GitHub Copilot SDK Node.js Instructions

Instruction file imported from tcardosoMSFT/ADOrepos2GHE (.github/instructions/copilot-sdk.instructions.md). Copyright stays with the author.

Architecture Boundary — CRITICAL

• Native app & CLI (apps/native-migrator/, scripts/, packages/migration-core/): Use @github/copilot-sdk
• Chrome extension (extension/): Does NOT support Copilot SDK. Uses GitHub Models API (OpenAI-compatible https://models.inference.ai.azure.com/chat/completions) via direct fetch() with GitHub PAT auth.
• NEVER import @github/copilot-sdk or @ado-migrator/migration-core in extension/ code.
• The extension runs in a browser service worker — Node.js-only modules like the Copilot SDK will not work there.

Core Principles

• The SDK is in technical preview and may have breaking changes
• Requires Node.js 18.0 or later
• Requires GitHub Copilot CLI installed and in PATH
• Built with TypeScript for type safety
• Uses async/await patterns throughout
• Provides full TypeScript type definitions

Installation

Always install via npm/pnpm/yarn:

npm install @github/copilot-sdk
# or
pnpm add @github/copilot-sdk
# or
yarn add @github/copilot-sdk

Client Initialization

Basic Client Setup

import { CopilotClient } from "@github/copilot-sdk";

const client = new CopilotClient();
await client.start();
// Use client...
await client.stop();

Client Configuration Options

When creating a CopilotClient, use CopilotClientOptions:

• cliPath - Path to CLI executable (default: "copilot" from PATH)
• cliArgs - Extra arguments prepended before SDK-managed flags (string[])
• cliUrl - URL of existing CLI server (e.g., "localhost:8080"). When provided, client won't spawn a process
• port - Server port (default: 0 for random)
• useStdio - Use stdio transport instead of TCP (default: true)
• logLevel - Log level (default: "debug")
• autoStart - Auto-start server (default: true)
• autoRestart - Auto-restart on crash (default: true)
• cwd - Working directory for the CLI process (default: process.cwd())
• env - Environment variables for the CLI process (default: process.env)

Manual Server Control

For explicit control:

const client = new CopilotClient({ autoStart: false });
await client.start();
// Use client...
await client.stop();

Use forceStop() when stop() takes too long.

Session Management

Creating Sessions

Use SessionConfig for configuration:

const session = await client.createSession({
    model: "gpt-5",
    streaming: true,
    tools: [...],
    systemMessage: { ... },
    availableTools: ["tool1", "tool2"],
    excludedTools: ["tool3"],
    provider: { ... }
});

Session Config Options

• sessionId - Custom session ID (string)
• model - Model name ("gpt-5", "claude-sonnet-4.5", etc.)
• tools - Custom tools exposed to the CLI (Tool[])
• systemMessage - System message customization (SystemMessageConfig)
• availableTools - Allowlist of tool names (string[])
• excludedTools - Blocklist of tool names (string[])
• provider - Custom API provider configuration (BYOK) (ProviderConfig)
• streaming - Enable streaming response chunks (boolean)
• mcpServers - MCP server configurations (MCPServerConfig[])
• customAgents - Custom agent configurations (CustomAgentConfig[])
• configDir - Config directory override (string)
• skillDirectories - Skill directories (string[])
• disabledSkills - Disabled skills (string[])
• onPermissionRequest - Permission request handler (PermissionHandler)

Resuming Sessions

const session = await client.resumeSession("session-id", {
  tools: [myNewTool],
});

Session Operations

• session.sessionId - Get session identifier (string)
• await session.send({ prompt: "...", attachments: [...] }) - Send message, returns Promise
• await session.sendAndWait({ prompt: "..." }, timeout) - Send and wait for idle, returns Promise<AssistantMessageEvent | null>
• await session.abort() - Abort current processing
• await session.getMessages() - Get all events/messages, returns Promise<SessionEvent[]>
• await session.destroy() - Clean up session

Event Handling

Event Subscription Pattern

ALWAYS use async/await or Promises for waiting on session events:

await new Promise<void>((resolve) => {
  session.on((event) => {
    if (event.type === "assistant.message") {
      console.log(event.data.content);
    } else if (event.type === "session.idle") {
      resolve();
    }
  });

  session.send({ prompt: "..." });
});

Unsubscribing from Events

The on() method returns a function that unsubscribes:

const unsubscribe = session.on((event) => {
  // handler
});
// Later...
unsubscribe();

Event Types

Use discriminated unions with type guards for event handling:

session.on((event) => {
  switch (event.type) {
    case "user.message":
      // Handle user message
      break;
    case "assistant.message":
      console.log(event.data.content);
      break;
    case "tool.executionStart":
      // Tool execution started
      break;
    case "tool.executionComplete":
      // Tool execution completed
      break;
    case "session.start":
      // Session started
      break;
    case "session.idle":
      // Session is idle (processing complete)
      break;
    case "session.error":
      console.error(`Error: ${event.data.message}`);
      break;
  }
});

Streaming Responses

Enabling Streaming

Set streaming: true in SessionConfig:

const session = await client.createSession({
  model: "gpt-5",
  streaming: true,
});

