Join the discussion

Write your take first — we'll ask for email only when you're ready to publish.

  • Hacker News
  • Hi everyone, I built Ponytrail, a small CLI and bundled agent skill for making AI coding sessions easier to inspect and undo.

    The problem I kept hitting: git diff shows what changed, but not why the agent changed it, what outcome it expected, what check it planned to run, or how to roll back one specific action. Final summaries are often too coarse, especially after a long coding session.

    Ponytrail adds a small local trail around file mutations. Before an agent creates, edits, moves, deletes, formats, or generates files, it records a pre-change snapshot: action, purpose, reason, files, expected outcome, verification plan, rollback path

    After the edit, it records a post-change snapshot with what changed, what checks ran, and the result.

    Everything is stored locally under .pony-trail/ as JSONL, session trees, and small before/after file copies. It does not replace git. It is more like per-action intent and rollback context for agent work.

  • Have you heard of git?
  • Not to be confused with "Ponytail", an award-winning puzzle-platform game by Jonathan Blow.
  • And what if the changes are remote?
  • Yuck. Use jj for this instead of a custom thing.
  • I get enough of this with a pre/post tool hook that takes a jj snapshot and teaching the agent about `jj evolog`.
    by dwb
  • My entire repo is an audit trail of AI coding-agent edits
  • Not only that, but Claude can undo its own changes I just never use the feature because git has me covered...