telemetry

Instruction file imported from aryanjsingh/oasis (.github/instructions/telemetry.instructions.md). Copyright stays with the author.

Patterns for GDPR-compliant telemetry in VS Code with proper type safety and privacy protection.

Implementation Pattern

1. Define Types

type MyFeatureEvent = {
    action: string;
    duration: number;
    success: boolean;
    errorCode?: string;
};

type MyFeatureClassification = {
    action: { classification: 'SystemMetaData'; purpose: 'FeatureInsight'; comment: 'The action performed.' };
    duration: { classification: 'SystemMetaData'; purpose: 'PerformanceAndHealth'; isMeasurement: true; comment: 'Time in milliseconds.' };
    success: { classification: 'SystemMetaData'; purpose: 'FeatureInsight'; isMeasurement: true; comment: 'Whether action succeeded.' };
    errorCode: { classification: 'SystemMetaData'; purpose: 'PerformanceAndHealth'; comment: 'Error code if action failed.' };
    owner: 'yourGitHubUsername';
    comment: 'Tracks MyFeature usage and performance.';
};

2.1. Send Event

this.telemetryService.publicLog2<MyFeatureEvent, MyFeatureClassification>('myFeatureAction', {
    action: 'buttonClick',
    duration: 150,
    success: true
});

2.2. Error Events

For error-specific telemetry with stack traces or error messages:

type MyErrorEvent = {
    operation: string;
    errorMessage: string;
    duration?: number;
};

type MyErrorClassification = {
    operation: { classification: 'SystemMetaData'; purpose: 'PerformanceAndHealth'; comment: 'The operation that failed.' };
    errorMessage: { classification: 'CallstackOrException'; purpose: 'PerformanceAndHealth'; comment: 'The error message.' };
    duration: { classification: 'SystemMetaData'; purpose: 'PerformanceAndHealth'; isMeasurement: true; comment: 'Time until failure.' };
    owner: 'yourGitHubUsername';
    comment: 'Tracks MyFeature errors for reliability.';
};

this.telemetryService.publicLogError2<MyErrorEvent, MyErrorClassification>('myFeatureError', {
    operation: 'fileRead',
    errorMessage: error.message,
    duration: 1200
});

3. Service Injection

constructor(
    @ITelemetryService private readonly telemetryService: ITelemetryService,
) { super(); }

GDPR Classifications & Purposes

Classifications (choose the most restrictive):

• SystemMetaData - Most common. Non-personal system info, user preferences, feature usage, identifiers (extension IDs, language types, counts, durations, success flags)
• CallstackOrException - Error messages, stack traces, exception details. Only for actual error information.
• PublicNonPersonalData - Data already publicly available (rare)

Purposes (combine with different classifications):

• FeatureInsight - Default. Understanding how features are used, user behavior patterns, feature adoption
• PerformanceAndHealth - For errors & performance. Metrics, error rates, performance measurements, diagnostics

Required Properties:

• comment - Clear explanation of what the field contains and why it's collected
• owner - GitHub username (infer from branch or ask)
• isMeasurement: true - Required for all numeric values flags used in calculations

Error Events

Use publicLogError2 for errors with CallstackOrException classification:

this.telemetryService.publicLogError2<ErrorEvent, ErrorClassification>('myFeatureError', {
	errorMessage: error.message,
	errorCode: 'MYFEATURE_001',
	context: 'initialization'
});

Naming & Privacy Rules

Naming Conventions:

• Event names: camelCase with context (extensionActivationError, commandExecuted)
• Property names: specific and descriptive (extensionId not id, durationMs not duration)
• Common patterns: success/hasError/isEnabled, sessionId/extensionId, type/kind/source

Critical Don'ts:

• ❌ No PII (usernames, emails, file paths, content)
• ❌ Missing owner field in classification (infer from branch name or ask user)
• ❌ Vague comments ("user data" → "selected language identifier")
• ❌ Wrong classification
• ❌ Missing isMeasurement on numeric metrics

Privacy Requirements:

• Minimize data collection to essential insights only
• Use hashes/categories instead of raw values when possible
• Document clear purpose for each data point