@postalsys/email-text-tools 1.4.1 → 2.0.1

package/index.js

@@ -5,15 +5,11 @@ const htmlToText = require('./lib/html-to-text');
 const inlineHtml = require('./lib/inline-html');
 const inlineText = require('./lib/inline-text');
 const mimeHtml = require('./lib/mime-html');
-const generateSummary = require('./lib/generate-summary');
-const riskAnalysis = require('./lib/risk-analysis');
 
 module.exports = {
     textToHtml,
     htmlToText,
     inlineHtml,
     inlineText,
-    mimeHtml,
-    generateSummary,
-    riskAnalysis
+    mimeHtml
 };

package/lib/html-to-text.js

@@ -6,7 +6,9 @@ const htmlToText = compile({
     selectors: [
         { selector: 'p', options: { leadingLineBreaks: 1, trailingLineBreaks: 1 } },
         { selector: 'pre', options: { leadingLineBreaks: 1, trailingLineBreaks: 1 } },
-        { selector: 'table.class#id', format: 'dataTable' }
+        { selector: 'table.class#id', format: 'dataTable' },
+        // ignore images
+        { selector: 'img', format: 'skip' }
     ],
     hideLinkHrefIfSameAsText: true
 });

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@postalsys/email-text-tools",
-    "version": "1.4.1",
+    "version": "2.0.1",
     "description": "Email text tools",
     "main": "index.js",
     "scripts": {
@@ -19,22 +19,18 @@
     "homepage": "https://github.com/postalsys/email-text-tools#readme",
     "devDependencies": {
         "eslint-config-nodemailer": "1.2.0",
-        "eslint-config-prettier": "8.8.0",
-        "mailparser": "3.6.4"
+        "eslint-config-prettier": "8.8.0"
     },
     "dependencies": {
-        "dompurify": "3.0.2",
-        "gpt3-tokenizer": "1.1.5",
-        "handlebars": "4.7.7",
+        "dompurify": "3.0.3",
         "he": "1.2.0",
         "html-to-text": "9.0.5",
         "js-beautify": "1.14.7",
-        "jsdom": "21.1.1",
+        "jsdom": "22.0.0",
         "juice": "9.0.0",
         "linkify-it": "4.0.1",
         "moment-timezone": "0.5.43",
         "node-html-parser": "6.1.5",
-        "tlds": "1.238.0",
-        "undici": "5.22.0"
+        "tlds": "1.238.0"
     }
 }

package/examples/risk.js

@@ -1,85 +0,0 @@
-'use strict';
-
-const fs = require('fs').promises;
-const { riskAnalysis } = require('..');
-const simpleParser = require('mailparser').simpleParser;
-const libmime = require('libmime');
-const Path = require('path');
-
-const BASE_PATH = process.argv[2] || '.';
-const MAX_ENTRIES = Math.abs(Number(process.argv[3]) || 0);
-
-async function analyzeEmail(path) {
-    const eml = await fs.readFile(path);
-
-    const parsed = await simpleParser(eml);
-
-    const result = await riskAnalysis(
-        {
-            headers: parsed.headerLines.map(header => libmime.decodeHeader(header.line)),
-            attachments: parsed.attachments,
-            html: parsed.html,
-            text: parsed.text
-        },
-        process.env.OPENAI_API_KEY,
-        {
-            //gptModel: 'gpt-3.5-turbo'
-            gptModel: 'gpt-4',
-            maxTokens: 6000
-        }
-    );
-
-    result.file = Path.basename(path);
-
-    console.log(JSON.stringify(result));
-}
-
-function shuffleArray(array) {
-    for (let i = array.length - 1; i > 0; i--) {
-        let j = Math.floor(Math.random() * (i + 1));
-        let temp = array[i];
-        array[i] = array[j];
-        array[j] = temp;
-    }
-    return array;
-}
-
-async function main() {
-    const stat = await fs.stat(BASE_PATH);
-    if (stat.isDirectory()) {
-        // process as directory
-
-        let processed = 0;
-        let list = await fs.readdir(BASE_PATH);
-
-        for (let file of shuffleArray(list)) {
-            if (Path.extname(file).toLowerCase() === '.eml') {
-                if (processed) {
-                    process.stdout.write(',');
-                }
-                try {
-                    await analyzeEmail(Path.join(BASE_PATH, file));
-                    console.error(`Processed ${file} [${processed + 1}]`);
-                } catch (err) {
-                    console.log(JSON.stringify({ file, error: err.message }));
-                    console.error(`Failed processing ${file} [${processed + 1}]`);
-                    console.error(err);
-                }
-                processed++;
-
-                if (MAX_ENTRIES && processed >= MAX_ENTRIES) {
-                    break;
-                }
-            }
-        }
-        console.error(`Processed ${processed} files from ${BASE_PATH}`);
-    } else {
-        // process as file
-        await analyzeEmail(BASE_PATH);
-    }
-}
-
-console.log('[');
-main().then(() => {
-    console.log(']');
-});

