Extract page copy

Skill darrenbounds/extract-page-copy/skills/extract-page-copy

Use when evaluating, reviewing, or critiquing copy or messaging on an HTML file or URL. Also use when the user wants to read, assess, or discuss the text on a page without wading through markup.From its SKILL.md

Install

npx -y skills add darrenbounds/extract-page-copy --skill extract-page-copy

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.7 KB, 611 tokens by cl100k_base, as published. Nobody here has run it

Extract Page Copy

Strip all markup, scripts, and styles from an HTML file and output only the text a visitor would read — grouped by semantic role. This reduces token usage by up to 98% compared to passing raw HTML.

When to Use

User says "review the copy on this page" with an HTML path or URL
User asks "what does this page say" pointing to a file or webpage
You need to evaluate messaging before providing feedback — raw HTML wastes context

How to Extract

Read the HTML source (from a file or URL), then apply these rules in order:

Strip completely — do not include any text from:

<script>, <style>, <svg>, <template>, <noscript> elements and all their contents
HTML comments ()
Elements with inline display:none or visibility:hidden styles

Extract and group by role:

Meta description — <meta name="description" content="..."> value
Headings — all <h1> through <h6> text, always include regardless of length
Body copy — <p>, <li>, <td>, <th>, <blockquote> text (≥25 characters)
CTAs — <button> text and <a> link text (keep even when short)
Nav / Header / Footer — text inside <nav>, <header>, <footer> (shorter minimum, ≥10 characters)
Image alt text — alt attribute values on <img> tags (it's copy)

Deduplication: Skip any string already seen (case-insensitive exact match).

Output Format

Grouped mode (default — best for copy audits)

META DESCRIPTION
----------------
[meta description text]

HEADINGS
--------
[h1] ...
[h2] ...

BODY COPY
---------
...

CTAS
----
...

NAV / HEADER / FOOTER
---------------------
...

IMAGE ALT TEXT
--------------
...

Reading-order mode (best for evaluating narrative arc)

Output all extracted text in top-to-bottom document order, prefixed with its type:

[h1] ...
[p] ...
[cta] ...
[h2] ...

Limitations

Dynamically injected content (React, Angular, etc.) won't appear in static HTML — fetch the rendered DOM if needed
CSS-hidden elements controlled by external stylesheets are not detected; only inline styles are evaluated
Class-based nav patterns (<div class="nav">) are not grouped as nav — only semantic <nav>, <header>, <footer>

Quick Tip

For JavaScript-rendered pages, get the live DOM: open browser DevTools → Elements → right-click <html> → Copy → Copy outerHTML → save to a file, then extract from that.

What ships with it: 1 file

8.8 KB alongside SKILL.md, 1 of them executable

extract_copy.pyruns8.8 KB