Dev tools

PR Lens draws every pull request as an architecture and data-flow diagram, and reached 428 stars in 18 days

September 7, 2026 at 2:20 PM PT

coldteadotai/pr-lens repository page on GitHub

Image: GitHub

Why it mattersReviewers get a picture of what a change touches before they read the diff, and the coding agent that wrote the change can draw its own diagram, so review starts with a map rather than a wall of unified patches.

Coldtea has open-sourced PR Lens, a GitHub App that posts a sticky comment on every pull request showing an architecture diagram of what the change touches, an animated data-flow walkthrough of the new path, and nested drill-down diagrams for the parts of the change that would otherwise get lost in a long diff. The repository has 428 stars on GitHub, was created on 20 August, and has been pushed to as recently as today. It is MIT licensed.

What the comment carries

The one comment on the pull request holds three things. The architecture view draws the components involved in the change and the calls between them, with green for new nodes, amber for changed nodes and red for anything the change removes. The data-flow view is an animated sequence: one dot crosses one arrow at a time, in the order the steps run. Nested <details> blocks then split the change into scoped views, so a reviewer can jump from the whole blast radius to the new path alone or to the code that is being retired.

Every diagram ships as a light and dark pair, and GitHub picks the one that matches the reader's theme. There is also an interactive canvas at prlens.dev with pan, zoom and a walkthrough player, linked from the comment.

Four ways to run it, one contract

The GitHub App is the intended path for most teams: install it on a repository, open a pull request, the comment appears and updates on every push. The App uses Coldtea's model key and is free for open-source projects.

Two other modes cover teams that want their own key. A GitHub Action does the same analysis inside the customer's CI with any Gemini, OpenAI or OpenAI-compatible endpoint. A CLI runs each step on the developer's machine and can render the diagram before the pull request exists, which the README frames as the shape of reviewing an agent's work: while the reviewer reads the diff, the agent that wrote it renders the same picture.

The fourth mode is a coding-agent skill, installed with npx skills add coldteadotai/pr-lens. The agent writes the graph document itself against a validator, so the diagram is drawn using the model that is already open on the review, without a separate provider key.

Corrections to the map are kept in .github/pr-lens.yml rather than the generated SVGs. Renames, exclusions, lane pins and groupings sit in that overlay so they survive as the code moves and the model renames things between runs.

Where the traction sits

The Hall of Fame in the README shows the same renderer replayed on real historical pull requests, including React Hooks, the Node fetch implementation and the first Kubernetes Ingress type. Every diagram in the gallery is live on prlens.dev, drawn from the diff at the merge base of the original branch.

For a team letting a coding agent open pull requests, the case is that the reviewer's first look at a change should be a map of what moved. The Hall of Fame examples were legible when they landed because a human wrote them and could shape the diff so a reader would follow. Agent-written changes come without that care, and a diagram gives the reviewer somewhere to start.

Source

coldteadotai/pr-lens on GitHub, with the live gallery at prlens.dev. Star count and repository dates from the GitHub REST API, 2026-09-07.

Source: GitHub

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

More from AI News

Engrim ships a local SQLite memory store that survives switching between Claude Code, Cursor, Antigravity and Codex

Engrim is a local-first SQLite memory engine for coding agents that carries architectural decisions and project state across sessions and across models, from Claude Code to Cursor to Google Antigravity to Codex CLI.

Source: Hacker NewsDev tools

GameForgeBench evaluates coding agents on building and repairing games in Godot, Unity, Roblox, Minecraft, Unreal and the web

GamePhanes Studio has published GameForgeBench, a Harbor-compatible benchmark that judges coding agents on the full loop of importing, building, running, probing and repairing interactive games in real game engines.

Source: GitHubDev tools

video-talkcraft is an Agent Skill that turns Claude Code and Codex into an explainer-video studio, 766 stars in 15 days

video-talkcraft takes a voiceover script and a recorded voice track, aligns them at the character level, and renders a Remotion explainer video with 108 pre-tuned motion cards, kinetic captions, and a camera system that refuses to hold a static frame.

Source: GitHubDev tools