Skip to content

A self-hostable bookmark-everything app (links, notes and images) with AI-based automatic tagging and full text search

feat: tagging prompt hardening

Mohamed Bassem9ceb077c8712
diff --git a/packages/shared/prompts.ts b/packages/shared/prompts.ts
index 0c800210..a0dbad82 100644
--- a/packages/shared/prompts.ts
+++ b/packages/shared/prompts.ts
@@ -26,13 +26,20 @@ export function buildImagePrompt(
);
return `
-You are an expert whose responsibility is to help with automatic text tagging for a read-it-later/bookmarking app.
+You are an expert whose responsibility is to help with automatic tagging for a read-it-later/bookmarking app.
Analyze the attached image and suggest relevant tags that describe its key themes, topics, and main ideas. The rules are:
-- Aim for a variety of tags, including broad categories, specific keywords, and potential sub-genres.
+- Prefer concrete, durable tags: named products, technologies, projects, standards, subject areas, and important concepts.
+- Include only retrieval-worthy tags that describe the saved item's intended content, not incidental UI or page chrome.
- The tags must be in ${lang}.
-- If the tag is not generic enough, don't include it.
+- Keep each tag short: ideally 1-3 words. Do not include parenthetical explanations, comma-separated examples, or long descriptive phrases inside a tag.
+- Prefer durable subject tags over one-off facts, examples, source organizations, page sections, or implementation details unless they are central to the whole item.
+- Do NOT generate any tags if the image is mainly:
+ - A screenshot of an error, unavailable, forbidden, unauthorized, not found, DNS, timeout, or service failure page
+ - A Cloudflare/security check, CAPTCHA, bot check, anti-DDoS challenge, browser verification, or access-blocked page
+ - Boilerplate content such as cookie consent, login walls, GDPR notices, navigation menus, or a blank/empty image
+ In these cases, return an empty tags array. Do not tag the failure/interstitial page itself.
- Aim for 10-15 tags.
-- If there are no good tags, don't emit any.
+- If there are no good tags, leave the array empty.
${curatedInstruction}
${potentialRelevantTagsInstruction}
${tagStyleInstruction}
@@ -60,12 +67,16 @@ export function constructTextTaggingPrompt(
return `
You are an expert whose responsibility is to help with automatic tagging for a read-it-later/bookmarking app.
Analyze the TEXT_CONTENT below and suggest relevant tags that describe its key themes, topics, and main ideas. The rules are:
-- Aim for a variety of tags, including broad categories, specific keywords, and potential sub-genres.
+- Prefer concrete, durable tags: named products, technologies, projects, standards, subject areas, and important concepts.
+- Include only retrieval-worthy tags that describe the saved item's intended content, not incidental page chrome.
- The tags must be in ${lang}.
-- If the tag is not generic enough, don't include it.
-- Do NOT generate tags related to:
- - An error page (404, 403, blocked, not found, dns errors)
- - Boilerplate content (cookie consent, login walls, GDPR notices)
+- Keep each tag short: ideally 1-3 words. Do not include parenthetical explanations, comma-separated examples, or long descriptive phrases inside a tag.
+- Prefer durable subject tags over one-off facts, examples, source organizations, page sections, or implementation details unless they are central to the whole item.
+- Do NOT generate any tags if the content is mainly:
+ - An error, unavailable, forbidden, unauthorized, not found, DNS, timeout, or service failure page
+ - A Cloudflare/security check, CAPTCHA, bot check, anti-DDoS challenge, browser verification, or access-blocked page
+ - Boilerplate content such as cookie consent, login walls, GDPR notices, navigation menus, or empty pages
+ In these cases, return an empty tags array. Do not tag the failure/interstitial page itself.
- Aim for 3-5 tags.
- If there are no good tags, leave the array empty.
${curatedInstruction}