vscode-extension-patterns

name	vscode-extension-patterns
description	Reusable patterns for VS Code extension development.

VS Code Extension Patterns Skill

Reusable patterns for VS Code extension development.

⚠️ Staleness Warning

VS Code APIs evolve with each monthly release. Patterns may become outdated or better alternatives may emerge.

Refresh triggers:

• VS Code major/minor releases
• New proposed APIs becoming stable
• Extension API deprecations
• Webview security policy changes

Last validated: February 2026 (VS Code 1.109+)

Check current state: VS Code API, Release Notes

---

Webview Dashboard

// Gather data in parallel, build HTML with async
const [health, knowledge, sync] = await Promise.all([
    checkHealth(true), getKnowledgeSummary(), getSyncStatus()
]);
panel.webview.html = await getWebviewContent(health, knowledge, sync);

Key: Make getWebviewContent async if it needs directory scanning or other async ops.

TreeDataProvider for Sidebar

class WelcomeViewProvider implements vscode.WebviewViewProvider {
    resolveWebviewView(webviewView: vscode.WebviewView) {
        webviewView.webview.options = { enableScripts: true };
        webviewView.webview.html = this.getHtmlContent();
        webviewView.webview.onDidReceiveMessage(async (message) => {
            switch (message.command) {
                case 'refresh': await this.refresh(); break;
            }
        });
    }
}

// Register in extension.ts
vscode.window.registerWebviewViewProvider('alex.welcomeView', new WelcomeViewProvider());

CSP-Compliant Webview Event Handling

Problem: Inline event handlers (onclick="...") violate Content Security Policy and can be blocked.

Solution: Use data-cmd attributes with delegated event listeners:

<!-- ❌ WRONG: Inline handlers (CSP violation) -->
<button onclick="handleClick()">Click</button>

<!-- ✅ CORRECT: data-cmd pattern -->
<button data-cmd="play">Play</button>
<button data-cmd="stop">Stop</button>

// Single delegated listener for all commands
document.addEventListener('click', (e) => {
    const target = e.target.closest('[data-cmd]');
    if (!target) return;
    
    const cmd = target.getAttribute('data-cmd');
    switch (cmd) {
        case 'play': audio.play(); break;
        case 'stop': audio.pause(); audio.currentTime = 0; break;
    }
});

Benefits:

• CSP-compliant (no inline scripts)
• Single event listener (better performance)
• Easy to add new commands
• Consistent pattern across webviews

Safe Configuration Pattern

Tiered settings: Essential (🔴) → Recommended (🟡) → Auto-Approval (🟠) → Extended Thinking (🧠) → Enterprise (🔵)

Safety rules:

• Additive only — never modify/remove existing
• Check config.inspect(key)?.globalValue before applying
• Preview JSON before changes
• User chooses categories
• Preserve higher user values when applying recommendations

async function applySettings(settings: Record<string, unknown>) {
    const config = vscode.workspace.getConfiguration();
    for (const [key, value] of Object.entries(settings)) {
        if (config.inspect(key)?.globalValue === undefined) {
            await config.update(key, value, vscode.ConfigurationTarget.Global);
        }
    }
}

Comprehensive Settings Management Pattern

Pattern: Documentation-first approach for complex settings ecosystems (VS Code 1.109+ chat settings).

Steps:

1. Research Phase — Comprehensive discovery

   • Grep search for all related settings in codebase
   • Read MS documentation for official settings
   • Identify experimental/unstable features
   • Document current user configuration

2. Documentation Phase — Create reference materials before implementation

   • GUIDE: Comprehensive reference (all settings, categories, use cases, warnings)
   • SUMMARY: Current state snapshot (what user has, what's missing, recommendations)
   • APPLIED: Change log document (what was applied, why, how to rollback)
   • JSON Template: Copy-paste ready configuration file

3. Implementation Phase — Safe automated application

   • Create PowerShell/shell script for automation (platform-specific)
   • Always backup before modifications (settings.json.backup-{timestamp})
   • Compare new vs existing, report changes (new, updated, skipped, preserved)
   • Preserve higher values (user has 150, recommending 100? Keep 150)
   • Exclude unstable features explicitly documented

