Codex CLI vs Claude Code vs Hermes

Codex CLI, Claude Code, and Hermes Agent all sit near the same architectural layer: agent harness / coding agent runtime. They can read repos, use tools, run commands, and execute coding tasks. The differences are product focus, openness, runtime customization, and operational model.

Quick comparison

Tool	Category	Best for
Codex CLI	Polished coding agent CLI	Day-to-day coding workflows in a repo
Claude Code	Polished coding agent CLI	Agentic coding in the Claude ecosystem
Hermes Agent	Open-source, hackable agent runtime	Custom/self-hosted/internal agent platforms

Similarities

All three can be used as the execution layer for workflow frameworks:

flowchart TB
    A[Codex CLI / Claude Code / Hermes] --> B[Spec Kit]
    A --> C[OpenSpec]
    A --> D[AI-DLC]
    A --> E[Superpowers]
    A --> F[GSD-like execution]

They all answer:

How does an AI agent interact with my repo and tools?

They do not, by themselves, fully answer:

What artifact is source of truth? What governance model should the team follow?

Key differences

Dimension	Codex CLI	Claude Code	Hermes Agent
Product style	Polished developer CLI	Polished developer CLI	Open-source runtime/CLI
Best default use	Coding and repo tasks	Coding and repo tasks	Custom agent platform or research
Runtime hackability	Lower	Lower	Higher
Self-host/local model fit	Depends on product support	Depends on product support	Stronger fit conceptually
Memory/skills/subagents	Product-dependent	Product-dependent	Central emphasis
Enterprise governance	Needs workflow/process	Needs workflow/process	Needs workflow/process
Best paired with	Spec Kit, OpenSpec, Superpowers	Spec Kit, OpenSpec, Superpowers	OpenSpec, AI-DLC, Superpowers, custom tools

Decision guide

Situation	Choose
You want the fastest path to high-quality repo coding	Codex CLI or Claude Code
You are already productive with Codex/Claude	Do not add Hermes unless you have a runtime requirement
You need local/self-hosted models or custom routing	Consider Hermes
You need custom internal tools and memory	Consider Hermes
You need spec/change workflow	Add OpenSpec or Spec Kit to any harness
You need enterprise governance	Add AI-DLC to any harness
You need TDD/review discipline	Add Superpowers to any harness

Typical stacks

Developer productivity stack

Codex CLI or Claude Code
  + OpenSpec or Spec Kit
  + Superpowers
  + repo tests/CI

Internal agent platform stack

Hermes Agent
  + model router or local LLM
  + internal tools
  + OpenSpec for change specs
  + AI-DLC for high-risk governance
  + CI/test evidence

Practical conclusion

If you already use Codex CLI or Claude Code successfully, Hermes is not automatically an upgrade. It is a different trade-off:

Codex CLI and Claude Code optimize for polished day-to-day developer productivity. Hermes optimizes for open, customizable agent runtime control.