Source code
Revision control
Copy as Markdown
Other Tools
/**
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/.
*/
/**
* Extract link destination from text starting at given position.
* Handles balanced parentheses and stops at optional title.
*
* @param {string} text - Text to parse
* @param {number} start - Position to start parsing from
* @returns {string} Extracted URL
*/
function extractLinkDestination(text, start) {
let i = start;
let parenDepth = 0;
let url = "";
while (i < text.length) {
const ch = text[i];
if (ch === "(") {
parenDepth++;
url += ch;
} else if (ch === ")") {
if (parenDepth === 0) {
break;
}
parenDepth--;
url += ch;
} else if (ch === " " || ch === "\t" || ch === '"' || ch === "'") {
break;
} else if (ch === "\n") {
break;
} else {
url += ch;
}
i++;
}
return url.trim();
}
/**
* Extract markdown links from response text.
* Skips links in code blocks and image syntax.
*
* @param {string} responseText - Raw response text with markdown links
* @returns {Array<object>} Array of {text, url, position} objects
*/
export function extractMarkdownLinks(responseText) {
if (!responseText) {
return [];
}
// Strip fenced code blocks (replace with spaces to preserve positions)
let cleaned = responseText.replace(/```[\s\S]*?```/g, match =>
" ".repeat(match.length)
);
// Strip inline code with variable-length backtick fences (e.g., `code`, ``code``)
cleaned = cleaned.replace(/(`+)(?!\1)([^`]|(?!\1)`)*\1/g, match =>
" ".repeat(match.length)
);
const links = [];
// Negative lookbehind to skip image links ()
const linkStartRegex = /(?<!!)\[([^\]]+)\]\(/g;
let match;
while ((match = linkStartRegex.exec(cleaned)) !== null) {
const text = match[1];
const startPos = match.index;
const urlStart = match.index + match[0].length;
const url = extractLinkDestination(cleaned, urlStart);
// Skip anchor links
if (url && !url.startsWith("#")) {
links.push({ text, url, position: startPos });
}
}
return links;
}
/**
* Normalize a URL for comparison purposes.
* Handles trailing slashes, host case, and default ports.
*
* @param {string} urlString - URL to normalize
* @returns {string} Normalized URL
*/
function normalizeUrl(urlString) {
try {
const url = new URL(urlString);
// Lowercase hostname
url.hostname = url.hostname.toLowerCase();
// Remove default ports
if (
(url.protocol === "http:" && url.port === "80") ||
(url.protocol === "https:" && url.port === "443")
) {
url.port = "";
}
// Remove trailing slash from pathname (except root)
if (url.pathname.length > 1 && url.pathname.endsWith("/")) {
url.pathname = url.pathname.slice(0, -1);
}
return url.href;
} catch {
return urlString;
}
}
/**
* Validate that cited URLs exist in the allowed URL list.
*
* @param {Array<string>} citedUrls - Array of citation URLs from response
* @param {Array<string>} allowedUrls - Array of allowed URLs from tool results
* @returns {object} Validation result:
* - valid: Array of valid citation URLs
* - invalid: Array of hallucinated citation URLs
* - validationRate: Percentage of valid citations (0-1)
*/
export function validateCitedUrls(citedUrls, allowedUrls) {
if (!Array.isArray(citedUrls) || citedUrls.length === 0) {
return { valid: [], invalid: [], validationRate: 1.0 };
}
if (!Array.isArray(allowedUrls)) {
return { valid: [], invalid: [...citedUrls], validationRate: 0 };
}
const normalizedAllowed = new Set(allowedUrls.map(normalizeUrl));
const valid = [];
const invalid = [];
for (const url of citedUrls) {
if (normalizedAllowed.has(normalizeUrl(url))) {
valid.push(url);
} else {
invalid.push(url);
}
}
const validationRate = valid.length / citedUrls.length;
return { valid, invalid, validationRate };
}