Prlense — AI skill for Claude Code

prlense

**Understand a pull request before you approve it.** More and more pull requests are vibe-coded: written by a teammate with an AI assistant, or opened by a coding agent on its own. They can be large and hard to follow, and the description may not match what the code does. `prlense` lets a human step back in: it checks the work against the code itself and explains the reason behind the change, so you can tell whether it makes sense instead of scrolling through thousands of lines of diff. It works with any model that can call tools.

Each report is one HTML page in six steps: why the change was made, what changed, which data models and contracts changed, which functions changed, the detailed changes with links into the PR, and notes for the reviewer. It explains. It does not review, judge, or comment on other reviewers.

Quick start

1. Make sure you have git, Node 22 or newer, the GitHub CLI logged in (`gh auth status`), and a model to run: Claude Code logged in (`claude auth status`) works with no setup, or point prlense at any other model (see [Engines](#engines)).

2. Install `prlense`:

sh -c "$(curl -fsSL https://raw.githubusercontent.com/anvk/prlense/main/tools/install.sh)"

1. Inside a checkout of the repo the PR belongs to:

prlense doctor
prlense explain https://github.com/acme/shop/pull/482

A typical run takes a few minutes, and prints the turns and time it took at the end.

Install

You need git, Node 22 or newer, the GitHub CLI logged in (`gh auth status`), and a model: Claude Code logged in, or an API key (or a local server) for any engine in [Engines](#engines). With `curl`:

sh -c "$(curl -fsSL https://raw.githubusercontent.com/anvk/prlense/main/tools/install.sh)"
prlense doctor

or with `wget`:

sh -c "$(wget -qO- https://raw.githubusercontent.com/anvk/prlense/main/tools/install.sh)"

This clones the repository into `~/.prlense-cli`, builds it, and links the `prlense` command globally. Se