Agent Harness Portability

Agent Harness Portability

Agent harness portability is the practice of organizing project files, instructions, skills, tools, and operating context so multiple coding-agent runtimes can work in the same directory without losing the workflow.

Key points

  • Nate Herk tested this by moving his Claude Code AIOS project into Codex (OpenAI), where Codex adjusted to the project structure in about two minutes [src-013].
  • In the Codex course, Nate frames the shared directory as the durable layer: folders, instructions, settings, skills, agents, and project files can be understood by Codex, Claude Code, Cursor, OpenClaw, or similar harnesses if the structure is explicit [src-048].
  • In the May 2026 tech-stack video, Nate makes the principle explicit: build directories as if they will outlive any specific tool, because Claude Code, Codex (OpenAI), Hermes Agent, OpenClaw, Cursor, or a future startup's harness can all plug into the same organized workspace [src-053].
  • Portability depends on keeping instructions in files rather than only in chat history; the harness reads project context, follows local conventions, and uses the same deterministic tools or APIs [src-048].
  • The strategic benefit is model/harness optionality. Nate uses Claude Code for exploratory planning and creativity, while using Codex (OpenAI) for pragmatic execution, long-plan following, troubleshooting, and review [src-048].
  • Portability does not remove the need for permissions, credentials, or connector setup. Each harness still needs safe access to APIs, files, browser automation, plugins, or MCP servers [src-048].
  • Jack Roberts adds a practical skill-migration recipe: export Claude skills plus overview/implementation notes into a folder, then ask Codex to recreate them as Codex skills so the same work patterns survive the harness move [src-058].
  • Roberts's Lovable-to-GitHub-to-Codex workflow is another portability example: the artifact moves through GitHub so a UI generated in one tool can be enhanced by another [src-058].
  • [src-064] adds an OpenClaw-specific version: a harness becomes powerful when the agent can inspect its own source code, documentation, tools, runtime, and model settings, then use those facts to debug or modify itself.
  • [src-074] extends the pattern into personal-assistant infrastructure: Nate keeps the same GitHub-backed files, skills, memory, and operating context usable across Claude Code, Hermes, OpenClaw, Codex, and other harnesses, with small compatibility edits when naming conventions differ.
  • [src-079] adds the runtime version of portability: Hermes Agent can act as a Telegram/mobile surface while reading the dashboard, logs, memory, and project state from a Claude/Codex-style desktop operating system.
  • [src-080] adds Cursor's product-side evidence that the harness layer itself is evolving: tab completion, local agents, and cloud agent teams are different interaction models over the same underlying codebase.
  • [src-084] adds OpenAI's internal terminology: Harness Engineering means making repositories agent-legible through bootstrap commands, validation harnesses, worktrees, docs, standards, skills, and decision records so multiple Codex agents can work safely.
  • The same source extends portability beyond code with a company-context repo pattern, where operating principles and non-code context become readable inputs for Codex planning and decision support [src-084].
  • [src-086] adds two more portability surfaces: n8n Desk can sit over Anthropic SDK, LangChain Deep Agents, local models, skills, and connectors; Roberts's Agentic OS dashboard spans Claude Code, Codex, OpenClaw, OpenRouter, local memories, and vector stores as one visible AI estate.
  • [src-087] makes the Claude Code to Codex mapping explicit: CLAUDE.md and .claude/ serve Claude Code, while AGENTS.md, .codex/, and .agents/ serve Codex; shared context files, scripts, references, skills, and project folders remain the durable layer both harnesses can read.
  • The same source reframes portability as a work-continuity pattern: if Claude Code stalls, Nate can hand the same project to Codex without changing the underlying context or duplicating the project.
  • [src-103] reinforces that a second brain should be tool-agnostic: Claude Code can route through CLAUDE.md, Codex can route through AGENTS.md, and both can read shared markdown memory and project files when the workspace structure is explicit.
  • The same source extends portability to Hermes-style assistants: the useful layer is not the interface itself, but shared files, skills, routing rules, and access paths that let different harnesses find the same context [src-103].

Related entities

Related concepts

Source references

  • [src-013] Nate Herk — "Build & Sell Claude Code Operating Systems (2+ Hour Course)" (2026-05-01)
  • [src-048] Nate Herk — "Master 97% of Codex in 1 Hour (full course)" (2026-05-06)
  • [src-053] Nate Herk — "Overwhelmed By AI? Just Copy My Tech Stack" (2026-05-08)
  • [src-058] Jack Roberts — "How to use Codex Better than 99% of People" (2026-05-06)
  • [src-064] Lex Fridman – "OpenClaw: The Viral AI Agent that Broke the Internet – Peter Steinberger | Lex Fridman Podcast #491" (2026-02-12)
  • [src-074] Nate Herk — "Hermes Agent: Zero to Personal AI Assistant (1 Hour Course)" (2026-05-10)
  • [src-079] Jack Roberts — "Hermes Agent just got 10X Better (Agentic OS)" (2026-05-15)
  • [src-080] Cursor — "The next era of AI coding" (2026-05-12)
  • [src-084] OpenAI Codex, Workspace Agents, Prompt Caching, and Superintelligence Policy cluster (2026-02-09 to 2026-05-08)
  • [src-086] Agent deployment, OpenClaw trading, n8n Desk, and Agentic OS cluster (2026-04-09 to 2026-05-15)
  • [src-087] Nate Herk late-May 2026 cluster (2026-05-17 to 2026-05-30)
  • [src-103] Nate Herk – "Every Level of a Claude Second Brain Explained" (2026-06-17)

Robin Cartier perspective

This page is part of Robin Cartier's working AI knowledge graph: a practical research layer for production AI, recommendation systems, experimentation, GEO, and agentic web readiness.

The useful next step is to connect this concept back to applied product leadership and operating models.

Recommended next

Keep reading from this thread

From 491 indexed pages and articles.

  1. Wiki concept Codex (OpenAI) OpenAI's coding and agentic work surface, available through the Codex app, VS Code extension, terminal, and a Claude Code plugin. Related by 048
  2. Wiki concept Global vs Project Skills Claude Code skills can be scoped to a single project (live in .claude/skills/) or installed globally (live in ~/.claude/skills/) and available from Related by portability
  3. Insight AI Beyond POCs How enterprise AI moves beyond proofs of concept through ownership, governance, measurement, adoption, and production operating models Readers have engaged with this next