@gaoren7716: Version control system for AI agents. Tracks what the agent did, which prompt wrote which line, and allows rollback to any step. Like .git has a .regent directory, but it records agent activity instead of human commits. Several core commands: rgt log —…

Version control system for AI agents. Tracks what the agent did, which prompt wrote which line, and allows rollback to any step. Like .git has a .regent directory, but it records agent activity instead of human commits. Several core commands: rgt log — View each step the agent took, with timestamps and change summaries rgt blame — Pinpoint to the line: which prompt made the agent write this line of code rgt show — View the full context of any step, including tool calls and conversation history Other aspects: Multi-session parallelism, each session has an independent ref with no conflicts; content-addressed storage using BLAKE3 hashing with automatic deduplication; SQLite index queries under 10ms; after rgt init, Claude Code / Codex / OpenCode automatically enable tracking with zero configuration; VSCode extension supports inline blame annotations and hover context. The most noteworthy scenario: rgt blame src/handler.go:42 doesn't just tell you who changed this line, but tells you which prompt made the agent write this line — you can see the conversation context, tool call parameters and results at that time. Code review is no longer "the agent wrote it, I don't know why", but every line has a traceable source of intent. http://github.com/regent-vcs/re_gent… #AgentVCS #版本控制 #审计 #ClaudeCode