Your agent wrote the code.
Know what you’re shipping.

A visual brief of what changed, what could break, and what still needs your judgment.

Read a real briefGet Brief · $19

Run locally or on every PR. Choose your model. Pay once.

Try a real pull request

nodejs/node #4174916 changed files

Adds a global fetch behind a flag

With --experimental-fetch, fetch, Request, Response and Headers exist as globals, served by a bundled copy of undici.

Risk score

20/100Lower

Provisional. Review priority, not failure probability.

Blast radius

5 observed areas

Bootstrap · undici · CLI options · Docs and tests · Other changed code & files

Downstream impact: unknown

Needs a human

2 review checks

Questions to resolve before merging.

Human checks
1·3fetch()1ReferenceError2undiciSocket2BeforeAfterretained stepremovedaddedStill presenthttp moduleFlag required
1·3fetch()12undiciReferenceErrorSocket2BeforeAfterretained stepremovedaddedStill presenthttp moduleFlag required

Start here

In lib/internal/bootstrap/pre_execution.js: Confirm the experimental warning prints once per process, not once per worker. Show a regression test and its result.

  1. Confirm the experimental warning prints once per process, not once per worker.

    lib/internal/bootstrap/pre_execution.js · lib/internal/main/worker_thread.js

  2. In test/parallel/test-fetch.mjs: Is one keep-alive-disabled test enough coverage for a new global. Show the options and evidence.

    test/parallel/test-fetch.mjs

Real PRs. Curated explanations. Scores derived from the stored diff.

One command. Or every PR.

Let GitHub run it

A new brief on each eligible PR update. Read it in the run summary; download the full report.

Run it when you need it

Analyze a GitHub PR, a commit, or a saved diff. Read the report. Copy a review prompt to your coding agent.

npm run analyze -- owner/repo#123

Anthropic · OpenAI · Ollama · MLX-LM · LM Studio

GitHub setup
  1. Add the kit in brief/ and copy brief/.github/workflows/brief.yml to your repository’s .github/workflows/ folder.
  2. Set BRIEF_PROVIDER and BRIEF_MODEL. Add your cloud provider’s API key as a repository secret, then set BRIEF_ENABLED to true.
  3. Merge the setup. Each subsequent PR update produces HTML, SVG, and JSON artifacts. Forks and drafts are skipped.

Node.js 22+. Set BRIEF_PROVIDER and BRIEF_MODEL; use BRIEF_BASE_URL for compatible endpoints. Local servers need a runner that can reach them.

Saved diffs & repository context

Brief reads related repository files where available. It shows which files it inspected. It never runs your code or tests.

npm run analyze -- --diff change.patch --repo .

Use a saved diff with committed local repository context. Add --context diff to review only the diff.

Brief: a sculptural review kit with connected change blocks
Digital download via Lemon Squeezy.

Keep the speed. Understand the change.

The working tool and its source, ready to make your own.

  • Visual reports & agent promptsRisk, blast radius, cited checks, and short prompts to act on.
  • Automation includedGitHub workflow, CLI, HTML, SVG, and JSON.
  • Editable source & examplesTypeScript, setup guide, and real PRs to explore.
$19

One-time · $39 before discount

Save $20 with BRIEF20. Applied at checkout.

Get Brief

Run it with your own cloud account or a local model.

A few things to know

What does Brief inspect?

The diff and a bounded selection of related files: callers, tests, and configuration where available. The report lists inspected files and gaps. Use a saved diff with --repo . for committed local context, or --context diff to skip repository context.

Which models can I use?

Choose Anthropic, OpenAI, Ollama, or an OpenAI-compatible server. Set BRIEF_PROVIDER and BRIEF_MODEL. MLX-LM and LM Studio have local presets; use BRIEF_BASE_URL for a different server address. The model needs to follow structured-output instructions.

Can I run it on my Apple Silicon Mac?

Yes. Start an MLX-LM, Ollama, or LM Studio server and point Brief at its endpoint. Brief connects to your model server; it does not install or manage the model.

Where does my code go?

Selected diff and repository content goes to your chosen provider. A local endpoint keeps that model call local. Reports stay on your computer or in your GitHub Actions outputs. Rival does not receive your code.

Does a low score mean it is safe to merge?

No. The score prioritizes review and shows its reasons. Brief reads code but never executes it or runs tests. Repository context can reveal more callers; it is not an exhaustive dependency audit.

What does it cost after buying?

Brief is a one-time purchase. It connects to your own model account or local server. Any cloud charges come directly from your chosen provider.

Can it run on every pull request?

Yes. The included GitHub Actions workflow produces a run summary and downloadable HTML, SVG, and JSON reports. Forks and drafts are skipped. A local model needs a runner with access to its server.

What am I buying?

The working CLI, local viewer, automation workflow, examples, and editable MIT-licensed source. Keep and modify the downloaded version. The example reports need no API key.

How do I hand a concern back to my coding agent?

Copy the short prompt beside a priority check. Paste it into your agent with the repository open. It gives the agent a specific question to investigate; you still decide whether the answer is enough to merge.