mirror of
https://github.com/ZeppelinBot/Zeppelin.git
synced 2025-07-06 02:27:19 +00:00
103 lines
3.3 KiB
TypeScript
103 lines
3.3 KiB
TypeScript
import escapeStringRegexp from "escape-string-regexp";
|
|
import z from "zod/v4";
|
|
import { normalizeText } from "../../../utils/normalizeText.js";
|
|
import { stripMarkdown } from "../../../utils/stripMarkdown.js";
|
|
import { getTextMatchPartialSummary } from "../functions/getTextMatchPartialSummary.js";
|
|
import { MatchableTextType, matchMultipleTextTypesOnMessage } from "../functions/matchMultipleTextTypesOnMessage.js";
|
|
import { automodTrigger } from "../helpers.js";
|
|
import { escapeInlineCode } from "discord.js";
|
|
|
|
interface MatchResultType {
|
|
word: string;
|
|
type: MatchableTextType;
|
|
}
|
|
|
|
const regexCache = new WeakMap<any, RegExp[]>();
|
|
|
|
const configSchema = z.strictObject({
|
|
words: z.array(z.string().max(2000)).max(1024),
|
|
case_sensitive: z.boolean().default(false),
|
|
only_full_words: z.boolean().default(true),
|
|
normalize: z.boolean().default(false),
|
|
loose_matching: z.boolean().default(false),
|
|
loose_matching_threshold: z.number().int().default(1),
|
|
strip_markdown: z.boolean().default(false),
|
|
match_messages: z.boolean().default(true),
|
|
match_embeds: z.boolean().default(false),
|
|
match_visible_names: z.boolean().default(false),
|
|
match_usernames: z.boolean().default(false),
|
|
match_nicknames: z.boolean().default(false),
|
|
match_custom_status: z.boolean().default(false),
|
|
});
|
|
|
|
export const MatchWordsTrigger = automodTrigger<MatchResultType>()({
|
|
configSchema,
|
|
async match({ pluginData, context, triggerConfig: trigger }) {
|
|
if (!context.message) {
|
|
return;
|
|
}
|
|
|
|
if (!regexCache.has(trigger)) {
|
|
const looseMatchingThreshold = Math.min(Math.max(trigger.loose_matching_threshold, 1), 64);
|
|
|
|
const patterns = trigger.words.map((word) => {
|
|
let pattern;
|
|
|
|
if (trigger.loose_matching) {
|
|
pattern = [...word].map((c) => escapeStringRegexp(c)).join(`[\\s\\-_.,!?]{0,${looseMatchingThreshold}}`);
|
|
} else {
|
|
pattern = escapeStringRegexp(word);
|
|
}
|
|
|
|
if (trigger.only_full_words) {
|
|
if (trigger.loose_matching) {
|
|
pattern = `\\b(?:${pattern})\\b`;
|
|
} else {
|
|
pattern = `\\b${pattern}\\b`;
|
|
}
|
|
}
|
|
|
|
return pattern;
|
|
});
|
|
|
|
const mergedRegex = new RegExp(patterns.map((p) => `(${p})`).join("|"), trigger.case_sensitive ? "" : "i");
|
|
|
|
regexCache.set(trigger, [mergedRegex]);
|
|
}
|
|
|
|
const regexes = regexCache.get(trigger)!;
|
|
|
|
for await (let [type, str] of matchMultipleTextTypesOnMessage(pluginData, trigger, context.message)) {
|
|
if (trigger.strip_markdown) {
|
|
str = stripMarkdown(str);
|
|
}
|
|
|
|
if (trigger.normalize) {
|
|
str = normalizeText(str);
|
|
}
|
|
|
|
for (const regex of regexes) {
|
|
const match = regex.exec(str);
|
|
if (match) {
|
|
const matchedWordIndex = match.slice(1).findIndex((group) => group !== undefined);
|
|
const matchedWord = trigger.words[matchedWordIndex];
|
|
|
|
return {
|
|
extra: {
|
|
type,
|
|
word: matchedWord,
|
|
},
|
|
};
|
|
}
|
|
}
|
|
}
|
|
|
|
return null;
|
|
},
|
|
|
|
renderMatchInformation({ pluginData, contexts, matchResult }) {
|
|
const partialSummary = getTextMatchPartialSummary(pluginData, matchResult.extra.type, contexts[0]);
|
|
const wordInfo = matchResult.extra.word ? ` (\`${escapeInlineCode(matchResult.extra.word)}\`)` : "";
|
|
return `Matched word${wordInfo} in ${partialSummary}`;
|
|
},
|
|
});
|