Skip to content

fix: replace deprecated prompt injection model #34

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Lubrsy706 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: replace deprecated prompt injection model #34

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
10 changes: 6 additions & 4 deletions workers/lib/ai.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,16 +21,18 @@ Return ONLY "NO" if it is a normal email (even if angry, confused, or containing

Respond with exactly one word: YES or NO.`;

const PROMPT_INJECTION_MODEL = "@cf/zai-org/glm-4.7-flash";

export async function isPromptInjection(ai: Ai, bodyHtml: string | null | undefined): Promise<boolean> {
if (!bodyHtml) return false;

const plainText = stripHtmlToText(bodyHtml).trim();
if (plainText.length < 10) return false;

try {
const response = (await ai.run(
// @ts-expect-error — model string not in generated union
"@cf/meta/llama-3.1-8b-instruct-fast",
PROMPT_INJECTION_MODEL,
{
messages: [
{ role: "system", content: INJECTION_PROMPT },
Expand All @@ -42,12 +44,12 @@ export async function isPromptInjection(ai: Ai, bodyHtml: string | null | undefi
)) as { response?: string };

const result = (response?.response || "NO").trim().toUpperCase();

if (result.includes("YES")) {
console.warn("Prompt injection detected in incoming email, blocking auto-draft");
return true;
}

return false;
} catch (e) {
console.error("Prompt injection scanner failed, skipping auto-draft:", (e as Error).message);
Expand Down