Handling Streaming Events

Handle both delta events (incremental) and final events:

await new Promise<void>((resolve) => {
  session.on((event) => {
    switch (event.type) {
      case "assistant.message.delta":
        // Incremental text chunk
        process.stdout.write(event.data.deltaContent);
        break;
      case "assistant.reasoning.delta":
        // Incremental reasoning chunk (model-dependent)
        process.stdout.write(event.data.deltaContent);
        break;
      case "assistant.message":
        // Final complete message
        console.log("\n--- Final ---");
        console.log(event.data.content);
        break;
      case "assistant.reasoning":
        // Final reasoning content
        console.log("--- Reasoning ---");
        console.log(event.data.content);
        break;
      case "session.idle":
        resolve();
        break;
    }
  });

  session.send({ prompt: "Tell me a story" });
});

Note: Final events (assistant.message, assistant.reasoning) are ALWAYS sent regardless of streaming setting.

Custom Tools

Defining Tools with defineTool

Use defineTool for type-safe tool definitions:

import { defineTool } from "@github/copilot-sdk";

const session = await client.createSession({
  model: "gpt-5",
  tools: [
    defineTool({
      name: "lookup_issue",
      description: "Fetch issue details from tracker",
      parameters: {
        type: "object",
        properties: {
          id: { type: "string", description: "Issue ID" },
        },
        required: ["id"],
      },
      handler: async (args) => {
        const issue = await fetchIssue(args.id);
        return issue;
      },
    }),
  ],
});

Using Zod for Parameters

The SDK supports Zod schemas for parameters:

import { z } from "zod";

const session = await client.createSession({
  tools: [
    defineTool({
      name: "get_weather",
      description: "Get weather for a location",
      parameters: z.object({
        location: z.string().describe("City name"),
        units: z.enum(["celsius", "fahrenheit"]).optional(),
      }),
      handler: async (args) => {
        return { temperature: 72, units: args.units || "fahrenheit" };
      },
    }),
  ],
});

Tool Return Types

• Return any JSON-serializable value (automatically wrapped)
• Or return ToolResultObject for full control over metadata:

{
    textResultForLlm: string;  // Result shown to LLM
    resultType: "success" | "failure";
    error?: string;  // Internal error (not shown to LLM)
    toolTelemetry?: Record<string, unknown>;
}

Tool Execution Flow

When Copilot invokes a tool, the client automatically:

1. Runs your handler function
2. Serializes the return value
3. Responds to the CLI

System Message Customization

Append Mode (Default - Preserves Guardrails)

const session = await client.createSession({
  model: "gpt-5",
  systemMessage: {
    mode: "append",
    content: `
<workflow_rules>
- Always check for security vulnerabilities
- Suggest performance improvements when applicable
</workflow_rules>
`,
  },
});

Replace Mode (Full Control - Removes Guardrails)

const session = await client.createSession({
  model: "gpt-5",
  systemMessage: {
    mode: "replace",
    content: "You are a helpful assistant.",
  },
});

File Attachments

Attach files to messages:

await session.send({
  prompt: "Analyze this file",
  attachments: [
    {
      type: "file",
      path: "/path/to/file.ts",
      displayName: "My File",
    },
  ],
});

Message Delivery Modes

Use the mode property in message options:

• "enqueue" - Queue message for processing
• "immediate" - Process message immediately

await session.send({
  prompt: "...",
  mode: "enqueue",
});

Multiple Sessions

Sessions are independent and can run concurrently:

const session1 = await client.createSession({ model: "gpt-5" });
const session2 = await client.createSession({ model: "claude-sonnet-4.5" });

await Promise.all([
  session1.send({ prompt: "Hello from session 1" }),
  session2.send({ prompt: "Hello from session 2" }),
]);

Bring Your Own Key (BYOK)

Use custom API providers via provider:

const session = await client.createSession({
  provider: {
    type: "openai",
    baseUrl: "https://api.openai.com/v1",
    apiKey: "your-api-key",
  },
});

Session Lifecycle Management

Listing Sessions

const sessions = await client.listSessions();
for (const metadata of sessions) {
  console.log(`${metadata.sessionId}: ${metadata.summary}`);
}

Deleting Sessions

await client.deleteSession(sessionId);

Getting Last Session ID

const lastId = await client.getLastSessionId();
if (lastId) {
  const session = await client.resumeSession(lastId);
}

Checking Connection State

const state = client.getState();
// Returns: "disconnected" | "connecting" | "connected" | "error"

Error Handling

Standard Exception Handling

try {
  const session = await client.createSession();
  await session.send({ prompt: "Hello" });
} catch (error) {
  console.error(`Error: ${error.message}`);
}

Session Error Events

Monitor session.error event type for runtime errors:

session.on((event) => {
  if (event.type === "session.error") {
    console.error(`Session Error: ${event.data.message}`);
  }
});

Connectivity Testing

Use ping to verify server connectivity:

const response = await client.ping("health check");
console.log(`Server responded at ${new Date(response.timestamp)}`);

