Skip to content

dcubasoy/puppeteer-theater

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

theater

Updates: 07/20/2019

  • Fixed the year below (lol).
  • Added new base class, extensions for infinite scrolling, improved some of the internal anit bot detection mechanisms due to challenge from https://arh.antoinevastel.com/reports/stats/menu.html The wrapper still fails (and nothing to my knowledge chromium based) will also fail on his latest 'am i chrome headless?' page.

Updates: 07/14/2019

  • Added JSDoc for everything!
  • Generated new API documentation using an automated tool (thankfully...)
  • Added some issues/improvements I'd like to see made in the project
  • Added: 4 Shows to demonstrate a use-case where theater could be / was used for a very interesting purpose related to credential re-use attacks. Don't be stupid and don't be evil please.

NOTE: To check out the logging (which is another huge plus for this project), make sure to set your env.

  • Upload to S3: Bucket name is defined by: process.env.S3_BUCKET_PREFIX-theater-logs
  • Upload to Firebase/Google Cloud Storage: process.env.GCLOUD_BUCKET_PREFIX-theater-logs

Create these resources or you wont see anything except the console.

Bots I've added for demo/baseline purposes:

/bots/run/experian-extractor: Extracts credit report from Experian + score. /bots/run/creditkarma-extractor: Extracts credit report from CreditKarma account + score. /bots/run/facebook-signin: Signs in using credentials, just grabs the session for now. Adding data extraction/more layers at some point. /bots/run/paypal-signin : Signs in using credentials, extracts data from account, saves session for later use.

    "username": "1337gamemode",
    "password": "njsnsuhfuf",
    "userId": "1337gamemode.njsnsuhfuf"
}

I tend to follow this convention with all of them. POST that using a REST client and boom, liftoff.

I am a fan of firebase so all outgoing/egress stuff such as reporting account credentials/session data/etc is done using firestore. The code is pretty simple and configuring this to hook up to your own firebase should be as easy as replacing the credential file in the config/ folder.

Purpose & Inspiration

In a sentence, Theater automates anything and everything a human being is capable of performing on a site. It does not wait for navigations, its looking for a set of conditions that when evaluated and return true, execute some particular code.

On the highest level, it achieves this by dealing with units of work as: Shows & Scenes.

A show might describe an entire site, like "Capital One".

Within this show, your scene sets can play - for example: SignIn (for linking a user's capital one account using a bot), ExtractStatements (for extracting pdf statements from account).

Scenes describe how the page looks and you decide what the bot does. It's simple! ❤️ And dangerous- its perfect for account takeovers (or worse). Note: I cannot endorse violating any terms of use of anything legally speaking. But hypothetically, do no harm and call it a day.

Lets dive in. Normally, when working with puppeteer, you will find yourself repeatedly calling- waitForNavigation().

What if there were a way to simply provide the bot what it should see (on the page), and instruct it what to do when such conditions are met?

  • This is a one of many enhancements over plain pupepteer script theater offers.
  • Another one: EventEmitter. Consider you want to just:
  • on('botCreatedAccount', (o) => doSomething(o));

You can define any custom events you wish.

One line, just told us how to handle whenever our bot creates an account succesfully. Maybe we store the data, or spawn another bot to provision the account! I love this pattern EventEmitter provides.

Theater also offers extensions: powerful and easy to use tools that can solve problems in two lines of code like

  1. recaptcha challenges
  2. generic captchas
  3. clicking all the annoying pop ups that screw up your automation (.spinner)
  4. Infinite Scrolling
  5. Custom function evaluations in browser context
  6. Delaying for a random or specified portion of time, to not appear like a bot.

The Scene.Extensions portion could be made way more powerful and I welcome PRs.

Read the docs for a detailed description of the whole API.

Fully Compatible with both [email protected] & [email protected].

References that are useful when reading this and any puppeteer code:

WTF is this repo?

