- Author
- Posted by cosmtrek, with replies from many commenters
- Published
- 12 July 2026
- Kind
- Show HN thread about an open source tool
- Original
- news.ycombinator.com/item?id=48878682
- Subject
- Agent tools and setups
What it is
The poster, cosmtrek, shared Mindwalk, a local tool for visualizing coding-agent session logs. The post has no text, so the README and thread are the sources. Commenters discuss uses, alternatives and missing features.
What it says
- The linked README says Mindwalk replays a session as light moving over a repository map.
- It says one Go binary reads Claude Code, Codex and pi session logs, fully local.
- The creator said the aim is to compare LLM task-solving by visualizing agent trajectories.
- Commenters asked for the use case and whether a replay beats reading diffs.
- Supporters said a visual overview shows at a glance which areas an agent changed.
- Commenters compared it to Gource and proposed uses such as comparing models or repeated runs.
- One user noted that history is kept per session and asked for a cumulative view.
What evidence it gives
The README describes features and an optional judge that scores sessions. The thread offers opinions and a few first-hand trials. No measured benefit is given.
Interests and limits
The creator wrote the tool and the README. The README says viewing is local. The optional judge sends a session summary to a model. The demo video and screenshots were not viewed. Commenters disagreed about whether the tool is more than eye candy. No one in the thread reported a case where it changed a decision.