Preview SDK

This SDK simplifies the process of building an GitHub Copilot Extension as an agent. Building Copilot Extensions previously required manual handling of request verification, response formatting, and API interactions. This SDK simplifies these tasks, allowing you to focus on your extension's core functionality rather than building boilerplate code. Use it to integrate your tools, APIs, or data sources directly into Copilot Chat.

We consider this SDK alpha software in terms of API stability, but we adhere to semantic-versioning, so it's safe to use today.

Note

This SDK does not apply to skillset-based extensions. It only applies to building an agent. For more info, see the docs on skillsets and agents.

Key features

Request payload verification
Payload parsing
Response building

Benefits

Handles security and response formatting requirements
Provides utilities for common extension tasks
Streamlines the development process

Usage

Verify a request

import { verifyRequestByKeyId } from "@copilot-extensions/preview-sdk";

const { isValid, cache } = await verifyRequestByKeyId(
  request.body,
  signature,
  keyId,
  {
    token: process.env.GITHUB_TOKEN,
  },
);
// isValid: true or false
// cache: { id, keys }

Build a response

import { createAckEvent, createDoneEvent, createTextEvent } from "@copilot-extensions/preview-sdk";

export default handler(request, response) {
  const ackEvent = createAckEvent();
  const textEvent = createTextEvent("Hello, world!");
  const doneEvent = createDoneEvent();

  response.write(ackEvent);
  response.write(textEvent);
  response.end(doneEvent);
}

Send a custom prompt

import { prompt } from "@copilot-extensions/preview-sdk";

try {
  const { message } = await prompt("What is the capital of France?", {
    token: process.env.TOKEN,
  });

  console.log(message.content);
} catch (error) {
  console.error(error);
}

API

Verification

`async verifyRequestByKeyId(rawBody, signature, keyId, requestOptions)`

Verify the request payload using the provided signature and key ID. The method will request the public key from GitHub's API for the given keyId and then verify the payload.

The requestOptions argument is optional. It can contain:

a token to authenticate the request to GitHub's API
a custom octokit request instance to use for the request
a cache to use cached keys

import { verifyRequestByKeyId } from "@copilot-extensions/preview-sdk";

const { isValid, cache } = await verifyRequestByKeyId(
  request.body,
  signature,
  keyId,
);

// with token
const { isValid, cache } = await verifyRequestByKeyId(
  request.body,
  signature,
  keyId,
  { token: "ghp_1234" },
);

// with custom octokit request instance
const { isValid, cache } = await verifyRequestByKeyId(
  request.body,
  signature,
  keyId,
  { request },
);

// with cache
const previousCache = {
  id: "etag_value",
  keys: [{ key_identifier: "key1", key: "public_key1", is_current: true }],
};
const { isValid, cache } = await verifyRequestByKeyId(
  request.body,
  signature,
  keyId,
  { cache: previousCache },
);

`async fetchVerificationKeys(options)`

Fetches public keys for verifying copilot extension requests from GitHub's API and returns them as an array. The request can be made without authentication, with a token, with a custom octokit request instance, or with a cache.

import { fetchVerificationKeys } from "@copilot-extensions/preview-sdk";

// fetch without authentication
const { id, keys } = await fetchVerificationKeys();

// with token
const { id, keys } = await fetchVerificationKeys({ token: "ghp_1234" });

// with custom octokit request instance
const { id, keys } = await fetchVerificationKeys({ request });

// with cache
const cache = {
  id: "etag_value",
  keys: [{ key_identifier: "key1", key: "public_key1" }],
};
const { id, keys } = await fetchVerificationKeys({ cache });

`async verifyRequest(rawBody, signature, keyId)`

Verify the request payload using the provided signature and key. Note that the raw body as received by GitHub must be passed, before any parsing.

import { verify } from "@copilot-extensions/preview-sdk";

const payloadIsVerified = await verifyRequest(request.body, signature, key);
// true or false

Response

All create*Event() methods return a string that can directly be written to the response stream.

