Install alongside Codex.
A local Codex plugin opens a separate learning companion in the built-in browser panel when available.
Prefer agent-assisted installation? Paste one URL into your coding agent.
Install
- Download the Codex macOS ZIP from Learning Harness and unzip the entire folder.
- Run Install Learning Harness.app. Enable Learning Harness on the plugin page.
- Start a new Codex coding chat. Restart the app if the plugin has not loaded.
Review the readiness hook
Open Settings → Hooks, or Review hooks above the composer. Review Learning Harness’s hook. Plugin enablement and hook trust are separate. The hook carries a fixed, rate-limited readiness instruction, not your lesson, source excerpts or learner profile.
Approvals can survive reinstalling; do not edit trust files to force a prompt or bypass review. Changed hook definitions may require another review.
Open and connect Learn
- Say Start Learning Harness, or choose the Start Learning Harness skill.
- Choose the project and coding chat, then Watch this chat.
- Keep Learn beside the coding chat while you begin a task.
Return, pause and stop
Say open Learn to return after closing the panel. Closing does not stop observation. Use Learn’s off/pause controls to stop learning. To shut down the local companion, ask Codex to Stop Learning Harness.
How model execution works
The adapter runs separate Codex model work through your existing ChatGPT sign-in. It is learning-subagent behavior implemented by a separate runtime, rather than a claim that every call is a native Codex subagent tool. It uses additional allowance. No new model API key or Learning Harness inference service is introduced.
Update or remove
Finish active work, update through the same installer/plugin manager, restart, and review changed hooks if requested. Updates preserve local memory. Remove Learning Harness through Codex’s plugin manager; do not delete live writer locks manually.
Local logs: ~/.local/share/dontdumb/codex/companion.log. See recovery steps.