Source linked

مسار التحقيقات المحلية الذي يسمح لك بالعودة إلى أي عامل AI Edit

Ponytrail يقوم بتسجيل كل تغييرات في الملفات من Claude، Copilot، و Codex إلى شجرة الفيديو المحلي؛ استعادة حالة المشروع باستخدام إجراء واحد.

ponytrailclaudecopilotcodexai coding agentsdeveloper tools

AI coding agents change files fast, and they don't leave an undo log. Ponytrail, a 14KB CLI from 0xroylee, fixes that by recording every file mutation into a local snapshot tree you can walk back through.

How Ponytrail Tracks Every Agent Edit

Ponytrail inserts itself as a skill into Claude, Copilot, and Codex. Run npx ponytrail skills install pony-trail and it wiretaps the agent's file operations. Every edit gets a snapshot entry in .pony-trail/, split into pre and post state. The history command prints a tree:

npx ponytrail history --details

You see the action type, summary, checks run, result, and even a rollback instruction. Snapshots are stored as JSONL in .pony-trail/snapshots.jsonl alongside session files. The folder is local runtime state, deliberately excluded from git.

Rolling Back Is a Single Command

Mistakes happen. Ponytrail's revert <snapshot-id> restores files from that snapshot's pre state. Run with --dry-run to preview what changes, then confirm. If a file didn't exist before the snapshot, revert deletes it. Non-interactive environments get a printed plan and exit without changes.

Snapshots are identified by timestamps and hashes. The tree shows the full lineage, so you can revert not just the last change but any point in the history. The system even records the skill install itself as a snapshot, making rollback of the agent configuration itself straightforward.

Designed for the AI Coding Workflow

Ponytrail is built for developers who let agents write code but want control. It's a companion to git, not a replacement. The .pony-trail/ folder stays local, so you get fine-grained undo without polluting your commit history. The tool is written in TypeScript with a bit of Shell and Python, and ships as a single npm package.

With 104 stars on day one, it's filling a gap that every agent-using developer has felt. Expect this pattern to become table stakes for coding agent tooling.


Source: I built Ponytrail, a local audit trail for AI coding-agent edits
Domain: github.com

Read original source ->

External source stays available while the OJO article and comment thread stay local.

Comments load interactively on the live page.