mindwalk

replay coding-agent sessions on a 3D map of your codebase

TLDR

Install and launch (scans Claude Code and Codex session dirs)

$ mindwalk

Serve the UI on a fixed port without opening a browser

$ mindwalk serve --port [8765] --no-open

Open one session file

$ mindwalk open [path/to/session.jsonl]

Build a repository citymap JSON

$ mindwalk build [path/to/repo] -o [out.json]

Normalize a session into a trace

$ mindwalk trace [session] -o [out.json]

Evaluate a session with a local agent CLI as judge

$ mindwalk analyze [session] --judge [claude|codex]

SYNOPSIS

mindwalk [command] [options] [args]

mindwalk visualizes how a coding agent explored a repository. It reads local Claude Code and Codex session logs, builds a deterministic layout of the repo (a "citymap"), and plays file reads, edits, and searches back as light on a 3D map in the browser.With no arguments it scans default session directories, serves a local UI on a random port, and opens a browser. Viewing is fully local; optional mindwalk analyze (or the Evaluate panel) can send a session summary to your own claude or codex CLI for judging.

PARAMETERS

serve

Start the local server (--port, --no-open, --claude-dir, --codex-dir)

open session.jsonl

Open a specific session (--no-open to skip the browser)

build repo

Write citymap JSON (-o output path)

trace session

Write normalized trace JSON

analyze session

LLM judge evaluation (--judge, --model)

CAVEATS

Session evaluation is opt-in and uses your local agent CLI, which may send a summary of that session to the model behind your account. Reports cache under ~/.mindwalk/reports.

RESOURCES

Source code