# PrunePage

> PrunePage reviews one public web page and returns a Craft score (0 to 100) with findings: how much of the page still looks the way a starter template or an AI site builder leaves it, plus accessibility and performance problems. It does not try to detect whether AI was used.

## API for agents

Check a page (the call waits for the result, usually under a minute):

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

- Result: JSON with `score`, `band`, `points_lost`, `areas`, `report_url`, `summary` and `findings`. `points_lost` is where the score went, by area: 100 minus those numbers is `score`. `areas` has each area's own 0 to 100 figure; Accessibility and Performance are scored on their own, outside `score`. `start_with` lists the three most important finding ids. Each finding has `id`, `title`, `area` and `kind`: `scored` (lowers the score; has `weight` strong, clear or mild), `accessibility`, `performance`, `note` or `editor`.
- If the review is still running after 100 seconds the response is `{"status": "running", "status_url": "..."}` with HTTP 202. Poll `GET status_url`, or add `?wait=1` to wait again.
- A failed review returns `{"status": "failed", "error": "..."}`. Pages that block automated visitors or return an error are not scored.
- Free: 3 checks a day per network; one review per page every 7 days (the same result comes back inside that week); checks only (no AI editor's read). The three `start_with` findings carry `evidence` (what was seen on the page) and `why`, and the first also carries `fix` and `prompt` as a sample; the rest are names only.
- With a Fix Pack token (`Authorization: Bearer <token>`, one page, 30 days, 30 re-scans): every finding carries `evidence`, `why`, `fix` and `prompt`; the result adds `combined_prompt`, `rules_file`, `rescans_left` and `since_first_review`.

Suggested loop with a token: check, fix scored findings strongest first using each finding's `prompt`, deploy, check again, compare `score`, stop when no scored findings remain.

## Pages

- [Use PrunePage from your coding agent](https://prunepage.com/agents): the API in full, with a snippet for agent instructions
- [How the Craft score works](https://prunepage.com/method): what the score measures, how it is tested, its limits
- [Home](https://prunepage.com/): review a page in the browser