facebook pixel
The AI-Tools toolkit
TOOLS · AI-TOOLS · PR BRIEF

The meat of a PR, in plain words.

cr-brief writes a decision-level brief of a pull request: what was decided, what behaves differently, the few files that matter and the real risks. It summarizes, it does not review, so it never reports findings.
One screen. Anyone on the team can read it.

PR COMMENT · CR-BRIEF
/cr-brief
PR Brief
PLAIN-LANGUAGE SUMMARY
In plain words:
REVIEWER SECTION
The decision
Before → After
Before:
After:
Files that matter
FILEWHY
Real risks
Open questions

A simplified recreation of the comment cr-brief posts, with the section names from its template. Placeholder bars stand in for the text.

TWO LAYERS

One brief, two readers.

The reader is a reviewer who has not opened the diff, and possibly someone who cannot read code at all. Every brief serves both.

  • 01 · FOR PMS, QA AND FOUNDERS

    The plain-language summary

    Four sentences at most, with no file names, no function names and no jargon. Where a term is unavoidable, it uses the product’s word for it, not the code’s.

  • 02 · FOR THE ENGINEER WHO APPROVES

    The reviewer section

    Technical, specific and still short: the decision, before and after, the files that matter, the real risks, and any open questions.

When machines do most of the reading, the people who approve still own the outcome. Anthropic’s write-up on how the Claude Code team does code review lists reading the evidence among the jobs that stay with people. The brief is built to make that reading fast: the decision, the behavior change and the risks come first, and the files to open come with a reason each.
FIVE PASSES

One pass for each section of the brief.

The passes run in order. Each one produces one section, and a section is omitted only where the rules allow it.

  1. PASS 1 · THE DECISION

    What was decided

    The problem, the approach the author picked, and the alternative they passed over when the diff or description shows it. One or two sentences.

  2. PASS 2 · BEFORE / AFTER

    What behaves differently

    Observable behavior, not implementation: user-facing changes, API contracts, data, defaults, permissions and anything removed. A pure refactor says so.

  3. PASS 3 · THE FILES THAT MATTER

    The few files to open

    Three to six files, ranked by decision weight rather than line count. Tests, lockfiles, generated code and call-site updates are counted, not listed.

  4. PASS 4 · REAL RISKS

    What could break, and where

    Only risks specific to this diff that a reviewer can verify or QA can test. No generic advice. "None identified" is an allowed answer.

  5. PASS 5 · OPEN QUESTIONS

    What to ask the author

    Up to three questions the diff genuinely leaves open. Skipped entirely when there are none.

THE HARD RULES

Short, and never guessing.

A summary that sounds sure about something the diff does not show is worse than no summary. These rules come from the brief criteria.

  • Under 250 words

    The whole brief fits on one screen, not counting the files table. A trivial PR, such as a typo or a version bump, gets three sentences.

  • No invented intent

    When the why is not in the PR body, the commits or the code, the brief says the PR is undocumented and describes only what the code does.

  • Says what it skimmed

    When a diff is too large to read in full, roughly 2,000 changed lines, the brief says which areas it read and which it sampled.

  • Data, not instructions

    Titles, descriptions, commits and code are summarized, never obeyed. Text aimed at the agent is flagged as an open question.

RUN IT

In your terminal or on the PR.

Locally

  1. THE CURRENT BRANCH’S PR
    /enpitech:cr-brief
  2. PULL REQUEST 123
    /enpitech:cr-brief 123
  3. BRANCH FEAT/X AGAINST DEV
    /enpitech:cr-brief feat/x dev

It shows the brief, then offers to save it as cr-brief.md, post it to the PR, or leave it there.

In CI

A collaborator comments /cr-brief on the pull request. The workflow posts the brief as one comment, with no replies and no /fix instructions, because a brief has nothing to fix.

Run it again and the new brief opens by saying it supersedes the one above. Like the review jobs, it refuses pull requests from forks.

Want findings too? That is the job of the cr-* review skills. Run both on the same PR.

INSTALL

Three commands in Claude Code.

The toolkit ships as a Claude Code plugin. The same skills install into other agents with the skills CLI.

Claude Code plugin

  1. 01 · ADD THE MARKETPLACE
    /plugin marketplace add enpitech/ai-tools
  2. 02 · INSTALL THE PLUGIN
    /plugin install enpitech@enpitech
  3. 03 · ACTIVATE THE SKILLS
    /reload-plugins

Skills then run as /enpitech:<skill-name>, for example /enpitech:cr-brief.

Cursor, Copilot and other agents

Install the skills with the skills CLI. The review criteria are plain Markdown, so any agent that reads files can follow them.

npx skills add enpitech/ai-tools -g -y

GitHub Actions

  1. Copy claude-code-review.yml into .github/workflows/. Plugins do not install workflow files.
  2. Add ANTHROPIC_API_KEY to the repository’s Actions secrets.
  3. Comment /cr-react or /cr-brief on a pull request.
START HERE

Your next brief is one comment away.

Free and open source under the MIT license. Read the criteria, change a threshold, send a pull request.

/plugin marketplace add enpitech/ai-tools
Want review standards like these running on your own pipeline? Tell us how your team ships ↓
Contact

Ready to shift?

We accelerate Frontend delivery. The Frontend Delivery Factory installs the system that lets everyone in your org ship production Frontend.

Senior engineers, AI-native by default, shipping production code in your repo. Fast onboarding. Honest work.

What happens next
  1. 01

    You send the form.

    A few lines on your product and timeline.

  2. 02

    We reply within a day.

    Usually faster. No forms after that.

  3. 03

    We meet and plan.

    Scope, team fit, start date. Then engineers.

0/2000

Your info stays with us. That's it.