dire niche
Built a small tool called agentdiff to answer a question I kept running into with AI coding agents:
“How do you know what and why the agent actually changed?”
agentdiff hooks into coding agents like Codex, Claude Code, Cursor, Copilot, Windsurf, OpenCode, and Gemini, then writes a repo-local attribution ledger for each commit. You can inspect who changed what, down to line-level blame, across your git history.
What it does:
- Tracks which agent wrote which lines
- Stores a commit-scoped audit trail locally in the repo
- Lets you query attribution with commands like list, blame, stats, and diff
- Supports signed ledger entries plus verification/policy checks for teams that want stronger governance
The main use case for me is teams using multiple agents who want visibility, accountability, and a clean audit trail without relying on a hosted service.
Open source here: https://github.com/codeprakhar25/agentdiff
