Mindwalk review: replay an AI coding session in 3D

Mindwalk is an open source tool that turns a codebase into a 3D city and replays activity from a coding assistant's session logs. The walkthrough uses an earlier Claude Code session: each file appears as a building, with taller buildings representing more code. A timeline lets the viewer follow which files the agent examined before making changes.

The presenter explains the colors: moss green marks a glance, blue marks a read, and amber marks an edit. Timeline symbols identify subagent launches, context compaction and new user instructions. Mindwalk uses a Go binary to read logs already on disk and a browser scene to display the replay. These events show the agent's recorded activity; they do not directly establish its internal reasoning.

An analyze command uses the user's Claude or Codex coding client to assess exploration, scope and verification. The presenter says each finding links to the relevant moment in the timeline. A separate map command renders a repository without a session replay.

The review covers version 0.3, which reads Claude Code and Codex logs but cannot replay Cursor sessions. The speaker sees it as useful for comparing how agents explore a task or reviewing a large change, while recommending a plain Git diff for a small fix.