Locates, ranks, and invokes repository symbols from natural language to streamline AI-driven code exploration.
The Code Discovery skill enhances Claude's ability to navigate complex codebases by implementing an agent-friendly packaging system. It automatically generates symbol-level manifests and structured maps, allowing AI agents to pinpoint specific logic and invoke functions without requiring full-file context. This approach minimizes token overhead and maximizes the precision of AI-assisted refactoring, debugging, and feature implementation within large repositories by providing a clear entry point for symbol-level retrieval.
Key Features
01Generation of per-function manifests for granular LLM access
021 GitHub stars
03Creation of CODE.md maps for high-level repository navigation
04Symbol-level retrieval to minimize context and token usage
05Automatic symbol discovery and ranking via natural language
06TypeScript generator for standardized code packaging
Use Cases
01Standardizing repository documentation for optimized AI-assisted coding
02Navigating large or unfamiliar repositories to find specific business logic
03Enabling LLMs to execute specific functions without reading entire source files