package/examples/summary.js

@@ -1,31 +0,0 @@
-'use strict';
-
-const fs = require('fs').promises;
-const generateSummary = require('../lib/generate-summary');
-const simpleParser = require('mailparser').simpleParser;
-const libmime = require('libmime');
-
-async function main() {
-    const eml = await fs.readFile(process.argv[2]);
-
-    const parsed = await simpleParser(eml);
-
-    const summary = await generateSummary(
-        {
-            headers: parsed.headerLines.map(header => libmime.decodeHeader(header.line)),
-            attachments: parsed.attachments,
-            html: parsed.html,
-            text: parsed.text
-        },
-        process.env.OPENAI_API_KEY,
-        {
-            //gptModel: 'gpt-3.5-turbo'
-            gptModel: 'gpt-4',
-            maxTokens: 6000
-        }
-    );
-
-    console.log(summary);
-}
-
-main();

package/lib/generate-summary.js

@@ -1,214 +0,0 @@
-'use strict';
-
-const packageData = require('../package.json');
-const htmlToText = require('../lib/html-to-text');
-const GPT3Tokenizer = require('gpt3-tokenizer').default;
-
-const { fetch: fetchCmd, Agent } = require('undici');
-const fetchAgent = new Agent({ connect: { timeout: 90 * 1000 } });
-
-const tokenizer = new GPT3Tokenizer({ type: 'gpt3' });
-
-const MAX_ALLOWED_TEXT_LENGTH = 32 * 1024;
-const MAX_ALLOWED_TOKENS = 4000;
-const OPENAI_API_URL = 'https://api.openai.com/v1/chat/completions';
-
-const SYSTEM_PROMPT = `
-I want you to act as an email client that processes emails and highlights important parts.
-You do not comment or explain anything.
-`
-    .replace(/\s+/g, ' ')
-    .trim();
-
-const USER_PROMPT = `
-Instructions:
-- You are an executive assistant scanning incoming emails to inform your superiors about what is important and what is not.
-- Your task is to monitor and analyze incoming emails, which consist of the message headers, a list of attachments, and text content
-- Describe the sentiment of the email using one word. Use either "positive", "neutral" or "negative". Include this value in the response as a "sentiment" property.
-- Generate a one-sentence summary of the email. Include this value in the response as a "summary" property.
-- Does it seem like the sender of the email would expect a reply to this email? Include this information in the response as a "shouldReply" property with the value "true" if they expect it and "false" if not.
-- If this email is a reply to a previous email, then extract the text content that only the email's sender wrote, and include this as a "replyText" property in the response.
-- Do not include message signatures in the extracted reply text
-- Respond with a JSON formatted structure. Do not write any other explanations.
-
-Facts:
-- An email consists of message headers, an attachments list, and text content
-- The email to analyze is formatted in a JSON format
-- An email is a reply to a previous email only if it includes a "in-reply-to" header
-- The email structure includes a property "headers" that contains an array of header values.
-- Each header contains two properties, "key" as the header field key name and "value" as the header value without the key prefix
-- The email structure includes a property "attachments" that contains an array of attachments.
-- Each attachment includes a "filename" property that describes the file name, a "contentType" property that describes the Content-Type value of the attachment
-- The email includes a "text" property for the text content
-
-Analyze the following email:
-`.trim();
-
-async function generateSummary(message, apiToken, opts) {
-    opts = opts || {};
-
-    let maxAllowedTokens = opts.maxTokens || MAX_ALLOWED_TOKENS;
-    let gptModel = opts.gptModel || 'gpt-3.5-turbo';
-
-    let text = message.text || '';
-    if (message.html && (!text || message.html.length > text.length * 2)) {
-        text = htmlToText(message.html);
-    }
-
-    let prompt;
-
-    const content = {
-        headers: [].concat(message.headers || []).filter(
-            // use a whitelist
-            header => ['from', 'to', 'cc', 'bcc', 'subject', 'mime-version', 'date', 'content-type', 'list-id'].includes(header.key)
-        ),
-        attachments: [].concat(message.attachments || []).map(attachment => ({ filename: attachment.filename, contentType: attachment.contentType }))
-    };
-
-    let charactersRemoved = 0;
-    let promptText = text;
-
-    if (promptText.length > MAX_ALLOWED_TEXT_LENGTH) {
-        charactersRemoved += promptText.length - MAX_ALLOWED_TEXT_LENGTH;
