Mindwalk: Replay Claude Code/Codex sessions as light trails on a 3D code map

Mindwalk is a visualization tool that replays coding-agent sessions on a 3D map of your codebase. Instead of staring at raw JSONL session logs, you get a night map where light moves as the agent searches, reads, and edits — untouched parts stay dark.
What it solves
A session log tells you what an agent did, but not how it understood the task: which parts of the repo it treated as relevant, where it explored before acting, or whether its footprint matched your intent. Reading JSONL line by line doesn't answer that.
Quick start
curl -fsSL https://raw.githubusercontent.com/cosmtrek/mindwalk/master/scripts/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"
mindwalkThe installer verifies the binary against checksums.txt and installs to ~/.local/bin (override with INSTALL_DIR; pin a release with VERSION). Windows archives are on GitHub Releases. To build from source: make setup && make build → bin/mindwalk.
Commands
With no arguments, mindwalk scans ~/.claude/projects and ~/.codex/sessions, serves the UI on a random local port, and opens a browser:
mindwalk serve [--port N] [--no-open] [--claude-dir DIR] [--codex-dir DIR]mindwalk open [--no-open] <session.jsonl>— open one specific sessionmindwalk build <repo> [-o out]— write the repository citymap JSONmindwalk trace <session> [-o out]— write the normalized trace JSON
Reading the visualization
- Tree / Terrain views — the repo as a radial tree or treemap; glow intensity ∝ how deeply and often a file was touched.
- Touch states — each file keeps its deepest touch: seen (moss green), read (moon white), edited (warm amber), unvisited (dark).
- HUD folds friction signals — error rate, churned files, edits after the last verify — into a review strip.
- Playback deck — scrub or play over a bucketed histogram. Bars on a cool/warm spectrum: observation (search, read, exec) stays cool, mutation (edit, verify) glows warm.
- Timeline marks — ◇ context compactions, ○ subagent launches, › user turns; click to jump.
- Inspector — click a file to pin its visit history; click a visit row to jump the playhead.
- Keyboard: Space play/pause · ←/→ step (⇧×10) · Home/End ends · S speed · E next edit · X next error · M next mark · ⌘B session rail.
Under the hood
Two separate artifacts: a trace (normalized stream of file-touch events from session logs via adapters in internal/adapter) and a citymap (deterministic layout of the repository in internal/citymap). A local Go server (internal/server) joins them and serves the React/Three.js frontend (web). Schema contracts live in schema/.
The same tree always produces the same map, so replays are comparable across sessions. No session data leaves your machine — fully local.
📖 Read the full source: HN LLM Tools
👀 See Also

Traversable Skill Graph for Persistent AI Agent Memory in Codebases
A developer built a three-layer skill graph system that lives inside a codebase, enabling AI coding assistants to maintain persistent memory across sessions. The system uses progressive disclosure with self-directing instructions instead of monolithic context files.

AI Code Review Benchmark: Claude, Gemini, Codex, Qwen, and MiniMax Compared
A benchmark tested five AI models on 15 Milvus pull requests with known bugs. Claude detected 53% of bugs in raw mode, while adversarial debate between models increased detection to 80%.

SimSense MCP Connector Gives Claude Artifacts Permanent URLs with Persistent State
SimSense is an MCP connector that lets Claude deploy generated HTML/JS artifacts to permanent URLs called 'sims' with persistent state storage. The tool addresses the limitation where Claude's output disappears when you close the chat window.

OpenClaw Implements Agent History Compression to Reduce Context Usage
OpenClaw now compresses agent history by replacing completed subtask logs with structured summaries, reducing ~1M tokens to ~30K. The system uses a 4-pass scanner to identify task lifecycles and generates masked summaries that maintain agent compatibility.