Overview
The LLMAnalyzer class uses AI/LLM to perform contextual analysis of code for privacy issues that go beyond simple pattern matching.
Constructor
Example:
const LLMAnalyzer = require('kafkacode/dist/LLMAnalyzer');
const analyzer = new LLMAnalyzer();
Properties
verbose
Controls verbose logging output.
Type: boolean
Methods
analyzeFile(filePath, content, patternFindings)
Performs AI-powered analysis on file content.
async analyzeFile(
filePath: string,
content: string,
patternFindings: Finding[]
): Promise<Finding[]>
filePath (string): Path to the filecontent (string): File contentpatternFindings (Finding[]): Findings from pattern scanner (context)
Returns: Promise resolving to array of findings
Example:
const LLMAnalyzer = require('kafkacode/dist/LLMAnalyzer');
const PatternScanner = require('kafkacode/dist/PatternScanner');
const patternScanner = new PatternScanner();
const llmAnalyzer = new LLMAnalyzer();
const content = fs.readFileSync('./src/config.js', 'utf-8');
// First, pattern scan
const patternFindings = patternScanner.scanContent('./src/config.js', content);
// Then, LLM analysis
const llmFindings = await llmAnalyzer.analyzeFile(
'./src/config.js',
content,
patternFindings
);
console.log(`LLM found ${llmFindings.length} additional issues`);
How It Works
The LLM Analyzer:
- Takes context from pattern findings - Uses initial pattern matches to focus analysis
- Analyzes code semantically - Understands code context and intent
- Identifies subtle issues - Finds privacy concerns that patterns miss
- Reduces false positives - Filters out non-issues based on context
- Provides specific recommendations - Gives actionable advice
Usage Examples
Basic LLM Analysis
const analyzer = new LLMAnalyzer();
const code = `
const config = {
apiKey: process.env.API_KEY, // Good
password: "hardcoded123" // Bad
};
`;
const findings = await analyzer.analyzeFile('config.js', code, []);
findings.forEach(f => {
console.log(`${f.severity}: ${f.description}`);
console.log(`Recommendation: ${f.recommendation}`);
});
Combined Analysis
const patternScanner = new PatternScanner();
const llmAnalyzer = new LLMAnalyzer();
async function fullAnalysis(filePath) {
const content = fs.readFileSync(filePath, 'utf-8');
// Pattern-based detection
const patternFindings = patternScanner.scanContent(filePath, content);
// AI-powered analysis
const llmFindings = await llmAnalyzer.analyzeFile(
filePath,
content,
patternFindings
);
return [...patternFindings, ...llmFindings];
}
The LLM Analyzer helps reduce false positives by understanding code context, such as distinguishing between test data and real credentials.
Next Steps