-        promptText = promptText.substr(0, MAX_ALLOWED_TEXT_LENGTH);
-    }
-
-    while (promptText.length) {
-        content.text = promptText;
-        prompt = `${USER_PROMPT}
-
-${JSON.stringify(content)}`;
-
-        let tokens = tokenizer.encode(prompt);
-        if (tokens.text.length <= maxAllowedTokens) {
-            break;
-        }
-        if (promptText.length > 2 * 1024 * 1024) {
-            promptText = promptText.substring(0, promptText.length - 1024 * 1024);
-            charactersRemoved += 1024 * 1024;
-        } else if (promptText.length > 2 * 1024) {
-            promptText = promptText.substring(0, promptText.length - 1024);
-            charactersRemoved += 1024;
-        } else if (promptText.length > 2 * 256) {
-            promptText = promptText.substring(0, promptText.length - 256);
-            charactersRemoved += 255;
-        } else if (promptText.length > 2 * 100) {
-            promptText = promptText.substring(0, promptText.length - 100);
-            charactersRemoved += 100;
-        } else if (promptText.length > 2 * 10) {
-            promptText = promptText.substring(0, promptText.length - 10);
-            charactersRemoved += 10;
-        } else if (promptText.length > 1) {
-            promptText = promptText.substring(0, promptText.length - 1);
-            charactersRemoved += 1;
-        } else {
-            throw new Error(`Prompt too long. Removed ${charactersRemoved} characters.`);
-        }
-    }
-
-    let headers = {
-        'User-Agent': `${packageData.name}/${packageData.version}`,
-        Authorization: `Bearer ${apiToken}`,
-        'Content-Type': 'application/json'
-    };
-
-    let payload = {
-        model: gptModel,
-        messages: [
-            {
-                role: 'system',
-                content: `${SYSTEM_PROMPT}`
-            },
-            {
-                role: 'user',
-                content: prompt
-            }
-        ]
-    };
-
-    let res;
-    let data;
-    let retries = 0;
-
-    let run = async () => {
-        res = await fetchCmd(OPENAI_API_URL, {
-            method: 'post',
-            headers,
-            body: JSON.stringify(payload),
-            dispatcher: fetchAgent
-        });
-
-        data = await res.json();
-
-        if (!res.ok) {
-            if (res.status === 429 && ++retries < 5) {
-                // try again
-                await new Promise(r => setTimeout(r, 1000));
-                return await run();
-            }
-
-            if (data && data.error) {
-                let error = new Error(data.error.message || data.error);
-                if (data.error.code) {
-                    error.code = data.error.code;
-                }
-
-                error.statusCode = res.status;
-                throw error;
-            }
-
-            let error = new Error('Failed to run API request');
-            error.statusCode = res.status;
-            throw error;
-        }
-
-        if (!data) {
-            throw new Error(`Failed to POST API request`);
-        }
-    };
-
-    const reqStartTime = Date.now();
-    await run();
-    const reqEndTime = Date.now();
-
-    let values;
-    const output =
-        data &&
-        data.choices &&
-        data.choices
-            .filter(msg => msg && msg.message && msg.message.role === 'assistant' && msg.message.content)
-            .sort((a, b) => ((a && a.index) || 0) - ((b && b.index) || 0))
-            .map(msg => msg.message.content)
-            .join('\n')
-            .trim()
-            .replace(/^\s*"|"\s*$/g, '')
-            .trim();
-
-    try {
-        values = JSON.parse(output);
-        if (typeof values.shouldReply === 'string') {
-            values.shouldReply = values.shouldReply === 'true';
-        }
-        if (typeof values.shouldReply !== 'undefined' && typeof values.shouldReply !== 'boolean') {
-            delete values.shouldReply;
-        }
-    } catch (err) {
-        let error = new Error('Failed to parse output from OpenAI API');
-        error.textContent = output;
-        throw error;
-    }
-
-    const response = Object.assign({ id: null, tokens: null, model: null }, values, {
-        id: data && data.id,
-        tokens: data && data.usage && data.usage.total_tokens,
-        model: gptModel,
-
-        _text: content.text,
-        _time: reqEndTime - reqStartTime,
-        _cr: charactersRemoved
-    });
-
-    return response;
-}
-
-module.exports = generateSummary;

