MindWalk

MindWalk helps developers review coding-agent sessions by showing where an agent explored and changed a repository. It turns Claude Code, Codex and pi logs into a visual replay for comparing the agent's work with the task you assigned.

• Repository maps: Tree and terrain views distinguish files the agent encountered, read or edited, while a review strip flags errors, repeated edits and edits after the last verification.
• Session playback: Scrub through a run, jump to user turns, context compactions or subagent launches, and replay individual subagent sessions.
• Comparison and export: An unchanged repository tree produces the same map across sessions, and you can export playback as WebM video entirely in your browser.
• Optional evaluation: Your own Claude or Codex CLI sends task wording, file paths and event summaries to its model to assess exploration, scope, wandering, verification and criteria drawn from your request.

Evaluation findings link to timeline evidence. The application calculates verdicts from finding severity; when the log can't establish whether a criterion was met, the report marks it as "no signal" rather than a failure.

MindWalk is open source under the MIT license. A local Go application serves the browser viewer, and session viewing sends no data elsewhere. Optional evaluation sends the session summary to the model behind your Claude or Codex CLI.