v0.3

README.md

@@ -1,11 +1,106 @@
-# Horseman
+# 👋
 
-This repository is for tracking issues and feature requests for the Horseman crawler.
+This repository is for detailed update notes, and for tracking issues and feature requests for the [Horseman](https://gethorseman.app) crawler.
 
 https://github.com/workeffortwaste/horseman/issues
 
-👀
+# Horseman v0.3
 
-## Current Version
+**tl;dr** 🕷️ *Crawl the web with GPT3.5 and use page content with prompts. 🤖Don't know JS? Create snippets with an AI helper instead. 🌌 Deeper exploration with new the Insights feature. 🤯 Huge number of new snippets and much more.*
 
-The latest version is v0.2.4 for Windows, MacOS (M1 and Intel), and Linux.
+## Patch Notes
+
+Let's get the obligatory patch notes out the way. A lot of bugs have been squashed, elements polished, and the edges removed from many cases.
+
+## New Features
+
+With that out the way let’s talk about the big new features in Horseman v0.3.
+
+### OpenAI GPT 3.5 Integration
+
+GPT3.5 has been integrated directly in Horseman in a couple of ways. 👇
+
+### Prompts In Snippets
+
+****************Requires an OpenAI API key (add yours under the main Settings menu).****************
+
+Query GPT with any prompt by returning the new `prompt` property from your snippet. Combine any piece of page data, or send the entire page to GPT for analysis.
+
+Example snippets using page content have been added to the *New* button in the Editor. You’ll also find newly created built-in snippets in the library to rewrite meta descriptions, write missing meta descriptions, and summarize page content.
+
+```jsx
+/* horseman-config enable-openai */
+
+/* Generate a beautiful poem from the meta description using GPT */
+
+/* Fetch the meta description */
+const meta = document.querySelector('meta[name="description"]')?.getAttribute('content')
+
+/* Skip the snippet if the page doesn't have a meta description */
+if (!meta) return
+
+/* Ask gpt-3.5-turbo to generate a poem */
+return {
+    prompt: `Rewrite the following as a short 160 character poem: ${meta}`
+}
+```
+
+Further basic examples can be found in the `examples` folder in this repository.
+
+### No-Code AI Helper
+
+****************************Does not require an OpenAI API key.****************************
+
+Don’t know JavaScript? That’s no longer an issue. You’ll find a new button in the Editor, *No-Code AI*, which will use the power of GPT to automatically create simple Horseman snippets from a description.
+
+Entering a brief description such as `the social meta image` will instantly create a snippet fetching and returning the image used for social shares in the open graph data.
+
+## Insights
+
+A fully explorable report showing statistics about the number of failing or passing pages per snippet. *Think of it as your own customisable audit.*
+
+All built in snippets have been updated to support this feature (where it makes sense) and your own snippets can use this feature with ease. Any snippet returning a `pass` / `fail` string (or the new State property) can enable Insights with the new configuration helper or the `enable-insights` directive.
+
+## States
+
+You are now no longer limited to defining a snippet as passing by only returning a `pass` /  `fail` string, with the new state property of the return object you can show the state alongside the data in the results table.
+
+```jsx
+/* horseman-config enable-state */
+
+/* Using the new state property to send a fail state if num is more than 1 */
+const num = 2
+
+return {
+	cell: num
+	state: num > 1 ? 'fail' : 'pass'
+}
+```
+
+## Deeper Explorations
+
+Explore additional data more easily by viewing it per snippet instead of just by page. Hover over the column header for a snippet with explore enabled to find the new option.
+
+What’s more, when in the additional table, hovering over the column headers you’ll then be able to filter the results by unique values.
+
+## Configuration Helper
+
+As the features of Horseman have grown trying to remember all the snippet directives can be a little tricky. A new configuration helper will toggle your `horseman-config` directives with no fuss.
+
+## Updated Chrome
+
+The version of Chromium used for crawling has been updated to v111 for a faster and more robust crawl. *My favourite new addition?* You can now use the `:has` selector in your snippets for more advanced element selection!
+
+## More Data Types
+
+Automatically split up CSV strings into tags with the new `data-type-tag`.
+
+Visualise colours directly in your crawl with `data-type-color` return any valid CSS colour value and render it as a coloured tag in the results table.
+
+## New Snippets
+
+*Over 40 new snippets have been added to the library!* Including ecommerce snippets which extract product information straight from the DataLayer, AI enhanced snippets to rewrite and analyse content, cookie consent platform discovery, and much more.
+
+## Updated Website
+
+I've given the website a small refresh as the tool get much closer to v1.0, check it out and download the latest version now. https://gethorseman.app/

examples/gpt__prompt__content_extraction.js

@@ -0,0 +1,25 @@
+/* horseman-config enable-openai, enable-external */
+
+/**
+ * This is an example of intelligently extracting the text content from a page
+ * (with readability.js) and using it with a prompt.
+*/
+
+/* Import the readability content extraction package from skypsack */
+const pkg = (await import('https://cdn.skypack.dev/@mozilla/readability')).default
+
+/* Create a clone of the document for readability to modify */
+const documentClone = document.cloneNode(true);
+
+/* Remove any elements from the clone that we definitely don't want to extract content from */
+['header', 'footer', 'nav'].forEach(selector => {
+  documentClone.querySelectorAll(selector).forEach(element => { element.remove() })
+})
+
+/* Extract the page content from the clone using readability */
+const content = (new pkg.Readability(documentClone).parse()).textContent
+
+/* Ask gpt-3.5-turbo to summarize the content */
+return {
+  prompt: \`Summarize the following content: \${ content }\`
+}

examples/gpt__prompt__raw_html.js

@@ -0,0 +1,13 @@
+/* horseman-config enable-openai */
+
+/**
+ * This is a simple example of using the raw HTML content with a prompt.
+*/
+
+/* Fetch the raw HTML */
+const html = document.documentElement.outerHTML
+
+/* Ask gpt-3.5-turbo to generate a new description */
+return {
+    prompt: \`Do something with the following HTML: \${ html } \`
+}

examples/gpt__prompt__raw_text.js

@@ -0,0 +1,13 @@
+/* horseman-config enable-openai */
+
+/**
+ * This is a simple example of using the raw text content with a prompt.
+*/
+
+/* Fetch the raw text */
+const html = document.body.innerText
+
+/* Ask gpt-3.5-turbo to generate a new description */
+return {
+    prompt: \`Do something with the following text: \${ html } \`
+}

examples/gpt__prompt__single_element.js

@@ -0,0 +1,16 @@
+/* horseman-config enable-openai */
+
+/**
+ * This is a simple example of using the content from a single HTML element with a prompt.
+*/
+
+/* Fetch the meta description */
+const meta = document.querySelector('meta[name="description"]')?.getAttribute('content')
+
+/* Skip the snippet if the page doesn't have a meta description */
+if (!meta) return
+
+/* Ask gpt-3.5-turbo to generate a new description */
+return {
+    prompt: \`Rewrite the following as a short 160 character poem: \${ meta } \`
+}