4. Extension Integration — Make it permanent

   • Update setupEnvironment.ts ESSENTIAL_SETTINGS, RECOMMENDED_SETTINGS, etc.
   • Add new categories if pattern discovered (e.g., AUTO_APPROVAL_SETTINGS)
   • Integrate into Initialize/Upgrade commands (offerEnvironmentSetup())
   • Update Welcome sidebar with accurate counts

5. Validation Phase

   • Test script execution (fix syntax errors iteratively)
   • Verify settings applied (check VS Code settings.json)
   • Document excluded settings with reasons (hooks not stable, platform-specific)
   • Commit documentation files to git

Key insights from Feb 2026 implementation:

• VS Code 1.109+ has 47+ chat-related settings across 6 categories
• Hooks (chat.hooks.enabled) marked experimental but not stable yet
• Auto-approval settings reduce friction (5 settings: autoRun, fileSystem.autoApprove, terminal.*)
• Initialize/Upgrade should apply Essential + Recommended + Auto-Approval (36 total)
• PowerShell inline commands unreliable — use .ps1 file for complex scripts
• Settings categories prevent overwhelming users (show 5 categories vs 47 individual settings)

When to use this pattern:

• Complex settings ecosystems (10+ related settings)
• Rapidly evolving features (experimental → stable transitions)
• User education needed (settings have non-obvious interdependencies)
• Safety-critical configuration (wrong settings break functionality)

Template structure:

docs/guides/
├── FEATURE-SETTINGS-GUIDE.md         # Comprehensive reference
├── FEATURE-SETTINGS-SUMMARY.md       # Current state snapshot  
└── FEATURE-SETTINGS-APPLIED.md       # Change log (gitignore if sensitive)

.vscode/
└── recommended-feature-settings.jsonc # Template (gitignore)

src/commands/
└── setupEnvironment.ts                # Settings constants + apply logic

Benefits:

• Users understand full capability before committing
• Documentation serves as reference post-application
• Safe rollback via timestamped backups
• Extension automatically applies for new users
• Audit trail of what was applied and why

Auto-Detection with Confidence

const PATTERNS = [
    { pattern: /learned|discovered|realized/i, confidence: 0.8 },
    { pattern: /key insight|the trick is/i, confidence: 0.85 },
];

Use confidence thresholds for auto-actions. Higher threshold = fewer false positives.

Duplicate Detection

function isDuplicate(newText: string, existing: string[]): boolean {
    const normalize = (s: string) => s.toLowerCase().replace(/[^\w\s]/g, '');
    return existing.some(e => calculateSimilarity(normalize(newText), normalize(e)) > 0.8);
}

Portability Rules

Extensions must work on any machine:

// ✅ CORRECT: Dynamic paths
const rootPath = vscode.workspace.workspaceFolders?.[0].uri.fsPath;
const globalPath = path.join(os.homedir(), '.alex');

// ❌ WRONG: Hardcoded paths
const rootPath = 'c:\\Development\\MyProject';  // Never!

Key utilities:

• vscode.workspace.workspaceFolders — Current workspace
• os.homedir() — Platform-independent home
• path.join() — Cross-platform path building

Publishing Workflow

# Load PAT from .env
$env:VSCE_PAT = (Get-Content .env | Select-String "VSCE_PAT" | ForEach-Object { $_.Line.Split("=",2)[1] })
vsce publish

Version collision: Increment patch → update package.json, README badge, CHANGELOG → retry.

Goals with Streak Tracking

interface LearningGoal {
    id: string;
    title: string;
    category: 'coding' | 'reading' | 'practice' | 'review';
    targetCount: number;
    currentCount: number;
    type: 'daily' | 'weekly';
    expiresAt: string;
}

// Auto-increment on activity
async function autoIncrementGoals(activityType: 'session' | 'insight') {
    const data = await loadGoalsData();
    for (const goal of data.goals) {
        if (shouldIncrement(goal, activityType) && !isExpired(goal)) {
            goal.currentCount = Math.min(goal.currentCount + 1, goal.targetCount);
        }
    }
    await saveGoalsData(data);
}

SecretStorage for Sensitive Tokens

Never store secrets in settings — use VS Code's SecretStorage API:

// Module-level cache
let secretStorage: vscode.SecretStorage | null = null;
let cachedToken: string | null = null;

// Initialize during activation
export async function initSecrets(context: vscode.ExtensionContext): Promise<void> {
    secretStorage = context.secrets;
    cachedToken = await secretStorage.get('myExtension.apiToken') || null;
    