The repo is just a basic RESTful API for you to execute bots and understand how the framework operates internally. To that end, you'll want to make sure to: export DEBUG=theater*

Just do it.

How to just rip and run?

  • Clone the repo
  • nvm use 10.5.3 (install if not present)
  • npm install

export DEBUG=theater*

npm start

curl --request POST \
  --url http://localhost/bots/paypal-signin/run \
  --header 'content-type: application/json' \
  --data '{
    "username": "[email protected]",
    "password": "hackedyourshit",
    "userId": "[email protected]"
}

Browser will launch, console output will show theater inner workings.

# Base Classes

class: Show

Example (Bot will login to discover.com)

class DiscoverShow extends Show {}
DiscoverShow.Scenes = Show.scenes(path.join(__dirname, 'discover/'));

const bot = new PuppeteerBot({
   preferNonHeadless: true,
});
await bot.init(); // starts the browser
bot.page.goto('https://portal.discover.com/customersvcs/universalLogin/ac_main'); // navigates to our page

const show = new DiscoShow({
     Scenes: DiscoShow.SceneSets.SignIn,
     bot,
     logger, // optional, for custom logger instance
});

show.on('accountLinkedResult', async (o) => { await reporter.onAccountLinkedResult(o); }); // will emit when sign in succeeds

await show.play();

await bot.deinit();

new Show({ Scenes, bot, timeout })

  • Scenes <?[Array]<[Class]<[Scene]>>> An array of type Scene that might play. SceneSets (a basic example) can represent a particular workflow, like Sign In or Extract Reports.

  • bot <[PuppeteerBot]> A bot that show will play on. Wrapper on top of puppeteer. Goal is to remove this dependency soon.

  • timeout <[number]> Time for Show to give up matching Scene (ms). Defaults to 30000.

I am working on a method to improve the timeout feature using a CDP feature that can determine if a given request is driving the page's navigation, in which case the show should not end.

Show.scenes(dir)

  • dir <[string]> Directory to search for /\.scene\.js$/ files.

  • returns: <[Object]<[string], [Class]<[Scene]>>>

  • Enumerate files in dir, look for all valid Scene files and load.

show.result()

  • returns: TBD

TBD. To return some result.

show.scene(TargetScene)

  • TargetScene <[Class]<[Scene]>>

  • returns: <[Promise]<[Scene]>>

If TargetScene specified, returns instance of that Scene for this show.

If no TargetScene specified, this method will returns matching Scene.

show.play()

  • returns: <[Promise]>

This method will play the show. Iterating through out all the Scenes and play()ing those with match()es.

class: Scene

This example scene will simply click #session_btn_continue. In the first example using Scene.Extensions.Click() and in the second using the basic approach.

class ClickExtendScene extends Scene {
  constructor(args) {
    super(Object.assign({
      elementQueries: {
        extendSession: {
          selector: '#session_btn_continue',
        },
      },
      extensions: [new Scene.Extensions.Click()],
    }, args));
  }
}

class RequestSessionExtensionScene extends Scene {
  constructor(args) {
    super(Object.assign({
      elementQueries: {
        extendSession: {
          selector: '#session_btn_continue',
        },
      },
    }, args));
  }

  async play() {
    await super.play();
    await this.elements.extendSession.click();
  }
}

new Scene({ show, elementQueries, extensions, generic })

  • show <[Show]>

  • elementQueries <[Object]<[string], [PuppeteerBotElementQuery]>>

  • extensions <[Array]<[SceneExtensions]>>

  • generic <[boolean]> If this value sets to false, conditions should be specified to determine when curtain will fall (ending the show). Defaults to true

scene.context(key)

  • key <[string]>

returns contextVariable[key]

scene.interaction()

  • returns: <[Interaction]>

scene.setContext(key, value)

  • key <[string]>

  • value <[*]>

  • sets var[key] = value

scene.match()

  • returns: <[Promise]<[boolean]>>

Criterion for determining whether this scene will match (returning true) in order:

