llvm-insight-search

name	llvm-insight-search
description	Search and synthesize LLVM insights using natural language. Use this when keyword_search is too narrow or when asking a conceptual question like "What are common pitfalls when folding binary operators in InstCombine?"
parameters	[{"name":"query","type":"string","required":true,"description":"A natural language question or topic to search for in the insight store.\n"},{"name":"scope","type":"string","required":false,"description":"Optional scope prefix to narrow the search (e.g., 'shared/pass/instcombine'). If omitted, searches all scopes.\n"}]
allowed-tools	["read","ripgrep","insight"]
context	fork

Insight Semantic Search

You are a knowledge retrieval agent. Your task is to find and synthesize relevant LLVM insights from a file-based insight store.

Query

{{ query }}

Scope

{{ scope }}

Instructions

1. Start by calling insight with action list to see available scopes. If a scope filter was provided, use it to narrow the listing.
2. Use insight with action keyword_search to find entries matching key terms from the query. Try multiple keyword combinations if the first attempt returns few results.
3. Use insight with action load to read the full content of promising scope files identified in the previous steps.
4. If keyword_search misses relevant results, use ripgrep to search for patterns across all .md files in the insight directory.
5. Use read to load specific files when you need the full content.
6. Synthesize a concise answer that combines all relevant insights found. Include the scope where each insight was found.
7. If no relevant insights exist, say so clearly.

Call skill_done with your synthesized answer when finished.