ryandonofrio3/osgrep

Overview

This skill provides semantic code search that finds code by meaning rather than exact text. Use it alongside grep: grep for literal matches, osgrep to locate concepts, logic, and orchestration across a codebase. It returns ranked snippets with context, labels, and call information to help you quickly understand relevant code paths.

How this skill works

You query with a natural language question or phrase and the tool returns about ten ranked code snippets with 15+ lines each, a relevance score, a category tag (e.g., ORCHESTRATION, DEFINITION), and calls made by the snippet. It supports tracing call graphs, generating compact file-only results, and producing skeleton outlines of large files. If the index is building, it will notify you so you can wait or accept partial results.

When to use it

• When you need to find where a behavior or concept is implemented (e.g., auth, caching, feature flags).
• When grep/ripgrep returns nothing because you’re searching by intent or architecture, not exact strings.
• When tracing how a request or action flows through services and functions.
• When you need compact pointers to where to read deeper instead of opening large files.
• When you want a quick summary of what a code range does before diving into implementation details.

Best practices

• Ask natural-language, specific questions: more words = better results (e.g., “where does the server validate JWT tokens”).
• Prioritize ORCHESTRATION results to understand logic and coordination points first.
• Read the exact line ranges osgrep returns instead of full files for focused context.
• If results look off, rephrase the query as you would ask a teammate.
• If you see indexing or syncing messages, decide whether to wait or accept partial/older results.

Example use cases

• Find where authentication or permission checks occur across the codebase.
• Locate entry points for incoming requests and understand request handling flow.
• Trace who calls a function and what that function calls using the trace command.
• Get a skeleton outline of a massive file to identify readable regions quickly.
• Produce compact lists of file paths when you only need locations for review or PRs.