  • Curtain has NOT fallen yet

  • PuppeteerBotElement all elementQueries validated

  • extensions[].match() all have returned true

scene.curtainFallen()

  • returns: <[Promise]<[boolean]>>

Check if scene finished playing; curtain fallen:

  • Not generic and no extensions[].curtainFallen() present in this scene.

scene.play()

  • returns: <[Promise]>

By default, this method will only call extensions[].play(). This is sort of confusing because it means if you are performing some work that needs to precede logic in play(), you need to be invoking the constructor await super.play() before you do your thing. Perfect example (found in the example shows) is captcha authentication. See the Docs.MD file for a detailed and complete reference.

class: PuppeteerBotElement

elementQueries: A series of selectors and conditionals that describe the state of the page that corresponds to the scene in which they lie.

This basically encapsulates the state of the page. Is this popup blocking the viewport? Is the deposit button visible and did we set the context to contain the coins? For one use-case I am actually doing this deposit/refill balance flow using [bitcoinjs-lib] (https://www.npmjs.com/package/bitcoinjs-lib).

query = {
  visibility: 'required'|'required:group-name'|'optional'|'forbidden',
  selector: '',
  visibilityAreaCheck: true|false,
}

This object (elementQueries) will be unique to a particular scene, when the scene is constructed it will target the results and map them into PuppeteerBotElement(s).

  • Don't mispell anything - your code will failure and you will spend hours tracking down the visblity that blocked you from proceeding.

  • If visibility is 'required', element will only match if that target element is visible. (Default)

  • 'required:group-name', match if one of element having same group-name is visible. You may specify as many as groups as you want.

  • 'optional', always match with or without target element visible.

  • 'forbidden', will only match if target element is not visible.

selector here is any valid CSS selector. CSS makes it easy to adjoin multiple selectors with the logical AND | , operator.

element.visibleElements()

  • returns: <[Promise]<[Array]<[Puppeteer.ElementHandle]>>>

element.visible()

  • returns: <[Promise]<[boolean]>>

element.getFrame()

  • Gets the content frame for a given element handle referencing iframe nodes
  • returns: <[Promise]<[Frame]>>

element.match()

  • Checks matching context (elementQuery)
  • returns: <[Promise]<[boolean]>>

element.innerText()

  • returns: <[Promise]<[string]>>

element.textContent()

  • returns: <[Promise]<[string]>>

element.attribute(opt)

  • opt will be evaluated to retrieve value of this attribute.

  • returns: <[Promise]<[string]>>

element.fill(opt)

  • opt will be passed to puppeteer-bot internally.

  • returns: <[Promise]<[boolean]>> whether element has been filled or not.

element.$select(opt)

  • opt will be passed to puppeteer-bot internally (dirty-select).

  • returns: <[Promise]<[boolean]>> whether element has been selected or not.

element.select(opt)

  • opt will be passed to puppeteer-bot interanlly (default-select).

  • returns: <[Promise]<[boolean]>> whether element has been selected or not.

element.check(checked)

  • checked <[boolean]> whether to check or uncheck.
  • returns: <[Promise]<[boolean]>> Whether element has been checked successfully or not.

element.click({ once } = {})

  • once <[boolean]> if true, will only click very first target element that matches. else, will click all the target elements that match.
  • returns: <[Promise]<[boolean]>> whether clicked or not.

element.screenshot()

  • returns a screenshot of the first element matching given elementQuery for given PupeeteerBotElement.
  • returns: <[Promise]<[Buffer]>> with screenshot file contents.

element.upload(file, opt = {})

  • Writes file to temporary directory on disk, uploads file on given elements inside this PuppeteerBotElement, cleans up.
  • returns: <[Promise]<[boolean]>> whether file uploaded or not.

element.tableContent(opts)

  • returns table as an array of JSON objects (good for scraping)
  • params: opts {} default arguments for tabletojson (refer to the documentation for this package)

element.eval(fn, ...args)

Convenience function to run puppeteerBot.$$safeEval. First argument of fn will be array of target elements, and from second argument will be ...args.

