Home › Learn
Learn
How Claude Code, Codex and OpenCode work underneath, and how to hook into them
Hooks, sessions, memory and helper agents in the three runtimes. Each tutorial lists the versions it was checked on, the date it was checked, and how we know each claim.
Tutorials
You can read them in order or start with the one you need. The first one needs nothing of ours installed.
-
01 · Tutorial
Claude Code hooks, Codex hooks and OpenCode plugins, desktop apps included
The Claude and Codex desktop apps run the same runtime as the terminal, and it reads your hook configuration. Log tool calls, block them, and add notes for the model with a short script.
-
02 · Tutorial
Spontaneous Memory: a second agent that reminds your session of past decisions
On each prompt, a helper searches the decisions you wrote down and hands the main agent a note when one applies. Build it with two Claude Code hooks and a Python script, then see it as a Crossing Guard profile.
Reference
Reference pages compare the configuration layer across runtimes. Planned pages are listed here but are not linked until they exist.
| Page | Covers | Status |
|---|---|---|
| Hooks across runtimes | Where hooks live, what they receive and how they answer, in Claude Code, Codex and OpenCode | Published |
| Claude Code memory, actually explained | What loads into a session, when, and from where | Planned |
| CLAUDE.md: what loads, when, and how big | Project instruction files and their limits | Planned |
| MCP config across runtimes | Declaring tool servers in each runtime | Planned |
| Skills across vendors | Packaged instructions, and how each runtime finds them | Planned |
How we write these
Each factual sentence in a tutorial has a label: Observed on a real install, In source (not yet public), from Vendor docs, or Not verified. Each page names the versions it was checked on and a date by which it will be checked again. When a vendor changes a hook, we update the page and its dates.