# roborev

Continuous code review for AI coding agents: a post-commit hook reviews every commit in the background as agents write code and surfaces issues in seconds. The framing is accountability for every line of generated code. Part of the [[kenn-software-suite]].

## What it does

Three steps. `roborev init` installs a post-commit git hook. Every commit then triggers a background review with no change to your workflow. Findings appear in an interactive TUI (vim-style navigation) or get fed back to agents. Built-in analysis types cover duplication detection, complexity, refactoring suggestions, test fixtures, dead-code identification, and security scanning. A verification pass filters false positives before findings surface.

`roborev fix` closes the loop: it feeds findings to an agent for automatic remediation, running the fix in an isolated git worktree so the working tree isn't disturbed; `roborev refine` iterates on a fix. Multi-agent support — Codex, Claude Code, Gemini, Copilot, and others — so the reviewer and the fixer can be different models.

## How it works technically

Go (~99%), v0.55. The review runs locally against the diff of each commit; there's no hosted service. Fix operations use isolated worktrees so an agent with full tool access can edit freely without touching your checkout. Configuration is `.roborev.toml`: agent selection, review guidelines, model specs, plus environment-variable overrides and extensible hooks for custom shell commands. Optional PostgreSQL sync persists review data (the same team-Postgres pattern as [[agentsview]]). For the Claude Code agent, reviews can route through a proxy (Ollama, LiteLLM).

## Basic usage

```bash
roborev init                 # install post-commit hook
# ... agents commit code ...
roborev tui                  # browse findings
roborev fix                  # feed findings to an agent, fix in a worktree
roborev refine               # iterate on the fix
roborev analyze <type>       # run a specific analysis on demand
```

Installs via shell script, Homebrew, Windows PowerShell, or `go install` — macOS/Linux/Windows.

## Limitations

- Requires the coding agents themselves already installed on the system.
- **Designed for trusted codebases.** Review agents have shell access and fix agents run with full tool access in isolated worktrees, so the README recommends running inside a sandbox when reviewing untrusted code.
- Finding quality depends on the underlying agent and model.

## Related

- [[charlie-daemons]] — markdown-defined autonomous background processes for operational hygiene; roborev is the narrower, commit-triggered code-review case of the same "agent runs continuously in the background" pattern.
- [[middleman]] — sibling tool for triaging *others'* PRs on the forge; roborev reviews *your own* commits locally before they get there.
- [[agentsview]] — sibling that tracks agent volume/cost; roborev tracks agent output quality. Both can sync to a shared Postgres.
- [[toolbox/re_gent]] — per-prompt agent VCS; same "accountability for generated code" instinct via provenance rather than review.
- [[kenn-software-suite]] — the rest of the suite.

## Repo

[github.com/kenn-io/roborev](https://github.com/kenn-io/roborev) — Go, MIT, 1.2k stars, v0.55.0 (95 releases). [roborev.io](https://roborev.io). Local-run, no hosted dependency.