Awesome Codex Plugins

Awesome Codex & ChatGPT Plugins

A curated list of awesome plugins, skills, and resources for OpenAI Codex and ChatGPT.

OpenAI documents plugins and skills for Codex, packaging skills, MCP servers, and app integrations into shareable, installable bundles across the Codex app, CLI, and IDE extensions.

Codex is now available inside the ChatGPT desktop app, while remaining a distinct coding workspace. OpenAI plugins can support workflows in ChatGPT, Codex, or both; this repository remains a Codex-compatible marketplace.

Contents

• Start Here
• Official Plugins
• Community Plugins
• Plugin Development
• Guides & Articles
• Related Projects
• Claim Your Plugin
• Plugin Trust Scores
• Plugin Quality
• Contributing

Start Here

New plugin workflow:

1. Create with $plugin-creator
2. Validate with plugin-scanner. The catalog-owned centralized scan must report a normalized score ≥ 75/100 before merge.
3. Optionally gate your source repo with the HOL scanner GitHub Action. Catalog admission uses the centralized scan; source-repo scanner CI is recommended but not required.
4. Ship or submit with confidence

Quick preflight

pipx run plugin-scanner lint .
pipx run plugin-scanner verify .

Scanner Requirements (Mandatory for This List)

All plugins submitted to Awesome Codex Plugins must pass the HOL AI Plugin Scanner:

See the full guide: SCANNER_GUIDE.md
See contributing requirements: CONTRIBUTING.md

This repo publishes a Codex repo marketplace at .agents/plugins/marketplace.json. The marketplace points at mirrored installable plugin bundles under ./plugins/, so a clone of this repo can act as a curated plugin source in Codex.

Use this marketplace in Codex

Install plugins directly from this curated list by pointing Codex at the repo marketplace:

CLI:

# Add this repo as a marketplace source (one-time setup)
codex plugin marketplace add \
  'https://github.com/hashgraph-online/awesome-codex-plugins.git' \
  --ref 'main' \
  --sparse '.agents/plugins' \
  --sparse 'plugins'

# Then browse and install (the marketplace name is derived from the repo name)
codex plugin list --source awesome-codex-plugins
codex plugin install <plugin-name> --source awesome-codex-plugins

Do not use the raw marketplace.json URL with codex plugin marketplace add. The Codex marketplace command clones a Git repository, so a raw GitHub file URL is treated like a repo URL and fails with remote: 404: Not Found.

Desktop App / IDE Extension:

1. Open Codex settings → Plugins → Next to search plugins input click on menu and select → +Add More...

   

2. Add this URL:

   https://github.com/hashgraph-online/awesome-codex-plugins.git
   

   

3. The curated plugin list appears as an available marketplace source.

Each plugin entry includes a source.path pointing at a mirrored bundle under ./plugins/ in this repo, so installations are fast and reproducible without hitting upstream repos at install time.

Official Plugins

Community Plugins

Third-party plugins built by the community. PRs welcome!

Development & Workflow

• A Team - A complete engineering team in one folder with 26 specialist agents, 20 workflow skills, an orchestrator, and quality gates; integration and hook support vary by coding platform.
• Aegis - An agentic skills framework & software development methodology that works: planning, TDD, debugging, and collaboration workflows.
• Agency Continuity Audit - Read-only audit that distinguishes durable agent goals, state, corrections, restart evidence, authority boundaries, scheduler claims, and commercial proof from self-reported health.
• Agent Deck - One MCP for context management: bind self-improving playbooks, MCP tools, and API keys to the session.
• Agent Guard - Local-by-default, no-telemetry hooks block or mask secret leaks in Claude Code and Codex, with opt-in PII filtering and Git/CI backstops.
• Agent Harness Skills - Designs agent-ready repository harnesses with entrypoints, validation surfaces, runtime evidence, delivery records, and atomic commit guidance.
• Agent Retro - Weekly retro of how you use Codex and Claude Code: 12 animated slides of usage stats with four evidence-backed coaching tips, built locally from your own logs.
• Agent Workflow System - 一套中文AI工作流系统：7个协作技能 + 行为规范宪法 + 会话恢复机制，模糊目标→可执行任务，全生命周期引导。Codex & Claude Code 双平台，新手友好。
• Agentic Readiness Assessment - Assess repository readiness for AI coding agents with an evidence-based scorecard and prioritized fixes.
• Agentizer - Turn any website into an AI-powered agentfront with split-pane
• AgentOps - DevOps layer for coding agents with flow, feedback, and memory that compounds between sessions.
• AgentPack - Ranks repo context for Codex with likely files, skill recommendations, agent rules, commands, warnings, and compact task-focused packs before editing.
• Agentry Observability - Agent-native product analytics, error logging, and deploy attribution for coding agents through one HTTP API.
• AgiFlow - Project management workflows for AI coding agents with planning, grooming, task execution, review, and AgiFlow MCP integration.
• AI-Native SDLC Loop - Six-stage SDLC in committed Markdown where hooks enforce the loop at the tool-call level: an artifact gate refuses source writes no approved plan covers and keeps a change declared low-risk away from sensitive paths, with 8 skills and 2 read-only subagents.
• AIBoarding - Generate, maintain, compress, and audit standard AI-agent onboarding files with AGENTS.md, CLAUDE.md, drift tracking, and lifecycle hooks.
• Alcove - Local-first MCP server for private project docs with hybrid BM25+vector search, tree-sitter code indexing, and automated linting for team-wide documentation standards.
• Anchor - Engineering discipline pack for Claude Code & Codex CLI with task-scope locking, anti-drift braking, condition-based codex review, project-CLAUDE.md pitfall writeback, and PreToolUse hooks that block irreversible bash patterns.
• Andon - Stop hooks that stop a coding agent from claiming "done" without proof, using a same-session verification log, warn or block mode, and a public defect ledger.
• Antigravity Workspace Template - Multi-agent codebase knowledge graph generator with context-aware planning and automatic scope management — turns codebases into coherent agent workspaces.
• Archcore - Spec-driven development and context engineering for Claude Code, Cursor, Codex, and GitHub Copilot — backed by project context in Git.
• ArmorCodex - Intent-based security for Codex with MCP plan registration, policy gating, CSRG cryptographic proofs, and audit logging on bash and apply_patch.
• assay - Opens a web page you changed, drives every control on it, and reports what it did and what happened in one line, clean or broken.
• AtomLane - Parallelizes only work proven safe by compiling resource-aware atomic plans for builds, tests, Make/Compose workflows, and long-running jobs with live progress and measured savings.
• Auto Pilot - Intake-first Codex plugin that turns a short project request into a structured, resumable execution workflow with one-question-at-a-time intake, project contract files, blocker triage, and resume support.
• AW-1 Circuit Breaker - Application-layer AST and argument policy gate for agent tool calls that checks configured rules before Python execution; complements, but does not replace, operating-system sandbox isolation.
• BABOK Analyst - BABOK v3 business analysis agent with 16 MCP tools, a 9-stage pipeline, and human-in-the-loop approval gates.
• BGS Modding Superpowers - Agentic Bethesda Game Studio modpack curation toolkit with MCP-driven xEdit conflict audit, MO2 control plane, BA2/BSA and Papyrus tooling, and skills for setup, dev-log, and release-changelog workflows.