    // Migration: Move token from settings to secrets
    const config = vscode.workspace.getConfiguration('myExtension');
    const settingsToken = config.get<string>('apiToken')?.trim();
    if (settingsToken && !cachedToken) {
        await secretStorage.store('myExtension.apiToken', settingsToken);
        cachedToken = settingsToken;
        await config.update('apiToken', undefined, vscode.ConfigurationTarget.Global);
        vscode.window.showInformationMessage('Token migrated to secure storage.');
    }
}

// Synchronous access to cached value
function getToken(): string | null {
    return cachedToken;
}

Key points:

• context.secrets.get() / store() / delete() are async
• Cache at module level for sync access
• Migrate existing settings tokens on first run
• Mark old setting as deprecated in package.json

Webview CSP Security

Always add Content-Security-Policy when enableScripts: true:

import { getNonce } from './sanitize';

function getWebviewHtml(webview: vscode.Webview): string {
    const nonce = getNonce();
    return `<!DOCTYPE html>
<html>
<head>
    <meta http-equiv="Content-Security-Policy" content="
        default-src 'none';
        style-src ${webview.cspSource} 'unsafe-inline';
        script-src 'nonce-${nonce}';
        img-src ${webview.cspSource} https: data:;
        font-src ${webview.cspSource};
    ">
</head>
<body>
    <script nonce="${nonce}">
        const vscode = acquireVsCodeApi();
        // ... your code
    </script>
</body>
</html>`;
}

// Nonce generator
function getNonce(): string {
    const chars = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789';
    return Array.from({ length: 32 }, () => 
        chars.charAt(Math.floor(Math.random() * chars.length))
    ).join('');
}

CSP Event Delegation (onclick → data-cmd)

Problem: After adding CSP with script-src 'nonce-${nonce}', all inline event handlers (onclick, onchange) stop working because CSP blocks inline JavaScript.

Solution: Replace inline handlers with data attributes and event delegation:

<!-- ❌ BLOCKED BY CSP -->
<button onclick="cmd('upgrade')">Upgrade</button>
<button onclick="cmd('launchSkill', {skill: 'code-review'})">Review</button>

<!-- ✅ CSP-COMPLIANT -->
<button data-cmd="upgrade">Upgrade</button>
<button data-cmd="launchSkill" data-skill="code-review">Review</button>

<script nonce="${nonce}">
    document.addEventListener('click', function(e) {
        const el = e.target.closest('[data-cmd]');
        if (el) {
            e.preventDefault();
            const command = el.getAttribute('data-cmd');
            const skill = el.getAttribute('data-skill');
            vscode.postMessage(skill ? { command, skill } : { command });
        }
    });
</script>

Benefits:

• Security — CSP blocks all inline scripts
• Performance — Single event listener vs many handlers
• Maintainability — Commands defined as data, not code

Webview Sandbox: postMessage Required

Problem: window.open(), location.reload(), and direct external links silently fail in sandboxed webviews.

Solution: WebView sends message to extension host; extension performs privileged action:

// In webview HTML
vscode.postMessage({ type: 'openExternal', url: 'https://example.com' });

// In extension host
panel.webview.onDidReceiveMessage(async (message) => {
    if (message.type === 'openExternal') {
        await vscode.env.openExternal(vscode.Uri.parse(message.url));
    }
});

Key insight: WebView ↔ Extension Host communication mirrors browser Content Script ↔ Background Script patterns.

Telemetry Opt-Out Compliance

Always respect VS Code's telemetry settings:

function isTelemetryEnabled(): boolean {
    // Check VS Code global setting first
    if (!vscode.env.isTelemetryEnabled) {
        return false;
    }
    // Then check extension-specific setting
    const config = vscode.workspace.getConfiguration('myExtension');
    return config.get<boolean>('telemetry.enabled', true);
}

function log(event: string, data?: Record<string, unknown>): void {
    if (!isTelemetryEnabled()) {
        return;
    }
    // Send telemetry...
}

Configuration Change Listeners

React to settings changes at runtime:

export function activate(context: vscode.ExtensionContext) {
    // Listen for configuration changes
    context.subscriptions.push(
        vscode.workspace.onDidChangeConfiguration(e => {
            if (e.affectsConfiguration('myExtension.featureA')) {
                // Refresh feature A
                refreshFeatureA();
            }
            if (e.affectsConfiguration('myExtension.telemetry')) {
                // Update telemetry state
            }
        })
    );
}

