cortextOS Multi-Agent Coding Orchestration System
source post: Video by agentic.james
Video by agentic.james
Source: instagram · James Goldbach Saved: 20260613 Tags: instagram Display: cortextOS Multi-Agent Coding Orchestration System — cortextOS unifies OpenAI Codex and Claude Code via shared config, instructions, and external markdown memory to prevent drift.
TL;DR
cortextOS is a personal agentic orchestration system built by James Goldbach that enables developers to run OpenAI Codex and Anthropic Claude Code in a unified, synchronized configuration. It introduces conventions — a shared agents.md instructions file, a /duplicate slash command for porting skills between harnesses, and external markdown-based memory — to prevent the two AI coding agents from drifting out of sync. Running multiple AI coding agent harnesses (Codex and Claude Code) in parallel creates config drift: separate instruction files, incompatible skill storage locations, and diverging memory mechanisms. cortextOS solves this by establishing a single canonical config layer and cross-agent memory strategy that both harnesses can consume without duplication.
What the post showed
Caption: If you run both Codex and Claude Code you've probably felt the pain of keeping their configs in sync. Here's how I actually solved it. agents.md is the canonical instructions file now, so instead of maintaining a separate CLAUDE.md, just import agents.md into it and you're done. Skills use the same yaml frontmatter but live in different folders, .agents for Codex and .claude for Claude Code, so I built a /duplicate slash command that ports a skill from one harness to the other in one shot. And for memory I stopped trusting either harness's internal mechanism, everything lives in plain markdown plus an external knowledge base both agents can read with their own tools. That's the whole trick to running two harnesses without them drifting apart. I go deep on orchestrating Codex and Claude Code together inside cortextOS in my Skool community. Link in bio.
Key claims from transcript: Here's the best way to share agent configurations between Codex and Cloud Code. First of all, agents.md is now your canonical instructions file for projects, and instead of writing it out in agents.md and Cloud.md separately, just at agents.md in your Cloud.md file with nothing else that automatically imports it. Similarly for skills, they use pretty much the same skill yaml front matter to actual
What it actually is
- What: cortextOS is a personal agentic orchestration system built by James Goldbach that enables developers to run OpenAI Codex and Anthropic Claude Code in a unified, synchronized configuration. It introduces conventions — a shared agents.md instructions file, a /duplicate slash command for porting skills between harnesses, and external markdown-based memory — to prevent the two AI coding agents from drifting out of sync.
- Who built it / maintained by: James Goldbach (independent creator, a.k.a. agentic.james), distributed via his Skool community
- Status: unknown
- Why it matters: Running multiple AI coding agent harnesses (Codex and Claude Code) in parallel creates config drift: separate instruction files, incompatible skill storage locations, and diverging memory mechanisms. cortextOS solves this by establishing a single canonical config layer and cross-agent memory strategy that both harnesses can consume without duplication.
- How it compares to alternatives:
- OpenAI Codex CLI
- Anthropic Claude Code
- Aider
- Continue.dev
- Cursor
- Cline
- GitHub stars: 0 · License: unknown · Archived: no
Links
- (no links found)
Kickstarter guide
cortextOS is not publicly available as an open-source repo; access is through James Goldbach's Skool community (link in his Instagram bio @agentic.james). The core conventions can be adopted independently: create an agents.md as your canonical instructions file and import it into CLAUDE.md, store skills in .agents/ and .claude/ folders respectively, add a /duplicate slash command to port skills between harnesses, and use plain external markdown files (or a vector DB / knowledge graph) for shared cross-agent memory. Following these patterns gives you the essential cortextOS workflow without the full system.