Resource Cleanup

Automatic Cleanup with Try-Finally

ALWAYS use try-finally or cleanup in a finally block:

const client = new CopilotClient();
try {
  await client.start();
  const session = await client.createSession();
  try {
    // Use session...
  } finally {
    await session.destroy();
  }
} finally {
  await client.stop();
}

Cleanup Function Pattern

async function withClient<T>(
  fn: (client: CopilotClient) => Promise<T>,
): Promise<T> {
  const client = new CopilotClient();
  try {
    await client.start();
    return await fn(client);
  } finally {
    await client.stop();
  }
}

async function withSession<T>(
  client: CopilotClient,
  fn: (session: CopilotSession) => Promise<T>,
): Promise<T> {
  const session = await client.createSession();
  try {
    return await fn(session);
  } finally {
    await session.destroy();
  }
}

// Usage
await withClient(async (client) => {
  await withSession(client, async (session) => {
    await session.send({ prompt: "Hello!" });
  });
});

Best Practices

1. Always use try-finally for resource cleanup
2. Use Promises to wait for session.idle event
3. Handle session.error events for robust error handling
4. Use type guards or switch statements for event handling
5. Enable streaming for better UX in interactive scenarios
6. Use defineTool for type-safe tool definitions
7. Use Zod schemas for runtime parameter validation
8. Dispose event subscriptions when no longer needed
9. Use systemMessage with mode: "append" to preserve safety guardrails
10. Handle both delta and final events when streaming is enabled
11. Leverage TypeScript types for compile-time safety

Common Patterns

Simple Query-Response

import { CopilotClient } from "@github/copilot-sdk";

const client = new CopilotClient();
try {
  await client.start();

  const session = await client.createSession({ model: "gpt-5" });
  try {
    await new Promise<void>((resolve) => {
      session.on((event) => {
        if (event.type === "assistant.message") {
          console.log(event.data.content);
        } else if (event.type === "session.idle") {
          resolve();
        }
      });

      session.send({ prompt: "What is 2+2?" });
    });
  } finally {
    await session.destroy();
  }
} finally {
  await client.stop();
}

Multi-Turn Conversation

const session = await client.createSession();

async function sendAndWait(prompt: string): Promise<void> {
  await new Promise<void>((resolve, reject) => {
    const unsubscribe = session.on((event) => {
      if (event.type === "assistant.message") {
        console.log(event.data.content);
      } else if (event.type === "session.idle") {
        unsubscribe();
        resolve();
      } else if (event.type === "session.error") {
        unsubscribe();
        reject(new Error(event.data.message));
      }
    });

    session.send({ prompt });
  });
}

await sendAndWait("What is the capital of France?");
await sendAndWait("What is its population?");

SendAndWait Helper

// Use built-in sendAndWait for simpler synchronous interaction
const response = await session.sendAndWait({ prompt: "What is 2+2?" }, 60000);

if (response) {
  console.log(response.data.content);
}

Tool with Type-Safe Parameters

import { z } from "zod";
import { defineTool } from "@github/copilot-sdk";

interface UserInfo {
  id: string;
  name: string;
  email: string;
  role: string;
}

const session = await client.createSession({
  tools: [
    defineTool({
      name: "get_user",
      description: "Retrieve user information",
      parameters: z.object({
        userId: z.string().describe("User ID"),
      }),
      handler: async (args): Promise<UserInfo> => {
        return {
          id: args.userId,
          name: "John Doe",
          email: "john@example.com",
          role: "Developer",
        };
      },
    }),
  ],
});

Streaming with Progress

let currentMessage = "";

const unsubscribe = session.on((event) => {
  if (event.type === "assistant.message.delta") {
    currentMessage += event.data.deltaContent;
    process.stdout.write(event.data.deltaContent);
  } else if (event.type === "assistant.message") {
    console.log("\n\n=== Complete ===");
    console.log(`Total length: ${event.data.content.length} chars`);
  } else if (event.type === "session.idle") {
    unsubscribe();
  }
});

await session.send({ prompt: "Write a long story" });

Error Recovery

session.on((event) => {
  if (event.type === "session.error") {
    console.error("Session error:", event.data.message);
    // Optionally retry or handle error
  }
});

try {
  await session.send({ prompt: "risky operation" });
} catch (error) {
  // Handle send errors
  console.error("Failed to send:", error);
}

TypeScript-Specific Features

Type Inference

import type { SessionEvent, AssistantMessageEvent } from "@github/copilot-sdk";

session.on((event: SessionEvent) => {
  if (event.type === "assistant.message") {
    // TypeScript knows event is AssistantMessageEvent here
    const content: string = event.data.content;
  }
});

Generic Helper

async function waitForEvent<T extends SessionEvent["type"]>(
  session: CopilotSession,
  eventType: T,
): Promise<Extract<SessionEvent, { type: T }>> {
  return new Promise((resolve) => {
    const unsubscribe = session.on((event) => {
      if (event.type === eventType) {
        unsubscribe();
        resolve(event as Extract<SessionEvent, { type: T }>);
      }
    });
  });
}

// Usage
const message = await waitForEvent(session, "assistant.message");
console.log(message.data.content);