search-strategy_skill

Overview

This skill orchestrates multi-source enterprise search by decomposing a natural language question into targeted, source-specific queries and returning a ranked, deduplicated synthesis. It translates user intent into the right mix of semantic and keyword searches, executes them in parallel, and applies relevance and authority heuristics to surface the best answers. It also handles ambiguity and fallback strategies to ensure robust results when sources are missing or queries are unclear.

How this skill works

The skill classifies the query type (decision, status, document, person, factual, temporal, exploratory) and extracts keywords, entities, intent signals, constraints, and negations. It then generates tailored sub-queries per connected source (chat, knowledge base, project tracker, cloud storage), choosing semantic or keyword search and mapping enterprise filters to source-specific syntax. Searches run in parallel, results are merged, scored by relevance/authority/freshness, deduplicated, and synthesized into a single coherent answer. When ambiguity or failures occur, it asks focused clarifying questions or applies fallback broadening rules.

When to use it

• You need a single answer synthesized from chat, docs, and trackers
• You want to find decisions or conclusions across conversations and meeting notes
• You need status or task updates across project trackers and chats
• You want authoritative policy or documentation surfaced quickly
• The query is exploratory and benefits from semantic search across sources
• A source returned no results or is temporarily unavailable

Best practices

• Prefer semantic search for conceptual or exploratory queries and keyword search for exact names and filters
• Always extract entities, time constraints, and negations before generating sub-queries
• Run source queries in parallel to minimize latency and merge results centrally
• Rank results using a query-type weighted relevance model (keywords, freshness, authority, completeness)
• Ask one concise clarifying question only when distinct interpretations would change results substantially
• When sources are unavailable, explicitly note gaps and broaden queries in a defined order (remove date, remove location, relax keywords)

Example use cases

• User: "What did we decide about the API migration timeline?" → searches chats, meeting notes, tracker tasks, and docs; returns prioritized decision evidence and links
• User: "What's the status of Project Aurora?" → pulls recent tracker updates, recent chat mentions, and status docs to synthesize current progress and blockers
• User: "Where's the spec for feature X?" → searches wikis, drives, and shared docs with exact-phrase and semantic variants and returns authoritative doc links
• User: "Who's working on the billing migration?" → checks task assignees, recent commit/authorship, and chat mentions to list likely owners and confidence levels