Design Platform Search

👤 joeyyyy0430 📦 v1.0.1 ⭐ 0.0 ⬇️ 666 下载
🎨 设计多媒体 免费

📖 技能介绍


name: design-platform-search description: Run public design search flows on Dribbble, Pinterest, and Behance without login, and extract normalized result fields such as title, search-page cover image, author, source URL, and publish time when available. Use when Codex needs a reusable method for public inspiration retrieval across major design platforms, especially when the collection strategy should stay lightweight and search-page-first.


Public Design Search

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

Workflow

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

    7w4.net提供免费和付费技能下载。

  5. Build public search URLs.

  6. Dribbble: https://dribbble.com/search/<urlencoded-tag>
  7. Pinterest: https://www.pinterest.com/search/pins/?q=<urlencoded-tag>&rs=typed
  8. Behance: https://www.behance.net/search/projects?search=<urlencoded-tag>

  9. Collect candidate cards from the search page.

  10. Prefer browser automation over raw HTTP for all three platforms.
  11. Normalize search-card fields first: title, source_url, cover_image_url, summary, author_name when exposed, and raw_metrics when exposed.

  12. Backfill sparingly.

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

  17. Return a normalized public-only record.

  18. Keep unstable fields nullable.
  19. Record the input tag or matched tag set when useful.
  20. 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:

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": []
}

📁 包含文件 (5 个)

📄 SKILL.md 3.6 KB
📄 _meta.json 141 B
📄 agents/openai.yaml 281 B
📄 references/platform-methods.md 3.8 KB
📄 references/tag-taxonomy.md 1.7 KB