AI Scan Summary

radar summary takes a completed scan result and produces a 3-line executive summary using Claude Sonnet. The summary tells you what changed, what the biggest risks are, and what to do next --- in a format you can paste into Slack, email, or a status report.

Plan requirement: AI Scan Summary requires the Solo plan or higher.

How It Works

1. Run radar scan to produce a deterministic analysis result
2. Run radar summary to send the result metadata to Claude Sonnet
3. Claude returns a structured 3-line summary:
   • Line 1: What changed in this scan (new violations, fixed violations, score delta)
   • Line 2: Biggest risk (the most severe finding and its production impact)
   • Line 3: Recommended action (what to fix first and why)

Only the violation metadata is sent to the AI --- rule IDs, severities, categories, file paths, and debt scores. Source code is not included in the summary prompt.

Usage

# Generate a summary after scanning
radar scan .
radar summary

# Pipe scan results directly
radar scan . --format json | radar summary

# Specify output format
radar summary --format slack

The 3-Line Format

Every summary follows the same structure, regardless of output format:

[What changed]  — New violations, resolved violations, score movement
[Biggest risk]  — The most severe finding and its potential impact
[Action item]   — What to fix first, with file path and line number

Example output:

This scan found 4 new violations (+12 debt points) and 2 resolved (-10 points).
Net delta: +2 points. Score: 34 → 36.

Biggest risk: unbounded findMany() on the orders table (XL, 1.2M rows) in
src/orders/order.repository.ts:18 — this will OOM under production load.

Fix the unbounded query first (add take: 100 + cursor pagination), then
address the sync file read in order.service.ts:42. Both are blocking violations.

Output Formats

Format	Flag	Description
Text	--format text	Plain text (default). Human-readable paragraphs.
JSON	--format json	Structured JSON with whatChanged, biggestRisk, and action fields.
Markdown	--format markdown	Markdown with headers and code references. Suitable for PR descriptions or wiki pages.
Slack	--format slack	Slack Block Kit JSON. Paste directly into a Slack webhook or bot message. Includes severity emoji and formatted code blocks.
Email	--format email	HTML email body with inline styles. Suitable for automated email reports via SendGrid, SES, or any SMTP provider.

JSON Format Example

{
  "whatChanged": "4 new violations (+12 points), 2 resolved (-10 points). Net: +2.",
  "biggestRisk": "unbounded-query on orders table (XL) in src/orders/order.repository.ts:18",
  "action": "Add pagination to order.repository.ts:18, then fix sync-fs in order.service.ts:42",
  "score": {
    "before": 34,
    "after": 36,
    "delta": 2
  },
  "violations": {
    "new": 4,
    "resolved": 2,
    "total": 11
  }
}

Slack Format Example

{
  "blocks": [
    {
      "type": "header",
      "text": {
        "type": "plain_text",
        "text": "Radar Scan Summary — my-api"
      }
    },
    {
      "type": "section",
      "text": {
        "type": "mrkdwn",
        "text": "*What changed:* 4 new violations (+12 pts), 2 resolved (-10 pts)\n*Biggest risk:* :red_circle: unbounded query on `orders` (XL) — `order.repository.ts:18`\n*Action:* Add pagination to `order.repository.ts:18` first"
      }
    }
  ]
}

Cost

Each summary consumes 1 AI credit, approximately $0.005 per summary. The prompt is small (violation metadata only, no source code), so API costs are minimal.

Integration Examples

Slack Notification on Every PR

# .github/workflows/radar.yml
- name: Radar Scan
  run: radar scan . --format json > scan-result.json

- name: Post Summary to Slack
  run: |
    radar summary --format slack > slack-payload.json
    curl -X POST $SLACK_WEBHOOK_URL \
      -H 'Content-Type: application/json' \
      -d @slack-payload.json

Daily Email Report

# Cron job — daily scan summary via email
radar scan . --format json | radar summary --format email > report.html
# Send report.html via your email provider