Code Semantic Search

Semantic search over source-code corpora. This skill is the specialized sibling of the general-purpose rag-implementation skill: same RAG mental model, but adapted for the three properties that make code corpora different from prose corpora.

DevAI-Hub's reference implementation is the internal devai-code-search MCP. v1.0.0 ships keyword-only search (inverted index + rapidfuzz, zero API keys, zero model downloads, zero outbound calls). v1.1.0 adds dense retrieval with local ONNX embeddings and a sqlite-vec vector store. Under the MCP Registry Policy in AGENTS.md, third-party semantic-code-search services are drop-class because their default flow ships source code to an external embedding endpoint.

When to Use This Skill

Use this skill when:

• The agent must answer questions about a codebase that exceeds the model's context window. Loading the full tree is not an option; retrieval is.
• Natural-language queries about the code ("where do we handle rate limiting for the billing endpoint?") are the dominant access pattern, and grep would return too many false positives.
• The team wants to eliminate "spelunking through the repo" as a recurring task before code reviews, onboarding, or incident response.
• A new feature needs to cite existing implementations of similar patterns in the codebase.
• You are planning a refactor that needs to find every caller of a function whose exact name you do not remember.

When NOT to use: