Chat Participant Patterns

Chat Participant Patterns Skill

VS Code Chat API patterns.

⚠️ Staleness Warning

Chat APIs evolve with VS Code releases. Last validated: March 2026 (VS Code 1.111+)

Check: Chat API, LM API, Tools API, AI Extensibility Overview, EXTERNAL-API-REGISTRY.md

When to Use Chat Participant vs. Skills/Tools

→ Prefer Skills for domain knowledge. Reserve participants for full prompt orchestration.

Create Participant

// package.json contribution
"contributes": {
    "chatParticipants": [{
        "id": "my-ext.participant",
        "name": "myparticipant",
        "fullName": "My Participant",
        "description": "What can I help with?",
        "isSticky": true,
        "commands": [{ "name": "help", "description": "Get help" }]
    }]
}

// In activate()
const participant = vscode.chat.createChatParticipant('my-ext.participant', handler);
participant.iconPath = vscode.Uri.joinPath(context.extensionUri, 'icon.png');

Handler Signature

const handler: vscode.ChatRequestHandler = async (
    request: vscode.ChatRequest,
    context: vscode.ChatContext,
    stream: vscode.ChatResponseStream,
    token: vscode.CancellationToken
): Promise<IChatResult> => {
    // Handle request
};

Key Operations

Response Types

// Markdown (supports CommonMark)
stream.markdown('# Title\n**bold** and _italic_');

// Code block with IntelliSense
stream.markdown('```typescript\nconst x = 1;\n```');

// Progress message
stream.progress('Processing...');

// Button (invokes VS Code command)
stream.button({ command: 'my.command', title: 'Run' });

// Command link in markdown
const md = new vscode.MarkdownString('[Run](command:my.command)');
md.isTrusted = { enabledCommands: ['my.command'] };
stream.markdown(md);

// File tree
stream.filetree([{ name: 'src', children: [{ name: 'app.ts' }] }], baseUri);

// Reference
stream.reference(vscode.Uri.file('/path/to/file.ts'));
stream.reference(new vscode.Location(uri, range));

LM Integration

const models = await vscode.lm.selectChatModels({ vendor: 'copilot' });
const response = await models[0].sendRequest(messages, {}, token);
for await (const chunk of response.text) {
    stream.markdown(chunk);
}

Tool Calling

// Using @vscode/chat-extension-utils library (recommended)
import * as chatUtils from '@vscode/chat-extension-utils';

const tools = vscode.lm.tools.filter(t => t.tags.includes('my-tag'));
const result = chatUtils.sendChatParticipantRequest(request, context, {
    prompt: 'System instructions here',
    responseStreamOptions: { stream, references: true, responseText: true },
    tools
}, token);
return await result.result;

Tool Registration

vscode.lm.registerTool('tool_name', {
    async invoke(options, token) {
        return new vscode.LanguageModelToolResult([
            new vscode.LanguageModelTextPart('result text')
        ]);
    }
});

Participant Detection (Auto-routing)

"chatParticipants": [{
    "id": "my-ext.participant",
    "disambiguation": [{
        "category": "my-domain",
        "description": "Questions about X domain",
        "examples": ["How do I do X?", "Explain Y concept"]
    }]
}]

Follow-up Provider

participant.followupProvider = {
    provideFollowups(result, context, token) {
        return [{ prompt: 'Tell me more', label: 'More details' }];
    }
};

Message History

// Get previous requests to this participant
const previousRequests = context.history.filter(
    h => h instanceof vscode.ChatRequestTurn
);

Best Practices