`createAckEvent()`

Acknowledge the request so that the chat UI can tell the user that the agent started generating a response. The ack event should only be sent once.

import { createAckEvent } from "@copilot-extensions/preview-sdk";

response.write(createAckEvent());

`createTextEvent(message)`

Send a text message to the chat UI. Multiple messages can be sent. The message argument must be a string and may include markdown.

import { createTextEvent } from "@copilot-extensions/preview-sdk";

response.write(createTextEvent("Hello, world!"));
response.write(createTextEvent("Hello, again!"));

`createConfirmationEvent({ id, title, message, metadata })`

Ask the user to confirm an action. The confirmation event should only be sent once.

The meta data object will be sent along the user's response.

See additional documentation about Copilot confirmations.

import { createConfirmationEvent } from "@copilot-extensions/preview-sdk";

response.write(
  createConfirmationEvent({
    id: "123",
    title: "Are you sure?",
    message: "This will do something.",
  }),
);

`createReferencesEvent(references)`

Send a list of references to the chat UI. The references argument must be an array of objects with the following properties:

id
type

The following properties are optional

data: object with any properties
is_implicit: a boolean
metadata: an object with a required display_name and the optional properties: display_icon and display_url

Multiple references events can be sent.

See additional documentation about Copilot references.

import { createReferencesEvent } from "@copilot-extensions/preview-sdk";

response.write(
  createReferencesEvent([
    {
      id: "123",
      type: "issue",
      data: {
        number: 123,
      },
      is_implicit: false,
      metadata: {
        display_name: "My issue",
        display_icon: "issue-opened",
        display_url: "https://github.com/monalisa/hello-world/issues/123",
    },
  ])
);

`createErrorsEvent(errors)`

An array of objects with the following properties:

type: must be one of: "reference", "function", "agent"
code
message
identifier

See additional documentation about Copilot errors.

`createDoneEvent()`

The done event should only be sent once, at the end of the response. No further events can be sent after the done event.

import { createDoneEvent } from "@copilot-extensions/preview-sdk";

response.write(createDoneEvent());

Parsing

`parseRequestBody(body)`

Parses the raw request body and returns an object with type support.

⚠️ It's well possible that the type is not 100% correct. Please send pull requests to index.d.ts to improve it

import { parseRequestBody } from "@copilot-extensions/preview-sdk";

const payload = parseRequestBody(rawBody);
// When your IDE supports types, typing "payload." should prompt the available keys and their types.

`transformPayloadForOpenAICompatibility()`

For cases when you want to pipe a user request directly to OpenAI, use this method to remove Copilot-specific fields from the request payload.

import { transformPayloadForOpenAICompatibility } from "@copilot-extensions/preview-sdk";
import { OpenAI } from "openai";

const openaiPayload = transformPayloadForOpenAICompatibility(payload);

const openai = new OpenAI({ apiKey: OPENAI_API_KEY });
const stream = openai.beta.chat.completions.stream({
  ...openaiPayload,
  model: "gpt-4-1106-preview",
  stream: true,
});

`verifyAndParseRequest()`

Convenience method to verify and parse a request in one go. It calls verifyRequestByKeyId() and parseRequestBody() internally.

import { verifyAndParseRequest } from "@copilot-extensions/preview-sdk";

const { isValidRequest, payload, cache } = await verifyAndParseRequest(
  request.body,
  signature,
  keyId,
  {
    token: process.env.GITHUB_TOKEN,
  },
);

if (!isValidRequest) {
  throw new Error("Request could not be verified");
}

// `isValidRequest` is a boolean.
// `payload` has type support.
// `cache` contains the id and keys used for verification.

`getUserMessage()`

Convenience method to get the user's message from the request payload.

import { getUserMessage } from "@copilot-extensions/preview-sdk";

const userMessage = getUserMessage(payload);

`getUserConfirmation()`

Convenience method to get the user's confirmation from the request payload (in case the user's last response was a confirmation).

import { getUserConfirmation } from "@copilot-extensions/preview-sdk";

const userConfirmation = getUserConfirmation(payload);

if (userConfirmation) {
  console.log("Received a user confirmation", userConfirmation);
} else {
  // The user's last response was not a confirmation
}

Prompt (Custom Chat completion calls)

`prompt(message, options)`

Send a prompt to the chat UI and receive a response from the user. The message argument must be a string and may include markdown.

