mirror of
https://github.com/brightdata/skills.git
synced 2026-09-17 14:18:09 +08:00
Shahar Carmi
cc87a228d8
Add three skills built on Bright Data's Discover API (intent-ranked semantic
web search), forming a sub-graph where discover-api is the foundation:
- discover-api: REST/CLI/SDK surfaces, trigger→poll, modes
(standard/zeroRanking/deep/fast), per-surface parameter matrix, result shape
- live-research: decompose → multi-angle Discover → dedup/rank → cited brief,
with a tested merge_corpus.sh helper and brief-template
- rag-pipeline: live retrieval + vector ingestion architectures, with runnable
JS/Python and a concrete (swappable) OpenAI+pgvector adapter
Method names, parameters, and result shapes were LIVE-TESTED against the real
Discover API and corrected accordingly:
- JS SDK client.discover() returns { success, data:[...] } (rows under .data),
NOT a bare array and NOT .results (the REST/CLI shape) — also fixed the same
wrong claim in js-sdk-best-practices/search.md
- CLI flag matrix verified against `bdata discover --help` (v0.3.1)
- added content-quality gating (high relevance_score can still be a 404 stub)
and transient-failure retry guidance
Bump both manifests to v1.8.0, update README (badge/nav/table/sections/tree/
resources) to 21 skills, add discover/rag/semantic-search keywords.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>