Readable re-implementations of real coding-agent harnesses, with one web UI that shows every hidden step, so students can learn and compare how the harnesses work.
Harness: The original coding-agent product that we study, for example Codex or pi. Avoid: original, upstream, agent
Backend: Our re-implementation of one harness, for example educodex (Codex) or edupi (pi). Avoid: server, port, clone
Visualiser: The shared web UI that connects to one backend at a time and shows what it does. Avoid: client, frontend, web client
Viewer: One open visualiser page that is connected to one backend. Avoid: tab, window, session
Launcher: The one program that serves the visualiser, lists the available backends and starts a backend for each viewer. Avoid: server, host, orchestrator
Adapter: The part of the visualiser that turns one backend's native protocol into the shared view. Avoid: translation layer, bridge
Trace event: A teaching-only message that a backend sends in addition to its harness's native protocol. Avoid: debug event, telemetry
Agent turn: One pass through the agent loop: one model call plus the tool calls it asks for. Avoid: step, loop iteration (pi calls it a turn; Codex has no name for it)
Agent run: One user message and all the agent turns after it, until the model replies with no tool calls. Avoid: turn (Codex calls this a turn), task
Agent session: All the agent runs that share one message history. Avoid: thread (Codex's name), conversation