• Bindry - Compiles public or workspace Binders into version-pinned agent skills, checks instruction drift, and supports optional MCP-based live mode.
• Boss - BMAD pipeline plugin that orchestrates a full requirements-to-deploy workflow across nine specialist agents with an auditable runtime DAG and quality gates, for Claude Code, Codex, OpenClaw, and Antigravity.
• Bring Your AI Migration Auditor - Read-only audit of AGENTS.md/CLAUDE.md scope, hooks, MCP config, skills, secret references, and validation notes before Codex edits migrated projects.
• Brooks Lint - AI code reviews grounded in six classic engineering books — decay risk diagnostics with book citations, severity labels, and four analysis modes (PR review, architecture audit, tech debt, test quality).
• Casefile - Persistent security case tracking for bug bounties, CTFs, and security audits.
• Changelog Forge - Conventional commits → CHANGELOG section + semver bump.
• ChatGPT Agent - Use ChatGPT as a reviewer, advisor, editor, or implementer in Codex, with ZIP packets by default and an opt-in MCP bridge.
• Claude Code Codex Plugin - Security-focused Codex plugin that connects to the local Claude Code CLI through MCP with login, status checks, safe-mode prompts, and no credential storage.
• Claude Code for Codex - Reverse of OpenAI's official Claude-hosted plugin: use Claude Code from Codex for reviews, rescue tasks, tracked background jobs, and hook-powered review gates.
• Claude Code Harness - Harness blueprint skill for turning vague agent ideas into concrete designs for request assembly, control loops, memory, permissions, recovery, and extension planes.
• Claude Code Skills - 223 production-ready skills, 23 agents, and 298 Python tools across 9 domains — engineering, marketing, product, compliance, and more.
• Claude Octopus - Multi-LLM orchestration dispatching to 8 providers (Codex, Gemini, Copilot, Qwen, Perplexity, OpenRouter, Ollama, OpenCode) with Double Diamond workflows, adversarial review, and safety gates.
• Claudemon - Turns your coding agents into pixel creatures you can watch work and collect, with a built-in Clawland .io brawler for breaks.
• Clean Room - Spec-first clean-room workflow for authorized source analysis, behavioral specs, role separation, and verification without replacement code.
• Click - Turn software intent into one approved contract, then let AI build and verify in one shot—with traceable evidence for each requirement.
• Clueless - Responsibility-flip mode for when the user cannot review the output: irreversible steps first, one verification step, every default named with its alternative, at most three questions only the user can answer.
• Codebase Recon - Analyze git history to understand a codebase before reading any code — auto-scales by repo size and cross-references hotspots with bug magnets to surface high-risk files, bus factor, and team momentum.
• CodeTruss - Local-first acceptance gate that checks coding-agent scope, sensitive surfaces, deterministic analyzers, and repository verification from immutable Git snapshots, then writes signed receipts before the PR.
• Codex Agenteam - Specialist AI agents (researcher, PM, architect, developer, QA, reviewer) orchestrated as a configurable team pipeline.
• Codex Agents Workflow - Build and run reusable agent workflows in a local Codex workbench with configurable Roles, Providers, and recoverable Runs.
• Codex Auto Resume - Unofficial Windows app and Codex plugin that waits out Codex usage limits, rate limits and temporary failures, then continues the exact interrupted conversation by itself, with a Dashboard and a tray icon.
• Codex Full-Stack Workflow - Turns rough product requests into staged, reviewable full-stack delivery with persistent requirements, change control, visual evidence, and completion gates.
• Codex How To - Engineering-first Codex curriculum and plugin with 9 skills, measured token-efficiency experiments, bounded orchestration, testing, review, and living knowledge maintenance.
• Codex Multi Auth - Multi-account OAuth manager for the official Codex CLI with switching, health checks, and recovery tools.
• Codex Process Jobs - Run long local builds, tests, benchmarks, and inference jobs as durable detached processes with tracked status, bounded results, and completion delivery across Codex surfaces.
• Codex Quota Optimizer - Reduces avoidable Codex usage with task-aware model routing, focused context, minimal change surfaces, and layered verification.
• Codex Quota Title - Shows live Codex quota usage and reset time directly in the Codex Desktop window title on Windows.
• Codex Reviewer - Second-pass review of Claude-driven plans and implementations.
