API v1

Use PrunePage from your coding agent

One request returns a page's Craft score and its findings as JSON. Free, you get the score, the areas, the three most important findings in full, the fix for the first one, and the rest by name. With a Fix Pack, the same request returns every finding in full with what to do about it, and lets your agent re-scan to check its own work.

Check a page

curl -s -X POST https://prunepage.com/api/v1/checks \
  -H 'content-type: application/json' \
  -d '{"url": "https://yourcompany.com"}'

The call waits for the result, which usually takes under a minute. If a review runs past 100 seconds you get "status": "running" and a status_url to poll. Send "wait": false to get that straight away.

What comes back

{
  "id": "wb7kc542",
  "status": "done",
  "url": "https://yourcompany.com/",
  "report_url": "https://prunepage.com/r/wb7kc542",
  "score": 42,
  "band": "Sloppy",
  "depth": "checks",
  "access": "free",
  "summary": { "scored": 7, "accessibility_and_performance": 1, "notes": 4 },
  "findings": [
    { "id": "…", "title": "No preview when shared", "area": "Details", "kind": "scored", "weight": "strong",
      "evidence": "No og:image, og:title or description.", "why": "A pasted link shows up as a bare URL…" },
    { "id": "…", "title": "Text too light to read", "area": "Accessibility", "kind": "accessibility" }
  ],
  "start_with": ["…", "…", "…"],
  "areas": { "Voice & copy": 86, "Visual identity": 74, "Credibility": 90, "Details": 74, "Accessibility": 93 },
  "unlock": { "message": "…", "url": "https://prunepage.com/r/wb7kc542#fixpack" }
}

kind says what a finding does: scored lowers the Craft score, accessibility and performance are scored in their own areas, and note or editor are reported without moving any number.
weight, on scored findings, is strong, clear or mild: how much it counts.
report_url is the same review as a page a person can read and share.

With a Fix Pack: fixes and re-scans

Free results carry evidence and why for the three findings in start_with, and the first of them also carries its fix and prompt as a sample. A Fix Pack covers one page for 30 days: send its token and every finding carries them, plus the fix and a prompt written for a coding agent. The result adds combined_prompt, rules_file, rescans_left and, after a re-scan, since_first_review with the score before and after.

curl -s -X POST https://prunepage.com/api/v1/checks \
  -H 'content-type: application/json' \
  -H 'Authorization: Bearer YOUR_FIX_PACK_TOKEN' \
  -d '{"url": "https://yourcompany.com"}'

With a token each call is a fresh review, including the AI editor's read, so an agent can fix, deploy, check again and see what moved. Your token is on your report, under “Your Fix Pack”.

Limits

Free: 3 checks a day per network, and one review per page every 7 days. Inside that week the same result comes back; a fresh look needs a Fix Pack.
Free checks run the checks only. The AI editor's read comes with reviews started on the website, and with a Fix Pack.
Public pages only, one page per check. A page that blocks automated visitors or returns an error isn't scored; the result says so.
A Fix Pack includes 30 re-scans of its page.

Paste this into your agent's instructions

For CLAUDE.md, AGENTS.md, Cursor rules or any agent that can run a command.

## Landing page check (PrunePage)

After you change the landing page and it is deployed, check the live page:

    curl -s -X POST https://prunepage.com/api/v1/checks \
      -H 'content-type: application/json' \
      -d '{"url": "<the live URL>"}'

The call waits for the result (usually under a minute) and returns JSON.
- Report "score" and "band" to me, and list the findings by title.
- Findings with "kind": "scored" lower the score; "weight" says how much. Fix those first, strongest first.
- Then fix "accessibility" and "performance" findings. Treat "note" findings as optional.
- Change only what a finding names. Give me the "report_url" when you're done.
- The findings listed in "start_with" carry "evidence" (what was seen on the page) and "why".
- The first of them also carries "fix" and "prompt", as a sample of what a Fix Pack adds to the rest.
- Free checks are limited to 3 a day, and a page reviewed in the last 7 days returns the same result.

The same description for machines is at /llms.txt. How the score is tested is on the method page.

Review a page in the browser