package/lib/risk-analysis.js

@@ -1,240 +0,0 @@
-'use strict';
-
-const packageData = require('../package.json');
-const htmlToText = require('../lib/html-to-text');
-const GPT3Tokenizer = require('gpt3-tokenizer').default;
-
-const { fetch: fetchCmd, Agent } = require('undici');
-const fetchAgent = new Agent({ connect: { timeout: 90 * 1000 } });
-
-const tokenizer = new GPT3Tokenizer({ type: 'gpt3' });
-
-const MAX_ALLOWED_TEXT_LENGTH = 32 * 1024;
-const MAX_ALLOWED_TOKENS = 4000;
-const OPENAI_API_URL = 'https://api.openai.com/v1/chat/completions';
-
-const SYSTEM_PROMPT = `
-I want you to act as are an IT security expert that monitors emails for suspicious and risky activity.
-`
-    .replace(/\s+/g, ' ')
-    .trim();
-
-const USER_PROMPT = `
-Instructions:
-- You are an IT security expert scanning incoming emails to detect and block fraud attempts.
-- Your task is to monitor and analyze incoming emails, which consist of the message headers, a list of attachments, and text content
-- Your analysis should contain (but is not limited to) the following risk factors:
-  - Does the email include links with domain names that contain typos or homoglyphs that might mislead the user about the actual target of the link
-  - Does the sender address of the email not seem to match the persona or organization the sender email claims to be
-  - Does the email promise the user an extremely good financial outcome
-  - Does the email promise unclaimed money or goods
-  - Does the email offer a vague business opportunity with no specific details
-  - Does the email suggest there is a severe penalty if the user does not act as requested
-  - Does the email claim that there are technical issues with the user's email account
-  - Does the email offer services or activities that are not suitable for underage persons
-  - Does the sender's email look like a throwaway address
-  - Does the sender's email look like it might be spoofed
-  - Does the email contain attachments where the name of the attachment might be misleading and suggest a different file format than is actually used
-  - Does the email contain attachments that might be executable files
-- Provide a risk score for the email using the following scale: 1 - 5 (where 1 is low risk, and 5 is high risk), taking into account what may happen if a user acts by the instructions given in the email.
-- Provide a short single-sentence assessment text that includes details about issues that increase the risk score, do not disclose details that decrease the risk score or do not affect it.
-- Respond with a JSON formatted structure with a numeric risk score as "risk" property and the assessment as the "assessment" property. Do not write any other explanations.
-
-Facts:
-- An email consists of message headers, an attachments list, and text content
-- An email might contain an "authentication-results" result header that indicates if the message passed SPF, DKIM, ARC, and DMARC checks
-- Having a valid SPF, DKIM, and DMARC increases the chances that the sender domain name is not spoofed
-- For valid SPF, the "authentication-results" header must include "spf=pass"
-- For valid DKIM, the "authentication-results" header must include "dkim=pass"
-- For valid DMARC, the "authentication-results" header must include "dmarc=pass"
-- For valid ARC, the "authentication-results" header must include "arc=pass"
-- Throwaway email addresses might contain a word or a name and more than one number in the username of the email address
-- Throwaway email addresses might use a randomly generated or hex text string as the username of the email address
-- Throwaway email addresses might use free email services like gmail.com, outlook.com, hotmail.com, yahoo.com, aol.com, etc
-- The email to analyze is formatted in a JSON format
-- The email structure includes a property "headers" that contains an array of header values.
-- Each header contains two properties, "key" as the header field key name and "value" as the header value without the key prefix
-- The email structure includes a property "attachments" that contains an array of attachments.
-- Each attachment includes a "filename" property that describes the file name, "contentType" property that describes the Content-Type value of the attachment
-- The email includes a "text" property for the text content
-
-Analyze the following email:
-`.trim();
-
-async function riskAnalysis(message, apiToken, opts) {
-    opts = opts || {};
-
-    let maxAllowedTokens = opts.maxTokens || MAX_ALLOWED_TOKENS;
-    let gptModel = opts.gptModel || 'gpt-3.5-turbo';
-
-    let text = message.text || '';
-    if (message.html && (!text || message.html.length > text.length * 2)) {
-        text = htmlToText(message.html);
-    }
-
-    let prompt;
-
-    const headerSeen = new Set();
-    const content = {
-        headers: []
-            .concat(message.headers || [])
-            .filter(
-                // use a whitelist
-                header =>
-                    ['from', 'to', 'cc', 'bcc', 'subject', 'mime-version', 'authentication-results', 'date', 'content-type', 'list-id'].includes(header.key)