• Codex rg Guard - Budgeted rg/grep replacement for Codex that narrows broad searches before they waste model context.
• Codex Skin Pack Installer - Codex plugin and Skill that stages verified desktop skin packs from GitHub releases, validates files, and keeps restore guidance visible.
• Codex Tasklist - A persistent per-session task list for Codex CLI with a live, scrollable panel across terminals through native integrations or a launcher with tmux fallback.
• Codex TUI Proof - Visually validate real local terminal UIs in Codex's in-app browser with screenshots and session evidence.
• Codex Usage and Resets - Turns Codex usage into planning facts with linear pace, projected exhaustion, banked-reset expirations, and conservative unexpected-reset detection.
• Codex–Claude Council - ChatGPT (through the Codex CLI) and Claude (through Claude Code) answer independently, then reply to each other round after round until both agree on one answer, read-only and on your own subscriptions.
• Commit Narrator - Generate semantic commit message from staged diff, including the why.
• Context Attachment Manager - Manage image history in Codex desktop and Claude Desktop Code with on-demand image retrieval, pinned references, and search by description or feedback.
• Context Guard - Preserves authoritative requirements and verification evidence across long-running Codex tasks and context compaction.
• Core Learning - Guided learning with Socratic practice, Feynman teach-back, spaced recall, evidence-based project assessment, and optional local Anki export.
• debt-ops - Catches AI-introduced tech debt at write-time: hooks log every deferral to a registry in your repo and a review skill ranks paydown by file churn.
• DeepSeek Mini-Router - Task-aware spec/react/flash working-contract plugin that fixes DeepSeek weak default reasoning inside Codex, with a measured BigCodeBench harness included.
• dejavu - Cross-session error gates that detect recurring tool-call failures and escalate remind to block, with one shared store across Codex, Claude Code, OpenCode, Gemini CLI, Cursor and more (MIT).
• Deps Doctor - Multi-ecosystem dependency audit (npm, pip, cargo, go) in one report.
• Delx Recovery - Free recovery and continuity plugin for AI agents: resume prior sessions, capture state, process failures into a recovery plan, and remember across sessions through a hosted MCP server (works in Codex, Claude Code, Cursor, and VS Code).
• design-helper - Settles CSS values, layout, and motion by eye: draws candidates that differ in one property side by side in the browser, then applies the one you pick to your code.
• Designer Skill - Plug-and-play MCP that gives your coding agent UI superpowers: design references, intent routing and a static anti-slop gate, no API key.
• Devin Plugin CC - Devin AI code reviews, delegated tasks, Fusion lead/sidekick workflows, and cloud handoffs to app.devin.ai for Codex, Claude Code, and other coding agents.
• Dev Skills - Team workflow skills for specs, plans, TDD, debugging, verification, review, branch finishing, and design context.
• Development Skills - Three-tier triage (PASS_THROUGH / LIGHT / FULL 4-phase) development workflow for Codex and Claude Code with language auto-detection (Python, Java, TypeScript, Swift, frontend) and a staff-reviewer subagent for fresh-eyes review on every change.
• DevRecap - Reconstruct evidence-backed developer work from authorized Codex and Claude history plus read-only Git, then turn it into daily standups, weekly reviews, sprint recaps, and memory reports.
• Docflow - Lightweight documentation memory for AI coding agents that scaffolds a 7-category docs tree, runs readiness checks, validates docs before finishing, and keeps a monthly changelog across Claude Code and Codex.
• done-gate - Stop hook that blocks "fixed/done" claims when no test output or exit code appeared in that turn (Claude Code & Codex, MIT).
• Dryforge - Bounded-autonomy plugin that asks only the decisions that belong to the user, builds nothing before the intent is approved, verifies with checks that actually ran, and keeps project decisions as plain Markdown across Codex, Claude Code, Grok Build, Copilot CLI, and Antigravity CLI.
• ejentum-mcp - MCP server exposing reasoning, code, anti-deception, and memory harness tools for Codex.
• Emulo - Mines selected evidence from local coding-agent sessions into private work, design, and writing profiles for Codex, Claude Code, and GitHub Copilot.
• EndoStack - Security-first workflows for coding agents that threat-model, security-review and verify code changes before they ship, across Codex, Claude Code, Cursor and Copilot.
