Public Design Search

Use public search pages as the primary source of truth and keep the workflow login-free.

Workflow

1. Choose search tags.
• Prefer English tags for these three platforms unless a platform clearly supports another language.
• Load references/tag-taxonomy.md for generic tag-selection guidance and example category/tag patterns.
• Keep tag choice independent from platform-specific extraction logic.

1. Build public search URLs.
• Dribbble: https://dribbble.com/search/
• Pinterest: https://www.pinterest.com/search/pins/?q=&rs=typed
• Behance: https://www.behance.net/search/projects?search=

1. Collect candidate cards from the search page.
• Prefer browser automation over raw HTTP for all three platforms.
• Normalize search-card fields first: title, source_url, cover_image_url, summary, author_name when exposed, and raw_metrics when exposed.

1. Backfill sparingly.
• Default: keep the search-page cover_image_url.
• Use detail pages primarily for publish_time.
• Only backfill author_name when the search card exposes nothing and the request budget allows it.
• On constrained servers or proxy-heavy environments, disable Pinterest detail backfill entirely and rely on search-page results.

1. Return a normalized public-only record.
• Keep unstable fields nullable.
• Record the input tag or matched tag set when useful.
• Do not fabricate authors, publish times, or engagement metrics.

Extraction Priorities

• title: search-card title first, then safe public meta/title fallback.
• cover_image_url: search-page image first; only do light URL upgrades when the site uses obvious small fixed-size variants.
• author_name: search-card author first; leave Unknown if public search does not expose it reliably.
• publish_time: detail-page public metadata when available; otherwise null.
• summary: normalized title or short public snippet.

Platform-specific rules live in:

• references/platform-methods.md
• references/tag-taxonomy.md

Network Guidance

• Pinterest and Behance may require VPN or proxy access in restricted regions.
• Browser traffic may work best through SOCKS5 while image download or pHash download may need an HTTP proxy.
• Verify network reachability before changing selectors.
• If a platform is reachable only through unstable proxies, reduce detail backfill before changing the tag set.

Guardrails

• Public pages only. No login-only APIs, cookies, or authenticated sessions.
• Prefer search-page fields over detail scraping whenever they are good enough.
• Do not chase high-resolution cover images unless the search-page image is clearly unusable.
• Treat missing metrics as normal.
• Expect DOM drift and anti-bot behavior.
• These three platforms are effectively English-first in most reusable workflows.

Output Shape

Normalize results into a shape close to:

{
  "title": "",
  "platform": "dribbble|pinterest|behance",
  "source_url": "",
  "cover_image_url": "",
  "author_name": "",
  "publish_time": null,
  "keyword": "",
  "matched_tags": []
}