Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
78 changes: 78 additions & 0 deletions src/server/features/ai-search/services/promptExplorer.test.ts
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
import { describe, expect, it, vi } from "vitest";
import type { LlmResponseResult } from "@/server/lib/dataforseoLlmSchemas";

vi.mock("cloudflare:workers", () => ({ waitUntil: vi.fn() }));

const { extractCitations } = await import("./promptExplorer");

// DataForSEO's LLM Responses payload nests references as untyped
// `{ title, url }` objects under items[].sections[].annotations — mirroring the
// SDK's AnnotationInfo, which has no citation-type discriminator.
function response(
annotations: Array<{ title?: string; url?: string }>,
): LlmResponseResult {
return {
model_name: "gpt-5",
web_search: true,
items: [
{
type: "reasoning",
sections: [{ type: "summary_text", text: "thinking" }],
},
{
type: "message",
sections: [{ type: "text", text: "answer", annotations }],
},
],
};
}

describe("extractCitations", () => {
it("keeps untyped annotations (no citation-type discriminator exists)", () => {
const citations = extractCitations(
response([
{ title: "Town & Country", url: "https://www.townandcountrymag.com/x" },
{ title: "Stylevana", url: "https://www.stylevana.com/y" },
]),
);
expect(citations.map((c) => c.url)).toEqual([
"https://www.townandcountrymag.com/x",
"https://www.stylevana.com/y",
]);
expect(citations[0]?.domain).toBe("townandcountrymag.com");
expect(citations[0]?.title).toBe("Town & Country");
});

it("dedupes repeated URLs and drops unsafe schemes", () => {
const citations = extractCitations(
response([
{ title: "A", url: "https://example.com/a" },
{ title: "A dup", url: "https://example.com/a" },
{ title: "evil", url: "javascript:alert(1)" },
{ title: "no url" },
]),
);
expect(citations).toHaveLength(1);
expect(citations[0]?.url).toBe("https://example.com/a");
});

it("ignores annotations outside message items and returns [] when absent", () => {
expect(extractCitations({ items: [] })).toEqual([]);
expect(
extractCitations({
items: [
{
type: "reasoning",
sections: [
{
type: "summary_text",
text: "t",
annotations: [{ title: "x", url: "https://x.test/1" }],
},
],
},
],
}),
).toEqual([]);
});
});
10 changes: 5 additions & 5 deletions src/server/features/ai-search/services/promptExplorer.ts
Original file line number Diff line number Diff line change
Expand Up @@ -207,7 +207,7 @@ function extractText(response: LlmResponseResult): string {
return textParts.join("\n\n").trim();
}

function extractCitations(
export function extractCitations(
response: LlmResponseResult,
): PromptExplorerCitation[] {
const seen = new Set<string>();
Expand All @@ -217,10 +217,10 @@ function extractCitations(
if (item.type !== "message") continue;
for (const section of item.sections ?? []) {
for (const annotation of section.annotations ?? []) {
if (annotation.type !== "citation") continue;
// Drop non-http(s) URLs — LLMs can be coaxed into emitting
// `javascript:` payloads as "citations" and we render these as
// <a href> in the UI.
// DataForSEO annotations are untyped `{ title, url }` reference
// objects (AnnotationInfo) — there is no citation-type discriminator
// to filter on. Guard on URL safety only: LLMs can be coaxed into
// emitting `javascript:` payloads, and we render these as <a href>.
const safeUrl = safeHttpUrl(annotation.url);
if (!safeUrl || seen.has(safeUrl)) continue;
seen.add(safeUrl);
Expand Down