• Env Lint - .env vs .env.example key parity — never prints values.
• Epic Harness - Auto-trigger quality skills + self-evolving agent harness — orbit (spec-to-ship), evolve (skill mutation), team (multi-agent), TDD, check, ship, simplify, debug, perf, secure.
• Espresso - Concise, evidence-preserving output with bundled Ponytail, for Claude Code, Codex, Pi and OMP.
• falsegreen-skill - Finds tests that stay green when the code they cover is broken, applying six ordered judgments over Python, TypeScript, JavaScript, and Robot Framework suites in Codex CLI and Claude Code.
• fanout - Tests parallel Codex and Claude Code sessions’ branches merged together before main, and stops two sessions doing the same task.
• Figma Maxxing - Eight Agent Skills for real Figma files, with Plugin API gotchas, a read-only preflight before any write, checks after the write and before handoff, and open comments turned into fixes with evidence.
• Flaky Detector - Run a test command N times, report per-test flakiness %.
• Frappe Agent - Frappe and ERPNext coding, customization, bench, and review intelligence for Codex.
• GCF Proxy - Save 71% on MCP tool call tokens by wrapping any server with GCF encoding, with session stats hook and setup skill.
• Generative Media Skills - 13 skills for image, video, and audio generation using 100+ models - FLUX, Midjourney v7, Veo3, Kling 3.0, Suno, and HunyuanVideo via muapi.ai.
• Goal Progress - A native progress bar for Codex Goals with checklist-based progress and token usage on Apple Silicon Macs.
• GrayMatter - Durable memory and shared graph state for Codex and OpenClaw agents, with live ValkyrAI schema awareness.
• Hera Agent Godot - Drives a live Godot 4.x editor through the low-token Hera CLI — scene, node, and signal edits, play control, and runtime QA with verifiable compact JSON output, with the companion addon published on the official Godot Asset Store.
• Hera Agent Unity - Controls and verifies a live Unity Editor through a low-token CLI, with scene, asset, Inspector, Play Mode, test, screenshot, and runtime C# workflows for Codex and other coding agents.
• Herdr Codex Subagents - Opens a read-only Herdr pane for every Codex subagent and closes it when the work is done.
• HOL Guard Plugin - AI antivirus workflow for Codex, Claude Code, Cursor, Gemini, OpenCode, MCP servers, skills, and plugin release checks with local approvals and receipts.
• Homie - Makes and remixes multiplayer web games, music and video, then publishes them from a studio site with public rooms on your own Cloudflare account.
• Honcho - Persistent cross-session memory for Codex powered by Honcho — lifecycle hooks capture each session and inject relevant context back at session start, so Codex remembers your preferences, projects, and decisions across restarts.
• HOTL Plugin - Human-on-the-Loop AI coding workflow plugin for Codex, Claude Code, and Cline with structured planning, review, and verification guardrails.
• Hypercore - Korean-first agent skills for planning, architecture, design, docs, research, issues, and Git delivery, each with declared triggers, workflows, and validation gates.
• Image & Media Skills - Eight skills that audit and fix image and video delivery for Core Web Vitals, plus Sirv image CDN, 360 spin, and AI Studio workflows.
• jev-use - Hands the steps of a Codex session that produce no text — is the build done, which of these elements do I click, is this shell command safe — to Jev, TypeSafe AI's judgment model, with a PreToolUse command gate and typed escalation back to the model for anything it cannot decide.
• Joxo - Setup skill that joins Codex to a team's shared Joxo project, where teammates' coding agents share one task board, decisions and handoffs.
• Keep Mac Awake - Lets Codex keep a MacBook awake with the lid closed while long tasks run, by starting and stopping timed sessions of the Keep Mac Awake macOS app (a paid app with a free trial).
• Kernel - Hook-enforced guardrails for Codex and Claude Code with blocked destructive commands, one-time human tokens for irreversible operations, context-blind verification agents, and SQLite agent memory recalled at session start.
• Krypton - Goal-based planning and proof gate for Codex and Claude Code that turns requests into ownership, cutover, review-gate, and acceptance-evidence plans.
• LLM Transpile - Auto-compress .md, .html, and .txt files via PostToolUse hook, cutting context usage by up to 40% with zero workflow change.
• LVTD Skills - Reusable Agent Skills for Codex, Claude Code, and compatible clients, covering Django, Rust, Cookiecutter, SEO, traction, product marketing, and nonfiction publishing workflows.
