Team PR Radar

A lightweight PR review brief generator for teams.

It uses GitHub CLI search to find relevant open PRs, classifies their review state, optionally adds AI-powered action insights, and sends a concise brief to Google Chat.

No dashboard. No server. No database. Just a command.

What it does

Team PR Radar helps answer:

• Which PRs are waiting for our team review?
• Which of my PRs are still open and need attention?
• Which PRs I reviewed/commented on may need follow-up?
• What is the likely next action for each PR?

It supports:

• Team review queue via team-review-requested:<org>/<team>
• My open PRs via author:<me>
• Review follow-up via reviewed-by:<me> and commenter:<me>
• Stale/urgent detection
• CI state detection
• Draft PR filtering
• Optional AI insights using API keys, custom commands, or logged-in Cursor Agent
• Google Chat dry-run and real send modes

Requirements

• Node.js 20+
• GitHub CLI: gh
• Authenticated GitHub/GHE access:

gh auth login --hostname github.example.com

For Cursor Agent AI support, optional:

cursor agent login
cursor agent status

Quick start

# 1. Generate config and GitHub Actions workflow
npx team-pr-radar init

# 2. Edit config
vim config/pr-radar.yml

# 3. Test locally without sending Google Chat
PR_RADAR_DRY_RUN=true npx team-pr-radar

# 4. Send to Google Chat
GOOGLE_CHAT_WEBHOOK_URL='https://chat.googleapis.com/...' npx team-pr-radar

For local development from source:

PR_RADAR_DRY_RUN=true npx tsx scripts/cli.ts --config=config/pr-radar.yml

CLI usage

Team review queue

npx team-pr-radar

Uses config/pr-radar.yml and shows open PRs currently requested from the configured reviewer team(s).

Example GitHub search:

team-review-requested:acme/ui-reviewers

My open PRs

npx team-pr-radar --mine

Shows open PRs authored by the current GitHub user in the configured repos.

Example GitHub search:

author:<me>

This includes PRs that no longer have pending review requests.

My review follow-ups

npx team-pr-radar --follow-up

Finds open PRs you reviewed or commented on, then flags PRs that may need your attention again.

Current follow-up signals:

• New commits after your last review/comment
• Your previous CHANGES_REQUESTED review is still waiting on author action

Example GitHub searches:

reviewed-by:<me>
commenter:<me>

Custom config path

npx team-pr-radar --config=config/my-team.yml
npx team-pr-radar --mine --config=config/my-team.yml
npx team-pr-radar --follow-up --config=config/my-team.yml

--mine and --follow-up are separate modes and cannot be used together.

Configuration

Example config/pr-radar.yml:

github:
  host: github.example.com
  org: acme
  repos:
    - web-app

filters:
  reviewers:
    include:
      - ui-reviewers
    exclude: []

rules:
  ignore_draft: true
  stale_after_hours: 24
  urgent_after_hours: 48
  max_open_prs_per_repo: 100
  max_prs_in_brief: 8
  max_prs_to_ai_summarize: 5
  include_approved_waiting_merge: true

ai:
  enabled: true
  provider: auto
  model: auto

chat:
  title: Team PR Review Brief
  timezone: America/Los_Angeles

Reviewer filters

Bare reviewer names are treated as team names:

filters:
  reviewers:
    include:
      - ui-reviewers

Equivalent GitHub search:

team-review-requested:<org>/ui-reviewers

You can also be explicit:

filters:
  reviewers:
    include:
      - team:ui-reviewers
      - team:acme/ui-reviewers
      - user:some-login

PR states

| State | Meaning | |---|---| | 🟢 Ready for Review | Review has been requested | | 🟠 Needs Reviewer | No reviewer is currently requested | | 🟡 Re-Review Needed | Author pushed after changes were requested | | 🔴 Author Action Needed | Reviewer requested changes | | ✅ Approved | Approved and waiting for merge/maintainer action |

Additional flags:

• CI failure / pending state
• Stale PRs
• Urgent PRs
• Draft PRs ignored by default

AI insights

AI is optional. When enabled, Team PR Radar adds structured insights to the top PRs:

AI: High risk · Owner: reviewer · Confidence: 82%
Summary: Remove moment.js from automation-ui; replace with day-js shim.
Next: Reviewers prioritize stale review; author must fix failing CI before merge.

AI returns:

• summary — what the PR appears to change
• nextAction — what should happen next
• owner — author, reviewer, maintainer, CI, or unknown
• risk — low, medium, high, or unknown
• confidence — AI confidence score
• evidence — short reasoning hints

AI provider: auto

Recommended local config:

ai:
  enabled: true
  provider: auto
  model: auto

Auto mode tries, in order:

1. ANTHROPIC_API_KEY
2. OPENAI_API_KEY
3. AI_API_KEY with ai.base_url
4. AI_COMMAND or ai.command
5. Logged-in Cursor Agent

If Cursor Agent is logged in, no API key is required:

cursor agent status

The tool automatically uses:

cursor agent -p --mode ask --model auto --trust --output-format text

AI provider: command

Use any local AI CLI that reads prompt from stdin and writes answer to stdout:

ai:
  enabled: true
  provider: command
  command: "cursor agent -p --mode ask --model auto --trust --output-format text"

Or via environment variable:

AI_COMMAND="cursor agent -p --mode ask --model auto --trust --output-format text" \
PR_RADAR_DRY_RUN=true \
npx team-pr-radar

When AI runs, logs look like:

AI enabled: using Cursor Agent
Generating AI insights for 5 PR(s)...
AI 1/5: automation #34994
AI insights completed: 5 attempted, 5 succeeded, 0 failed.

Google Chat

Dry run:

PR_RADAR_DRY_RUN=true npx team-pr-radar

Send for real:

GOOGLE_CHAT_WEBHOOK_URL='https://chat.googleapis.com/...' npx team-pr-radar

GitHub Actions

npx team-pr-radar init generates .github/workflows/pr-radar.yml.

Example:

name: PR Radar

on:
  schedule:
    - cron: '0 10 * * 1-5'
  workflow_dispatch:

jobs:
  pr-radar:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Send PR brief
        run: npx team-pr-radar
        env:
          GOOGLE_CHAT_WEBHOOK_URL: ${{ secrets.GOOGLE_CHAT_WEBHOOK_URL }}
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

For GitHub Enterprise, ensure your workflow can authenticate to the configured host.

Development

npm install
npm run typecheck
npm run build

# dry-run message generation
npm run test:message

# source CLI dry-run
PR_RADAR_DRY_RUN=true npx tsx scripts/cli.ts --config=config/pr-radar.yml

Useful validation commands:

# Team queue
PR_RADAR_DRY_RUN=true npx tsx scripts/cli.ts --config=config/pr-radar.yml

# My open PRs
PR_RADAR_DRY_RUN=true npx tsx scripts/cli.ts --mine --config=config/pr-radar.yml

# My review follow-ups
PR_RADAR_DRY_RUN=true npx tsx scripts/cli.ts --follow-up --config=config/pr-radar.yml