# Overview

git-lrc is designed around one idea: review at the moment of commitment.

Instead of relying only on later review stages, it places an AI-assisted review decision near `git commit`, where developer intent and accountability are strongest.

## Core Concepts

### Commit-time guardrail

The check runs before commit lands. This catches problems while context is fresh and before risky code spreads.

### Review loop

Developers iterate:

1. Generate or edit code
2. Run review
3. Fix findings
4. Repeat

### Responsibility choices

- `review`: run AI review
- `vouch`: no AI run, explicit ownership
- `skip`: bypass review and ownership attestation

### Traceability in git log

Each commit can include status metadata:

- `ran (iter:N, coverage:X%)`
- `vouched (iter:N, coverage:X%)`
- `skipped`

This gives teams visibility into review behavior over time.

## Practical Difference vs IDE-only flows

IDE integrations are useful but tool-specific. Git-level integration gives one review control point regardless of editor preference across a team.

## Security-Oriented Framing

The model is aimed at high-impact diff risks such as:

- credential leakage
- sensitive logging
- behavior-changing logic removal
- expensive cloud operation insertion

For detailed risk categories, see [Threat Model](/content/livereview/git-lrc/docs/security/index.html).

Last updated on March 26, 2026.