• macbox - A real Mac for Codex: build, test, run and tap through iOS apps on the iOS simulator from any machine, including Linux and cloud sandboxes.
• Maestro - Opt-in local multi-CLI fusion engine and orchestration doctrine that fans a prompt across model CLIs, then judges and synthesizes one grounded answer.
• Magents - Shared session bus MCP + skill so Codex can list, read, hand off, send, and spawn with Claude, Cursor, Grok, and friends without pasting context.
• MailAgent - Temporary inboxes for Codex — OTP, magic links, signup QA, simulate-first autotests (23 MCP tools).
• Maintainer Mode - Evidence gates for AI-generated pull requests that catch stale issues, overlapping work, and unsupported test claims.
• Markdown Reader - Navigate Markdown structure and retrieve only the relevant section through a portable Codex MCP plugin.
• Masaka for Codex - Keeps each ChatGPT account and workspace in an isolated CODEX_HOME, shows official usage windows, and selects the identity used by the next native Codex process.
• memi - Interface understanding and design-system memory for Codex, Claude Code, Cursor, and MCP agents with UI audits, Tailwind token extraction, shadcn registry workflows, and a bundled Codex plugin.
• memtomem - Markdown-first long-term memory for Codex that searches, recalls, saves and indexes project notes, and resumes work from a handoff another session or agent wrote.
• MiniCut - Agent skills for Claude Code, Codex, Cursor and Gemini CLI that make a multi-shot AI short film from an idea (story, shot sheet, takes, voice, music, edit) and run Seedance, Kling and Wan video models, with the price shown before every paid job.
• ML Pipeline - Enforces a strict 16-step data-first ML workflow — data inspection, EDA, and problem definition before any model training — with explicit user-permission gates, marimo notebooks, and matplotlib figures at every step; also works in Claude Code and Kimi.
• Model Salt - Gives the model randomness from outside it before a creative task, so one prompt gives different answers that still do what was asked.
• Multi-LLM - Parallel plan, code, and commit reviews across 19 coding CLIs, with validated findings, task generation, implementation, and automated revision.
• Nightshift - Accountable, time-bounded Codex shifts with a persistent punch list, parked decisions, and reviewable run history.
• Off the Wall - Quotes your own past corrections back to Codex, word for word, before each answer; otw report shows what you correct most from your local history, and a routing rule asks the orchestrator to name a model for each subagent. No API key.
• Oh My Cassette - Turn raw local clips into a finished cut by chatting: beat-synced edits, auto-matched music, subtitles, and transitions through a local stdio MCP server, with a timeline delta and contact-sheet preview every turn before anything renders.
• oh-my-openspec - Spec-driven development workflow for Codex CLI and OpenCode, with planning, parallel task execution, and mechanical OpenSpec validation.
• Ontoly - Deterministic Software Graph workflows for Codex: architecture review, dependency analysis, request tracing, configuration analysis, and impact analysis.
• Open Dynamic Workflows - Local-first MIT dynamic multi-agent workflows for Codex, OpenCode, Antigravity, Cursor, and VS Code with a daemon, MCP bridge, Codex skills, OpenCode plugin, and bring-your-own-model support.
• Personal Data Protection - Engineer-facing personal-data-protection compliance reference — Singapore PDPA, Thailand PDPA, Indonesia UU PDP, Malaysia PDPA (Act 709 + 2024 Amendments), Philippines DPA — organised by where in the stack each obligation lands, with checklists, breach-response runbook, and a developer-view divergence table across all five.
• Pixeltable - Declarative multimodal AI data engine for tables, computed columns, embedding search, agents, and FastAPI services.
• Plugin Stack - Six agent skills for getting an AI plugin into the ChatGPT, Claude and Cursor stores and found there: submission checks, listing copy, discoverability audits, widget critique and competitor tracking, on live store data from OpenApps.
• PMC - Give Codex lasting project context in an Obsidian vault you control.
• Ponder - Coverage-gated, multi-pass deliberative review CLI and skill for Codex and Claude Code where four perspectives argue across passes, retract unverified findings, and block stopping until every unit in scope is graded.
• PR Storyteller - PR title + body + test plan from commits and diff vs base branch.
• Praxis - Intent-driven workflow skills for coding agents: describe what done looks like, not the steps. Triage-first design keeps token costs low across design, TDD, debug, review, and release.
• Pro Launch

…