Example Scene (1):

/**
 * Scene will play IFF:
 * - h1 exists
 * - #clickMeFirst or #clickMeNext exists
 * - .spinner is hidden

* Scene play:
 * - if any of h1 element match /click me/,
 * - click all #clickMeFirst if clickable
 * - click all #clickMeSecond if clickable
 */
class TestScene extends Scene {
  constructor() {
    super({ elementQueries: {
      title: {
        visibility: 'required',
        selector: 'h1',
      },
      button1: {
        visibility: 'required:buttons',
        selector: '#clickMeFirst',
      },
      button2: {
        visibility: 'required:buttons',
        selector: '#clickMeNext',
      },
      spinner: {
        visibility: 'forbidden',
        selector: '.spinner',
        visibilityAreaCheck: true,
      },
    }});
  }

  async play() {
    await super.play();
    if (/click me/.test(await this.elements.title.innerText())) {
      this.elements.button1.click();
      this.elements.button2.click();
    }
  }
}

Example Scene (2):

/**
 * Scene will play IFF:
 * - username, password, loginBtn are visible

* Scene play:
 * - check if errors visible, if visible => log details, prevent continous play of this scene when errors are visible
 * - fill: #userid-content,#userid
 * - fill: #password-content,#password
 * - check: #id-checkbox-content,#id-checkbox
 * - click: #log-in-button
 */
class TestScene2 extends Scene {
 constructor(args) {
    super(Object.assign({
    elementQueries: {
      errors: {
        selector: '#info-err-msg',
        visibility: 'optional',
        visibilityAreaCheck: true,
      },
      username: {
        selector: '#userid-content,#userid',
      },
      password: {
        selector: '#password-content,#password',
      },
      rememberMe: {
        selector: '#id-checkbox-content,#id-checkbox',
        visibility: 'optional',
      },
      loginBtn: {
        selector: '#log-in-button',
      },
    }});
  }

  async play() {
   await super.play();
    this.setContinousPlayLimit('errors', 1);

    if (await this.elements.errors.visible()) {
      const errorMessage = await this.elements.errors.innerText();
      this.log('login-error-message: ', errorMessage);
    }

    await this.elements.username.fill(this.context('username'));
    await this.elements.password.fill(this.context('password'));
    await this.elements.rememberMe.check(true);

    await this.elements.loginBtn.click();
  }
}

Scene.Extensions

class: Scene.Extensions.Click

new Scene.Extensions.Click(opt)

  • opt <[string]> Simply click element by name specified by opt. If left empty, will click all elements.

class: Scene.Extensions.Delay

new Scene.Extensions.Delay(ms)

  • ms <[number]> Giving a delay for ms (ms). If ms left empty, value will be a random number between 2000-7500.

class: Scene.Extensions.Fork

new Scene.Extensions.Fork(forks)

  • forks <[Array]> fork configurations (managing when a scene itself can be forked into multiple options.
new Scene.Extensions.Fork([

{

fork: async() => this.elements.manageStatements.click(),

Scenes: [StatementsSummaryScene, StatementsDetailScene],

},

{

fork: async() => this.elements.accountDetails.click(),

Scenes: [AccountProfileInformationScene, AccountContactInformationScene],

},

]);

Use this extension if Scene can be forked into multiple options. If a given Scene has been matched, a given fork that has Scene whose curtain has not fallen will play via fork.fork().