The options argument is optional. It can contain a token to authenticate the request to GitHub's API, or a custom request.fetch instance to use for the request.

import { prompt } from "@copilot-extensions/preview-sdk";

const { message } = await prompt("What is the capital of France?", {
  model: "gpt-4",
  token: process.env.TOKEN,
});

console.log(message.content);

In order to pass a history of messages, pass them as options.messages:

const { message } = await prompt("What about Spain?", {
  model: "gpt-4",
  token: process.env.TOKEN,
  messages: [
    { role: "user", content: "What is the capital of France?" },
    { role: "assistant", content: "The capital of France is Paris." },
  ],
});

Alternatively, skip the message argument and pass all messages as options.messages:

const { message } = await prompt({
  model: "gpt-4",
  token: process.env.TOKEN,
  messages: [
    { role: "user", content: "What is the capital of France?" },
    { role: "assistant", content: "The capital of France is Paris." },
    { role: "user", content: "What about Spain?" },
  ],
});

⚠️ Not all of the arguments below are implemented yet. See #5 sub issues for progress.

await prompt({
  model: "gpt-4o",
  token: process.env.TOKEN,
  system: "You are a helpful assistant.",
  messages: [
    { role: "user", content: "What is the capital of France?" },
    { role: "assistant", content: "The capital of France is Paris." },
    {
      role: "user",
      content: [
        [
          { type: "text", text: "What about this country?" },
          {
            type: "image_url",
            image_url: urlToImageOfFlagOfSpain,
          },
        ],
      ],
    },
  ],
  // GitHub recommends using your GitHub username, the name of your application
  // https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api?apiVersion=2022-11-28#user-agent
  userAgent: "gr2m/my-app v1.2.3",
  // set an alternative chat completions endpoint
  endpoint: "https://models.inference.ai.azure.com/chat/completions",
  // compare https://platform.openai.com/docs/guides/function-calling/configuring-function-calling-behavior-using-the-tool_choice-parameter
  toolChoice: "auto",
  tools: [
    {
      type: "function",
      function: {
        name: "get_weather",
        strict: true,
        parameters: {
          type: "object",
          properties: {
            location: { type: "string" },
            unit: { type: "string", enum: ["c", "f"] },
          },
          required: ["location", "unit"],
          additionalProperties: false,
        },
      },
    },
  ],
  // configuration related to the request transport layer
  request: {
    // for mocking, proxying, client certificates, etc.
    fetch: myCustomFetch,
    // hook into request life cycle for complex authentication strategies, retries, throttling, etc
    // compare options.request.hook from https://github.com/octokit/request.js
    hook: myCustomHook,
    // Use an `AbortController` instance to cancel a request
    signal: myAbortController.signal,
  },
});

`prompt.stream(message, options)`

Works the same way as prompt(), but resolves with a stream key instead of a message key.

import { prompt } from "@copilot-extensions/preview-sdk";

const { requestId, stream } = await prompt.stream("What is the capital of France?", {
  token: process.env.TOKEN,
});

for await (const chunk of stream) {
  console.log(new TextDecoder().decode(chunk));
}

`getFunctionCalls()`

Convenience metthod if a result from a prompt() call includes function calls.

import { prompt, getFunctionCalls } from "@copilot-extensions/preview-sdk";

const result = await prompt(options);
const [functionCall] = getFunctionCalls(result);

if (functionCall) {
  console.log("Received a function call", functionCall);
} else {
  console.log("No function call received");
}

Copilot Extensions Documentation

Dreamcode

While implementing the lower-level functionality, we also dream big: what would our dream SDK for Coplitot extensions look like? Please have a look and share your thoughts and ideas: