A CLI tool that records which AI agent wrote each code change and why

When multiple AI coding agents (Cursor, Copilot, Claude Code, Devin, aider) touch a codebase, there is no way to know which agent made which change, what prompt drove it, or why a decision was made. Developers lose accountability at the commit level and cannot audit, rollback, or blame AI-authored changes with any more precision than a co-author tag. This tool wraps the version control layer with per-change agent attribution, prompt capture, and a queryable audit log.

Demand Breakdown

GitHub

531

216

127

Social Proof 3 sources

regent-vcs/re_gent – Version-Control for AI coding agents

@gh:regent-vcs · 2026-04-10

531 HN

Show HN: Git for AI Agents

@doshay · 2026-05-07

216 PH

Re_gent – Version Control for AI agent Activity

@re_gent · 2026-04-10

127

Gap Assessment

CompetitiveMultiple tools exist but differentiation opportunities remain

4 tools exist (re_gent, Origin, Cursor Blame (Cursor Enterprise), Korso (YC)) but gaps remain: No prompt logging, no cross-agent support (vendor-agnostic attribution), no queryable audit API, no team/enterprise export, no CI/CD integration gate.; Commit-level only (no per-hunk or per-line prompt context), early stage with minimal traction, no cross-agent unified dashboard, no compliance export..

Features7 agent-ready prompts

Vendor-agnostic agent hook layer

▶

Per-change prompt and session capture

▶

Queryable audit log CLI

▶

Team sync and shared audit store

▶

CI/CD attribution gate

▶

Web dashboard for team leads

▶

Rollback and bisect by agent

▶

Competitive LandscapeFREE

Product	Does	Missing
re_gent	Tracks AI agent sessions in a .regent directory alongside .git. Captures what changed per session.	No prompt logging, no cross-agent support (vendor-agnostic attribution), no queryable audit API, no team/enterprise export, no CI/CD integration gate.
Origin	Git blame for AI agents. Tags each commit with which AI assistant wrote it.	Commit-level only (no per-hunk or per-line prompt context), early stage with minimal traction, no cross-agent unified dashboard, no compliance export.
Cursor Blame (Cursor Enterprise)	Extends git blame within Cursor IDE to distinguish tab completions, agent runs, and human edits.	Cursor-only (no Copilot, Claude Code, Devin, aider attribution), locked to Cursor Enterprise tier, no exported audit log, no prompt capture.
Korso (YC)	YC-backed AI agent with built-in audit trail showing what was done and why.	Proprietary to their own agent only, not a vendor-neutral attribution layer that works across all AI coding tools a team already uses.