searchcans-rag-source-curatorlisted

# SearchCans RAG Source Curator Build a deliberately bounded source manifest rather than crawling the web. The Skill searches Google or Bing, chooses diverse candidate domains, fetches only the selected pages with Reader, accepts explicit PDF/Office file URLs, and can retain screenshot URLs for visual review. ## Create the source set Start with a concrete question, market, source budget, and the minimum number of successfully extracted sources needed before the pack is considered ready. Add direct documents only with `--file-url`; do not infer file URLs from a page. ```bash python scripts/rag_source_curator.py "What is the current SERP API market?" \ --engine google --country us --language en \ --source-budget 4 --min-claim-ready 2 --include-content \ --out rag-sources.json ``` Use `--query` for a small query matrix. Use `--screenshot 1` or `2` only when visual review is required. `--include-content` intentionally makes the output larger; omit it when you need a metadata manifest before a second retrieval step. ## Apply the evidence quality gate The output marks a source `claim_ready` only when the Reader/File Extraction response has content. Its `evidence_gate.status` is `passed` only when the requested minimum is met. Never treat a SERP snippet, an unread URL, an empty response, or a screenshot as sufficient evidence for a consequential answer. Every source begins with `authority_assessment: unassessed`; assign authority, recency, permissions, and organization