
    Q1 -->|Need external data| L3[Level 3: MCP Servers]
    Q1 -->|Complex workflows| L4[Level 4: Subagents]
    Q1 -->|Cross-tool teams| L5[Level 5: Plugins + Bridges]

    L1 -->|"Start here"| A1["Write AGENTS.md<br/>Browse openai/skills<br/>Use /skills in TUI"]
    L2 --> A2["Write SKILL.md files<br/>Install via skills.sh<br/>Browse awesomeskills.dev"]
    L3 --> A3["Configure config.toml<br/>[mcp_servers] section<br/>Add Figma, Jira, Datadog"]
    L4 --> A4["Create .codex/agents/*.toml<br/>Browse VoltAgent collection<br/>Start with 2-3 agents"]
    L5 --> A5["Evaluate CliGate, mco<br/>Build plugin bundles<br/>Use AGENTS.md as shared base"]

Level 1: AGENTS.md and official skills

If you are just getting started, write an AGENTS.md file for your project and browse the official openai/skills catalog. This costs nothing and delivers immediate productivity gains.

Level 2: Custom skills

When you find yourself repeating the same instructions across sessions, extract them into a SKILL.md. Check awesomeskills.dev and skills.sh before writing from scratch — someone has likely solved your problem already4.

Level 3: MCP servers

When your agent needs external data (Jira tickets, Figma designs, Datadog metrics, documentation), add MCP servers to your config.toml. Each server adds roughly 3–4K tokens of schema overhead to your context window9, so be selective.

Level 4: Subagents

For complex workflows requiring parallel execution or specialist knowledge, define subagent TOML files. The VoltAgent collection provides battle-tested starting points5. Start with two or three agents before scaling to larger swarms.

Level 5: Plugins and cross-tool bridges

For teams running multiple AI coding tools (Codex + Claude Code is the most common pairing), invest in plugins and bridges that unify configuration and enable cross-model review loops.

Ecosystem Health Metrics

The Codex CLI ecosystem shows strong vital signs as of April 2026:

  • 245+ tools catalogued across 20 categories1
  • 1,370+ cross-platform skills available via community installers1
  • 136+ pre-built subagents in the largest single collection5
  • 3M+ weekly active Codex CLI users as of 8 April 202611
  • 150+ members of the Agentic AI Foundation backing the AGENTS.md standard6

The ecosystem’s growth rate — from a handful of tools in Q4 2025 to 245+ in Q1 2026 — mirrors the trajectory that npm saw in its early years. The key difference is that agent skills are inherently more portable: a well-written SKILL.md works across Codex, Claude Code, Gemini CLI and a dozen other tools without modification6.

What to Watch

Three trends are shaping the ecosystem’s near-term direction:

  1. Consolidation around skills.sh — Vercel’s package manager is becoming the de facto distribution channel, and its install counts increasingly influence which skills gain traction4.

  2. Cross-model orchestration — Tools like mco and CliGate that abstract across multiple AI coding agents are gaining adoption as teams realise no single tool wins at everything1.

  3. Enterprise plugin registries — The plugin system’s marketplace.json format supports private, organisation-scoped registries8. Enterprise teams are beginning to build internal plugin directories that bundle approved skills, MCP configurations and subagent definitions for their specific technology stacks.

Citations

  1. RoggeOhta, “Awesome Codex CLI — curated list of 150+ ecosystem tools”, GitHub Discussion #16329, openai/codex, April 2026. https://github.com/openai/codex/discussions/16329 ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 ↩10 ↩11 ↩12 ↩13 ↩14 ↩15 ↩16 ↩17 ↩18 ↩19 ↩20 ↩21 ↩22 ↩23 ↩24 ↩25 ↩26 ↩27 ↩28 ↩29 ↩30 ↩31

  2. OpenAI, “Agent Skills – Codex”, developers.openai.com, 2026. https://developers.openai.com/codex/skills ↩ ↩2

  3. VoltAgent, “awesome-agent-skills: 1,000+ agent skills from official dev teams and the community”, GitHub, 2026. https://github.com/VoltAgent/awesome-agent-skills ↩ ↩2

  4. awesomeskills.dev, “Awesome Agent Skills — Claude Code, Codex & Cursor Skill Directory”, 2026. https://www.awesomeskills.dev/en ↩ ↩2 ↩3

  5. VoltAgent, “awesome-codex-subagents: 130+ specialized Codex subagents”, GitHub, 2026. https://github.com/VoltAgent/awesome-codex-subagents ↩ ↩2 ↩3

  6. Agentic AI Foundation (AAIF), “AGENTS.md Open Standard”, Linux Foundation, 2025-2026. https://github.com/agentsmd/agents.md ↩ ↩2 ↩3 ↩4

  7. ComposioHQ, “awesome-codex-skills: practical Codex skills for automating workflows”, GitHub, 2026. https://github.com/ComposioHQ/awesome-codex-skills ↩

  8. OpenAI, “Plugins – Codex CLI”, developers.openai.com, 2026. https://developers.openai.com/codex/plugins ↩ ↩2 ↩3

  9. OpenAI, “Model Context Protocol – Codex”, developers.openai.com, 2026. https://developers.openai.com/codex/mcp ↩ ↩2 ↩3

  10. shinpr, “sub-agents-skills: Cross-LLM sub-agent orchestration as Agent Skills”, GitHub, 2026. https://github.com/shinpr/sub-agents-skills ↩

  11. Sam Altman, “Codex crosses 3M weekly active users”, OpenAI announcement, 8 April 2026. https://openai.com/index/introducing-upgrades-to-codex/ ↩