ChunkHound

What I Do

Provide universal patterns and best practices for using ChunkHound MCP server tools for semantic code search, regex pattern matching, and deep architectural research across codebases.

Universal ChunkHound Usage Patterns

Tool Selection Guide

Semantic Search Patterns

# Universal pattern for conceptual code discovery
from chunkhound import search_semantic

# Find authentication-related code by concept
results = search_semantic(
    query="how does user authentication work in this codebase?",
    page_size=10,
    max_response_tokens=20000
)

# Narrow search to specific directory
results = search_semantic(
    query="error handling patterns",
    path="src/components",
    page_size=5
)

Regex Search Patterns

# Universal pattern for exact pattern matching
from chunkhound import search_regex

# Find all function definitions
results = search_regex(
    pattern=r"def \w+\(",
    page_size=20
)

# Find all class definitions in Python files
results = search_regex(
    pattern=r"class \w+:",
    include="*.py"
)

Code Research Patterns

# Universal pattern for architectural exploration
from chunkhound import code_research

# Deep architectural analysis
report = code_research(
    query="how does the payment processing system work?"
)

# Research specific component relationships
report = code_research(
    query="map the data flow from API to database"
)

When to Use Me

Use this skill when:

• Exploring unfamiliar codebases for architectural understanding
• Finding existing patterns before implementing new features
• Debugging by mapping complete system flows
• Refactoring preparation with dependency analysis
• Code archaeology in legacy systems

Universal Examples

Architecture Discovery Workflow

# Step 1: Broad semantic search to understand concepts
auth_concepts = search_semantic(query="authentication implementation")

# Step 2: Extract key symbols for comprehensive search
key_symbols = extract_symbols_from_results(auth_concepts)

# Step 3: Find all references with regex
for symbol in key_symbols:
    references = search_regex(pattern=symbol)

# Step 4: Deep research for complete understanding
full_report = code_research(query="complete authentication architecture")

Debugging Pattern Matching

# Find error handling patterns
error_patterns = search_semantic(query="error handling and logging")

# Search for specific error types
validation_errors = search_regex(pattern=r"ValidationError|InvalidInput")

# Research complete error flow
error_flow = code_research(query="how errors propagate through the system")

Refactoring Preparation

# Understand current implementation
current_impl = code_research(query="current caching strategy")

# Find all usage patterns
cache_usage = search_semantic(query="cache usage patterns")

# Identify all cache-related code
cache_symbols = search_regex(pattern=r"(?i)cache")

Best Practices

Search Strategy

1. Start Broad: Use semantic search for conceptual understanding
2. Narrow Down: Use regex search for precise symbol locations
3. Go Deep: Use code research for architectural relationships

Performance Optimization

• Use path parameter to limit search scope when possible
• Adjust page_size based on expected result volume
• Use maxresponsetokens to control output size

Result Interpretation

• Semantic search finds conceptually related code
• Regex search finds exact matches and references
• Code research provides structured architectural reports

Compatibility Notes

This skill works with:

• Any codebase with ChunkHound MCP server configured
• OpenCode agent framework
• Claude-compatible MCP clients
• Projects requiring deep code understanding

Integration with Other Skills

With knowledge-management: Store research findings as memories

store_memory(
    type="code_pattern",
    title="Authentication architecture discovered",
    content=research_report,
    tags=["architecture", "authentication"]
)

With issue-tracking: Create tasks based on research findings

create_issue(
    title="Refactor authentication based on research",
    description=f"Research shows: {key_findings}"
)