Feature request: Add support for OpenAI Codex (port plugin beyond Claude Code)

[adairlcasey]

Summary

I'd love to see an official version of the Compound Knowledge plugin that works with OpenAI Codex CLI. Today the plugin is built specifically for Claude Code — installation goes through the Claude Code plugin marketplace (/plugin marketplace add ...), the workflows are exposed as Claude Code skills/agents, and the customization layer reads CLAUDE.md. None of that carries over to Codex, so Codex users can't adopt the compounding-knowledge loop.

Problem / motivation

The core value of this plugin — the /kw:brainstorm → plan → confidence → review → work → compound loop, with learnings saved to docs/knowledge/ and surfaced automatically next time — is agent-agnostic in principle. But the current implementation is tied to Claude Code:

• Install path: requires the Claude Code plugin marketplace; there's no equivalent entry point for Codex.
• Commands/agents: the 6 skills, 2 review agents, and 3 research agents are defined as Claude Code components.
• Customization: the strategic-alignment and data-accuracy reviewers read CLAUDE.md for business context, data hierarchy, and style guides.

As a Codex user, I can't take advantage of any of this, even though my projects already use the same plans/ + docs/knowledge/ markdown convention the plugin relies on.

Proposed solution

Provide a Codex-supported distribution of the plugin. A few possible directions (open to whatever fits your architecture best):

1. Map the /kw:* workflows onto Codex's command/prompt mechanism so the same loop is callable in a Codex session.
2. Read project context from a config-agnostic source (e.g. also support AGENTS.md / a generic config file) in addition to CLAUDE.md, so the reviewers work regardless of which agent is running.
3. Document a Codex install/setup path equivalent to the current Claude Code marketplace flow.

Since the knowledge files are already plain, git-tracked, greppable markdown, the underlying knowledge layer should port cleanly — the work is mostly in the agent-integration layer.

Alternatives considered

• Manually copying the prompts into Codex by hand — works in a pinch but loses the parallel research/review agents, the auto-surfacing of past learnings, and the install/update experience.

Additional context

• Plugin currently documents Claude Code as the only supported host.
• Happy to help test a Codex build or provide feedback on the integration. Thanks for building this — the compounding-knowledge concept is great and I'd like to use it in my Codex workflow.
• The plugin is cited/recommended to use within this Every guide to Codex here: https://every.to/guides/codex-for-knowledge-work > Level 5: Compound your Codex system.