The attacker host receives the request (including the dataset_id) and controls the response body that is surfaced back to the model/user — enabling response spoofing and, in an agentic setting, indirect prompt-injection content delivered under the guise of a trusted-portal tool.
Contributes to SSRF surface: while makeCkanRequest blocks private/internal IPs, this bypass removes the domain restriction that the code intends to enforce for these tools.
The @-userinfo variant is the most severe form because validation passes on a string whose actual host is fully attacker-chosen.
Proof of concept
poc/mqa-allowlist-poc.mjs runs the verbatim regex over benign and malicious URLs:
Validate the parsed host, not the raw string. For example:
function isValidMqaServer(serverUrl) {
let u; try { u = new URL(serverUrl); } catch { return false; }
if (u.protocol !== "https:") return false;
const h = u.hostname.toLowerCase();
return h === "dati.gov.it" || h === "www.dati.gov.it";
// or: h === "dati.gov.it" || h.endsWith(".dati.gov.it")
}
Anchoring the regex end-to-end (/^https:\/\/(www\.)?dati\.gov\.it(\/|$)/i) also closes the suffix trick, but URL-parsing + exact host comparison is the robust fix and also neutralizes the @-userinfo variant.