-            )
-            .filter(header => {
-                // only keept the latest
-                if (['authentication-results'].includes(header.key)) {
-                    if (headerSeen.has(header.key)) {
-                        return false;
-                    }
-                    headerSeen.add(header.key);
-                }
-                return true;
-            }),
-        attachments: [].concat(message.attachments || []).map(attachment => ({ filename: attachment.filename, contentType: attachment.contentType }))
-    };
-
-    let charactersRemoved = 0;
-    let promptText = text;
-
-    if (promptText.length > MAX_ALLOWED_TEXT_LENGTH) {
-        charactersRemoved += promptText.length - MAX_ALLOWED_TEXT_LENGTH;
-        promptText = promptText.substr(0, MAX_ALLOWED_TEXT_LENGTH);
-    }
-
-    while (promptText.length) {
-        content.text = promptText;
-        prompt = `${USER_PROMPT}
-
-${JSON.stringify(content)}`;
-
-        let tokens = tokenizer.encode(prompt);
-        if (tokens.text.length <= maxAllowedTokens) {
-            break;
-        }
-        if (promptText.length > 2 * 1024 * 1024) {
-            promptText = promptText.substring(0, promptText.length - 1024 * 1024);
-            charactersRemoved += 1024 * 1024;
-        } else if (promptText.length > 2 * 1024) {
-            promptText = promptText.substring(0, promptText.length - 1024);
-            charactersRemoved += 1024;
-        } else if (promptText.length > 2 * 256) {
-            promptText = promptText.substring(0, promptText.length - 256);
-            charactersRemoved += 255;
-        } else if (promptText.length > 2 * 100) {
-            promptText = promptText.substring(0, promptText.length - 100);
-            charactersRemoved += 100;
-        } else if (promptText.length > 2 * 10) {
-            promptText = promptText.substring(0, promptText.length - 10);
-            charactersRemoved += 10;
-        } else if (promptText.length > 1) {
-            promptText = promptText.substring(0, promptText.length - 1);
-            charactersRemoved += 1;
-        } else {
-            throw new Error(`Prompt too long. Removed ${charactersRemoved} characters.`);
-        }
-    }
-
-    let headers = {
-        'User-Agent': `${packageData.name}/${packageData.version}`,
-        Authorization: `Bearer ${apiToken}`,
-        'Content-Type': 'application/json'
-    };
-
-    let payload = {
-        model: gptModel,
-        messages: [
-            {
-                role: 'system',
-                content: `${SYSTEM_PROMPT}`
-            },
-            {
-                role: 'user',
-                content: prompt
-            }
-        ]
-    };
-
-    let res;
-    let data;
-    let retries = 0;
-
-    let run = async () => {
-        res = await fetchCmd(OPENAI_API_URL, {
-            method: 'post',
-            headers,
-            body: JSON.stringify(payload),
-            dispatcher: fetchAgent
-        });
-
-        data = await res.json();
-
-        if (!res.ok) {
-            if (res.status === 429 && ++retries < 5) {
-                // try again
-                await new Promise(r => setTimeout(r, 1000));
-                return await run();
-            }
-
-            if (data && data.error) {
-                let error = new Error(data.error.message || data.error);
-                if (data.error.code) {
-                    error.code = data.error.code;
-                }
-
-                error.statusCode = res.status;
-                throw error;
-            }
-
-            let error = new Error('Failed to run API request');
-            error.statusCode = res.status;
-            throw error;
-        }
-
-        if (!data) {
-            throw new Error(`Failed to POST API request`);
-        }
-    };
-
-    const reqStartTime = Date.now();
-    await run();
-    const reqEndTime = Date.now();
-
-    let values;
-    const output =
-        data &&
-        data.choices &&
-        data.choices
-            .filter(msg => msg && msg.message && msg.message.role === 'assistant' && msg.message.content)
-            .sort((a, b) => ((a && a.index) || 0) - ((b && b.index) || 0))
-            .map(msg => msg.message.content)
-            .join('\n')
-            .trim()
-            .replace(/^\s*"|"\s*$/g, '')
-            .trim();
-
-    try {
-        values = JSON.parse(output);
-        values.risk = Number(values.risk) || -1;
-    } catch (err) {
-        let error = new Error('Failed to parse output from OpenAI API');
-        error.textContent = output;
-        throw error;
-    }
-
-    const response = Object.assign({ id: null, tokens: null, model: null }, values, {
-        id: data && data.id,
-        tokens: data && data.usage && data.usage.total_tokens,
-        model: gptModel,
-
-        _text: content.text,
-        _time: reqEndTime - reqStartTime,
-        _cr: charactersRemoved
-    });
-
-    return response;
-}
-
-module.exports = riskAnalysis;