Session M6-01: Build reliable agent applications with the GitHub Copilot SDK¶

Module: Module 6: Advanced Topics & Capstone

Difficulty: Advanced

Prerequisites: Sessions M1-01 to M1-03, M2-01 to M2-04, M3-01 to M3-05

Duration: 3 hours (1 hour trainer content, 2 hour lab)

Overview¶

Build a TypeScript CLI around an agent to assess whether a synthetic software change is ready to proceed.

You write the reliability controls in application code. The CLI manages the client and session lifecycle, streams progress, and limits tool use. It also shows denied actions, records failures, and closes resources. Tests check that control path. An OpenTelemetry Collector sends traces to Jaeger so you can inspect the same run outside the terminal.

Learning outcomes¶

manage CopilotClient and session startup, use, and shutdown;
subscribe to streaming session events without printing duplicate final output;
define a typed custom tool over deterministic synthetic data;
separate tool availability, permission handling, and pre-tool policy;
use pre-tool, failed-tool, error, and session-end hooks;
make a denied action visible to the user and trace;
test the allowed, denied, failed, and cleanup paths;
send SDK traces through an OpenTelemetry Collector to Jaeger.

Required access¶

Learners need:

authenticated GitHub Copilot SDK access;
Node.js and npm;
Docker with Docker Compose;
a terminal that can run the supplied TypeScript project.

If GitHub Copilot access, Node.js, npm, Docker, or Docker Compose is unavailable, stop before the lab. Fix the environment or reschedule the hands-on work.

Scenario¶

The application is a change readiness assistant. It reads deterministic synthetic change records, asks Copilot to assess the selected change, and returns ready or needs-review with the supporting gaps.

One request targets a restricted synthetic record. The application denies that lookup before execution and shows the reason. No customer data, production repository, or external business system is used.

Session structure¶

Block	Duration	Content
Trainer content	60 min	Slides, concepts, and prepared demo
Lab	120 min	Build and check the CLI

Materials¶

Resource	Location
Trainer guide	`trainer-content/README.md`
Slides	`slides.md`
Lab guide	`lab/README.md`
Starter project	`lab/starter/sdk-readiness-assistant/`
Reference solution	`lab/solution/sdk-readiness-assistant/`

Scope¶

The lab covers:

local CopilotClient and session lifecycle;
TypeScript streaming event handlers;
one application-owned readiness tool;
recorded permission decisions;
pre-tool, post-tool-failure, error, and session-end hooks;
a visible denied action;
focused automated tests;
OTLP export through an OpenTelemetry Collector;
trace inspection in Jaeger.

The session does not cover dynamic workflows, fleet, cloud or remote sessions, BYOK, plugin packaging, production deployment, external business systems, or implementations in other languages.

Product references¶

These official GitHub sources were checked on October 5, 2026. The GitHub Copilot SDK became generally available on June 2, 2026. Trainers should still recheck API names and package requirements before delivery.

Trainer note¶

Keep the lesson on application reliability. Do not expand the demo into orchestration, remote execution, provider selection, or deployment.