query

Query

Ask natural language questions about a codebase and receive detailed answers with citations to specific files. The agent iteratively searches and reads files to gather context before generating a comprehensive answer.

How It Works

The query agent:

1. Analyzes the repository structure to understand the codebase
2. Searches for relevant files using regex and/or vector search
3. Reads and inspects files that might contain answers
4. Synthesizes information and generates a detailed answer
5. Includes citations to files where information was found
6. Returns the list of files that were analyzed

Usage

query "How does the authentication system work?"

# Complex query with hard LLM (more thorough)
query -h "Explain the complete data flow from API request to database response"

# With initial files to include in context
query --initial-files=src/auth.ts,src/middleware.ts "How does this authentication flow work?"

Flags:

• -h: Use hard LLM for complex queries (slower but more thorough)
• --initial-files=file1,file2: Comma-separated list of initial files to include in context
• -r: Resume from previous agent run

Output Format

Returns a detailed text answer with:

• Comprehensive explanation answering the query
• Citations to specific files where information was found
• List of all files that were analyzed
• Confidence level indicator

Example Queries

Architecture & Design:

• "What is the overall architecture of this application?"
• "What design patterns are used in the codebase?"

Implementation Details:

• "How does the caching mechanism work?"
• "Where is error handling implemented?"

Debugging:

• "Where would authentication failures be logged?"
• "What happens when a database connection fails?"

Notes

• Uses multiple search strategies (file system tree with per-file summaries, regex and vector search if available)