Return ranked code chunks with file paths, scope labels, and confidence percentages.
npx skills add https://github.com/auyelbekov/rawq --skill rawq-agent-usage-guide
Please help me install this Agent Skill. Skill: rawq — Agent Usage Guide Source: https://github.com/auyelbekov/rawq/tree/main Command: npx skills add https://github.com/auyelbekov/rawq --skill rawq-agent-usage-guide
This guide teaches agents to avoid reading entire codebases by letting rawq surface the most relevant code chunks with metadata, so they stop wasting tokens on irrelevant files.
rawq map, search, diff, and the daemon, plus JSON/stream outputs for integration with agent pipelines.Tell the agent to run rawq search "how does authentication failure handling work" ./testdata to receive ranked code snippets with file scopes and confidence scores.
High-intent search queries and answers about installing and using this skill.
Hybrid semantic and BM25 code search returns ranked code chunks with file paths, scope labels, and confidence percentages for descriptive queries. This avoids reading entire repositories by surfacing only the most relevant snippets to save tokens.
Hybrid semantic and lexical code search fuses embedding-based retrieval with BM25 ranking to adaptively weight queries for both conceptual intent and identifier-driven lookups. It is needed when exploring unfamiliar repositories without preidentifying relevant files.
Code search supports contextual filtering through language filters and exclusion globs. You can restrict retrieval to specific programming languages and exclude unwanted file patterns to ensure results match your agent workflow and token budget requirements.
Diff-restricted queries allow you to search for code changes within a specific git diff. The tool applies hybrid semantic and BM25 retrieval directly to the diff output, returning ranked code chunks scoped to only the modified portions of the repository.
The daemon supports caching for repeated agent queries using daemon-backed embedding models. It maintains an indexed context map of the repository, enabling faster subsequent searches and structured JSON or NDJSON streaming outputs for pipeline integration.
Confidence scoring in code retrieval provides percentage-based relevance rankings for returned chunks, but relies on adaptive hybrid weighting. If queries lack descriptive intent or target highly fragmented code, confidence percentages may not accurately reflect contextual relevance.
aliyunBuild Hologres knowledge bases with vector and full-text indexes.
erencodingSubscribe to Feishu/Lark events via WebSocket and output NDJSON.
h4vzzOptimize raw context chunks for AI models within a token budget.
ricardorocheStandardizes chunking, vector database integration, embedding management, and retrieval reranking for RAG pipelines.