Kb Scout

Then scan for structural signals of gaps:

• Open Questions sections in wiki articles -- these are explicit holes
• Broken wikilinks -- articles referenced but not yet written
• Thin articles -- wiki articles with sparse content or few sources
• Stale coverage -- areas where the newest source is old relative to the field's pace

Use kb-search for semantic context if it helps identify related coverage across domains:

uv run --project <kb-root>/_search kb-search --json --top 10 "relevant query"

2. Search for candidates

Based on the KB assessment and the user's focus instructions, search for new material using web search. Target:

For recent scouting:

• arXiv / Semantic Scholar / Google Scholar for preprints and papers
• Product Hunt, TechCrunch, HN, relevant blogs for products/tools
• GitHub trending / Papers With Code for implementations
• Search queries should be specific to the domain's terminology (use the index and existing article titles as vocabulary)

For gap filling:

• Search for topics identified in Open Questions sections
• Search for foundational work that's assumed but not cited
• Search for survey/review papers that could anchor thin areas
• Look for seminal papers missing from the raw sources

For competitive/product landscape:

• Search for companies, tools, and projects in the space
• Look for recent funding rounds, launches, or pivots
• Check for open-source alternatives and benchmarks

Adjust search recency based on context:

• "Recent scouting" -> last 1-3 months unless user specifies otherwise
• "Gap filling" -> no time restriction, prioritize quality and relevance
• Respect any time bounds the user gives ("since 2025", "this quarter")

3. Evaluate and filter

For each candidate found, assess:

• Relevance -- does it connect to existing KB concepts or fill an identified gap?
• Quality -- is it from a reputable venue/source? Is it substantive enough to warrant a KB entry?
• Novelty -- does the KB already cover this? Check against existing raw sources by title/author/DOI to avoid duplicates.
• Actionability -- can the user actually get and read this? (open access, available preprint, etc.)

Filter aggressively. A good scouting report has 5-15 high-signal items, not 50 marginal ones.

4. Produce the scouting report

Create a report as a markdown artifact at: <domain>/artifacts/scout-report-<YYYY-MM-DD>.md

Report structure:

---