However, if all Scenes' curtain have fallen (show is over), this extension's curtain will also fall.

class: Scene.Extensions.PreventCurtainFall

new Scene.Extension.PreventCurtainFall({ playCount = 1 })

  • playCount <[number]> curtain will not fall until this extension played for playCount times. If not specified, playCount will default to 1. This extension is really just a stop-gap for development- as it is sometimes not clear whether or not the bot should continue working. For development, I recommend using it liberally.

class: Scene.Extensions.Captcha

new Scene.Extension.Captcha(targetElementName, targetAnswerElementName)

  • targetElementName is element name of captcha image to solve, targetAnswerElementName is element name of captcha solution input field. This extension will solve the captcha and fill the solution into targetAnswerElementName. This extension will decide whether to: screenshot element on DOM, get src attribute containing base64 image, or use rp to get image from url.

class: Scene.Extensions.ReCAPTCHAv2

new Scene.Extension.ReCAPTCHAv2(targetElementName, siteKeyFn)

  • targetElementName is element name wherein #g-recaptcha-response is contained in child nodes of this element. Most often seen as .g-recaptcha. This extension will determine site-key from provided siteKeyFn(), solve recaptcha, use getFrame() if needed to properly set response, and invoke the callback function (if present) to trigger the result after captcha has been solved.

class: Scene.Extensions.Scroll

new Scene.Extension.Scroll(time, repeats)

  • Simply Scrolls for the given time (in ms), and repeats scrollRepeats (default = 5) times. Designed for infinite scrolling scenarios.

Core Components

Some of the syntax within Theater can seem intimidating but it is pretty simple once you understand the underlying components.

  • promise-condition: supports or, and, not, strictEqual nested evaluations for promises. Extremely useful, as illustrated in two examples.

Here the scene will match() when either challenge or header is visible(). Other puppeteer-bot-element queries will have no impact since we have not invoked super.match() inside our conditional. The match() after shows how that would look.

 async match() {
    return PromiseCondition.or(
      this.elements.challenge.visible(),
      this.elements.header.visible(),
    );
  }

async match() {
    return PromiseCondition.and(
	  super.match(), // now the first is no longer needed
  }

Here the scene will match() when all puppeteer-bot-element queries have validated, and context signedIn has been set to true.

  async match() {
    return PromiseCondition.and(
      this.context('signedIn'),
      super.match(),
    );
  }

Here is a more complicated example.

async match() {
    return PromiseCondition.and(
      super.match(),
      PromiseCondition.or(
        this.elements.errorMessages.visible(),
        ...Object.keys(this.elements).filter(k => /^input/.test(k)).map(e => PromiseCondition.and(
          this.elements[e].visible(),
          PromiseCondition.not(this.elements[e].value()),
        )),
      ),
    );
  }
  • Interaction: At some point you may find yourself needing to provide some information a bot after it has already begun running (anytime you require input after the bot initializes). Perhaps you need to prompt for a security question's answer, code, etc. Interaction is a very simple class that uses Redis to maintain a real-time interaction link between the user and the bot.

Credits to @vhain, who wrote this code.

You will find a basic React boilerplate that demonstrates how this works (see classes/theater/shows/jokerstash) for an example of interaction being used to communicate with a user to search for their information.

Common Design Patterns

When working with complex automation tasks, I've found setting up a few boilerplate Scenes before I do anything else saves me a lot of time. These scenes are:

  • JustClickAwareScene (exports JustClickAwareScene.WithSpinner, JustClickAwareScene.WithoutSpinner

  • JustClickScene: will just-click any selector specified - perfect for closing annoying modals, CTAs, extending session, etc

  • SpinnerAwareScene: (will wait for any spinners/loading modals to dissapear before continuing toplay() the Show.

You can find some real-examples of this in the classes/theater/shows folder.

Test Driven Development

These modular nature of Theater lends itself well to a TDD approach - for each particular Scene just grab the html for that page and add a test assertion in show-tests/example folder where example is your Show name.

Run node index.js after running cd classes/theater/show-tests/. This is my primitive test runner.

Contribution guidelines

This is my first real open-source project that I'll be maintaining, I'd love contributions, questions, criticism, or guidance!

Contact

https://webscrapers.slack.com for a prompt response, message me here!

https://keybase.io/nicomee

About

automation framework powered by headless chrome

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published