Knowledge Graph

Knowledge Graph Skill

Access to GitLab Knowledge Graph (gkg) for semantic code discovery. Indexes code structure including files, classes, functions, and their relationships.

Quick Reference

• Templates: See templates.md for reporting knowledge graph results

Available Actions

search_codebase

Search for code entities by name or pattern.

Parameters:

• query (required): Search query (function name, class name, etc.)
• node_types (optional): Filter by types - function, class, file, module
• language (optional): Filter by language - python, typescript, rust, ruby
• limit (optional): Maximum results (default: 20)

Example:

{
  "action": "search_codebase",
  "parameters": {
    "query": "handleAuth",
    "node_types": ["function"],
    "language": "typescript"
  }
}

find_symbol_references

Find all usages of a symbol across the codebase.

Parameters:

• symbol_name (required): Name of the function, class, or variable
• repository (optional): Limit search to specific repository

Example:

{
  "action": "find_symbol_references",
  "parameters": {
    "symbol_name": "processPayment"
  }
}

get_code_structure

Get the file and directory structure of a repository.

Parameters:

• repository (required): Name of the repository
• path (optional): Specific path within the repository

Example:

{
  "action": "get_code_structure",
  "parameters": {
    "repository": "agent-bot",
    "path": "src/services"
  }
}

find_dependencies

Find what a code entity imports, calls, or inherits from.

Parameters:

• node_id (required): ID of the code entity from a previous search
• direction (optional): outgoing (what this uses) or incoming (what uses this)

Example:

{
  "action": "find_dependencies",
  "parameters": {
    "node_id": "uuid-of-module",
    "direction": "outgoing"
  }
}

find_code_path

Find the relationship path between two code entities.

Parameters:

• source_id (required): ID of the source entity
• target_id (required): ID of the target entity

Example:

{
  "action": "find_code_path",
  "parameters": {
    "source_id": "uuid-of-caller",
    "target_id": "uuid-of-callee"
  }
}

get_code_neighbors

Get neighboring code entities at a specified depth.

Parameters:

• node_id (required): ID of the code entity
• edge_types (optional): Filter by relationship - calls, imports, inherits, uses
• depth (optional): How many levels to traverse (default: 1)

Example:

{
  "action": "get_code_neighbors",
  "parameters": {
    "node_id": "uuid-of-class",
    "edge_types": ["inherits", "implements"],
    "depth": 2
  }
}

get_graph_stats

Get statistics about the knowledge graph.

Example:

{
  "action": "get_graph_stats",
  "parameters": {}
}

MCP Integration

This skill integrates with the knowledge-graph MCP server running on port 9005. The MCP server provides the following tools:

Node Types

Edge Types

Workflow Example

1. Search for a function:

   search_codebase(query="processPayment", node_types=["function"])
   

2. Get its dependencies:

   find_dependencies(node_id="<result_id>", direction="outgoing")
   

3. Find what calls it:

   find_dependencies(node_id="<result_id>", direction="incoming")
   

4. Explore neighbors:

   get_code_neighbors(node_id="<result_id>", depth=2)