code-search-assistant

Code Search Assistant

Overview

Search codebases to find code related to a given snippet using multi-dimensional similarity analysis: call chain patterns, textual structure, and functional behavior. Results are ranked and presented with matching code snippets.

Workflow

1. Analyze Input Snippet

Extract key characteristics from the provided code snippet:

Structural elements:

• Function/method calls made
• Classes/types used
• Control flow patterns (loops, conditionals, try-catch)
• Data structures (arrays, objects, maps)

Functional elements:

• Purpose/intent of the code
• Input/output behavior
• Side effects (I/O, state changes, API calls)
• Domain concepts (authentication, validation, transformation)

Textual elements:

• Variable and function names
• String literals and constants
• Comments and documentation
• Code tokens and keywords

2. Define Search Scope

Determine where to search:

• Full repository: Search all code files
• Specific directories: Focus on relevant modules
• File type filter: Limit to specific languages

Use Glob to identify candidate files:

**/*.js, **/*.py, **/*.java, etc.

3. Search by Call Chain Similarity

Find code with similar function call patterns and dependencies.

Search strategy:

1. Extract function/method calls from input snippet
2. Use Grep to find files containing those function calls
3. Read matching files to analyze call sequences
4. Score based on:
   • Number of shared function calls (weight: 40%)
   • Order of function calls (weight: 30%)
   • Shared imported modules/libraries (weight: 30%)

Example:

// Input snippet calls: fetch(), JSON.parse(), setState()
// High match: Code that calls fetch() → JSON.parse() → setState()
// Medium match: Code that calls fetch() and setState() in different order
// Low match: Code that only calls fetch()

4. Search by Textual Similarity

Find code with similar structure and token patterns.

Search strategy:

1. Extract key identifiers from input snippet (function names, variable names)
2. Use Grep to find files with similar identifiers
3. Read matching files to compare code structure
4. Score based on:
   • Shared variable/function names (weight: 35%)
   • Similar control flow structure (weight: 35%)
   • Shared keywords and operators (weight: 30%)