Implement summarization internals

This commit is contained in:
Guido Kraemer 2026-01-02 20:39:35 +01:00
parent f29d8337ad
commit 7595c70a38

View file

@ -58,7 +58,10 @@ import {
} from './js/mzta-utils.js';
import { taPromptUtils } from './js/mzta-utils-prompt.js';
import { mzta_specialCommand } from './js/mzta-special-commands.js';
import { getSpamFilterPrompt } from './js/mzta-prompts.js';
import {
getSpamFilterPrompt,
getSpecialPrompts
} from './js/mzta-prompts.js';
import { taSpamReport } from './js/mzta-spamreport.js';
import { taWorkingStatus } from './js/mzta-working-status.js';
import { addTags_getExclusionList, checkExcludedTag } from './js/mzta-addatags-exclusion-list.js';
@ -1017,54 +1020,63 @@ browser.menus.onClicked.addListener( (info, tab) => {
processEmails(getMessages(info.selectedMessages), _add_tags, _spamfilter);
}
if(_summarize) {
// info.selectedMessages is of type MessageList
summarizeEmails(getMessages(info.selectedMessages));
}
});
async function summarizeEmails(messages) {
taWorkingStatus.startWorking();
// generate the special prompt for summarization
// - make a long string of all the emails
let emails_content = '';
for await (let message of messages) {
////// TODO: make this customizable through placeholders!
// plan: make a second option field where placeholders can be used and
// simply add the result from that tepmlate to the end of the prompt option
// field
emails_content += "\n\n---\n\n";
let curr_fullMessage = await browser.messages.getFull(message.id);
// from
curr_fullMessage
// to
// cc
// bcc
// subject
// date
// attachment info
// body
let msg_text = getMailBody(curr_fullMessage)
let body_text = htmlBodyToPlainText(msg_text.html);
if( body_text.length == 0 ){
taLog.log("No HTML found in the message body, using plain text...");
body_text = msg_text.text.replace(/\s+/g, ' ').trim();
};
emails_content += body_text;
};
emails_content += "\n\n---\n\n";
// - join prompt and emails
// we have two prompts, the actual assignment for the LLM and the email
// template prompt.
const specialPrompts = await getSpecialPrompts();
const prompt = specialPrompts.find((prompt) => prompt.id === 'prompt_summarize');
const prompt_email = specialPrompts.find((prompt) => prompt.id === 'prompt_summarize_email_template');
const tabs = await browser.tabs.query({ active: true, currentWindow: true });
const chatgpt_lang = await taPromptUtils.getDefaultLang(prompt);
// assemble all email messages into one string and add the assignment prompt
let messages_list = [];
for await (let curr_message of messages) {
// extract body of current message as text
let curr_message_full = await browser.messages.getFull(curr_message.id);
let curr_body_full_html = getMailBody(curr_message_full);
let curr_body_full_text = htmlBodyToPlainText(curr_body_full_html.html);
if( curr_body_full_text.length === 0) {
taLog.log("No HTML found in the message body, using plain text...");
curr_body_full_text = curr_message_full.text;
}
messages_list.push(await taPromptUtils.preparePrompt({
curr_prompt: prompt_email,
curr_message: curr_message,
chatgpt_lang: chatgpt_lang,
body_text: curr_body_full_text,
subject_text: curr_message_full.headers.subject,
msg_text: curr_body_full_html,
}));
};
let messages_string = messages_list.join("\n\n--- Next Email ---\n\n");
let full_prompt = prompt.text + "\n\n\n" + messages_string;
console.log(full_prompt);
// send the prompt to the chat interface
openChatGPT(
full_prompt,
prompt.action,
tabs[0].id,
prompt.name,
prompt.need_custom_text,
prompt
)
taWorkingStatus.stopWorking();
// send to LLM
console.log(emails_content);
return {ok : '1'};
}