Key points:

• Use affectsConfiguration() to filter relevant changes
• Push listener to context.subscriptions for cleanup
• Re-read config values, don't cache indefinitely

VS Code 1.109+ Agent Platform Capabilities

VS Code 1.109 introduces a native agent platform that extensions can leverage:

Agent Files (AGENTS.md)

Extensions can ship agent definitions that VS Code auto-discovers:

<!-- .github/agents/my-agent.agent.md -->
---
name: "MyAgent"
description: "Specialized agent for domain X"
---

# MyAgent Instructions

Agent-specific instructions and knowledge go here.

Setting: chat.useAgentsMdFile: true enables automatic loading.

Skills Loading

Extensions can define skills in .github/skills/ that are auto-loaded into chat:

Setting: chat.agentSkillsLocations: [".github/skills"]

Each skill folder contains a SKILL.md (knowledge) and optional synapses.json (connections).

Chat Participant API

Register custom chat participants that users can @mention:

const participant = vscode.chat.createChatParticipant('myext.agent', async (request, context, stream, token) => {
  // Access to request.prompt, request.command
  // Stream responses with stream.markdown(), stream.button(), stream.reference()
  stream.markdown('Hello from my agent!');
});

participant.iconPath = vscode.Uri.joinPath(context.extensionUri, 'icon.png');
context.subscriptions.push(participant);

Tool Registration

Register tools that any chat participant can invoke:

const tool = vscode.lm.registerTool('myext-searchDocs', {
  async invoke(options, token) {
    const query = options.input.query;
    // Perform tool action
    return new vscode.LanguageModelToolResult([
      new vscode.LanguageModelTextPart(JSON.stringify(results))
    ]);
  }
});
context.subscriptions.push(tool);

Declare in package.json:

{
  "contributes": {
    "languageModelTools": [{
      "name": "myext-searchDocs",
      "displayName": "Search Documentation",
      "modelDescription": "Searches project documentation for relevant content",
      "inputSchema": {
        "type": "object",
        "properties": {
          "query": { "type": "string", "description": "Search query" }
        },
        "required": ["query"]
      }
    }]
  }
}

Extended Thinking

Models supporting extended thinking can be configured per-model:

{
  "claude-opus-4-*.extendedThinkingEnabled": true,
  "claude-opus-4-*.thinkingBudget": 16384
}

MCP Integration

VS Code 1.109+ supports Model Context Protocol servers:

Setting: chat.mcp.gallery.enabled: true

MCP servers extend AI capabilities with external tools (Azure, GitHub, databases).

Key 1.109 Settings Summary

Setting	Value	Purpose
chat.agent.enabled	true	Enable custom agents
chat.agentSkillsLocations	[".github/skills"]	Auto-load skills
chat.useAgentsMdFile	true	Use AGENTS.md
chat.mcp.gallery.enabled	true	MCP tool access

Integration Audit Checklist

10-category audit scoring system (5 points each, 50 total):

#	Category	What to Check
1	Activation Events	package.json activationEvents match actual needs
2	Extension Context	context.subscriptions, secrets, globalState usage
3	Disposable Management	All disposables pushed to subscriptions
4	Command Registration	Commands in package.json match registerCommand
5	Configuration Access	getConfiguration usage, onDidChangeConfiguration
6	Webview Security	CSP policies, nonce usage, enableScripts
7	Language Model/Chat	vscode.lm patterns, tool registration
8	Telemetry	vscode.env.isTelemetryEnabled respected
9	Error Handling	try/catch patterns, error type handling
10	File System	vscode.workspace.fs vs Node.js fs

Quick wins (high impact, low effort):

• Telemetry opt-out: Check vscode.env.isTelemetryEnabled
• CSP on webviews: Add Content-Security-Policy with nonce
• Config listeners: Add onDidChangeConfiguration for runtime updates
• Secret storage: Use context.secrets instead of settings for tokens

Scoring:

• 45-50: Excellent — Ready for publish
• 40-44: Good — Minor fixes
• 35-39: Fair — Address before publish
• <35: Needs Work — Major refactoring

When to apply: Before marketplace publishing, after major features, quarterly reviews.

Synapses

See synapses.json for connections.