We want this session to end with code another engineer can read, test, and use.

In the past year, Copilot CLI became generally available, skills and MCP spread across Copilot surfaces, and Squad reached 1.0. Martin and Haflidi show how native GitHub Copilot CLI features and Squad work together on an existing AKS Terraform codebase. The outcome is a reusable infrastructure module, documented and tested around a clear contract, with a consumption example for an existing Azure landing zone. The primary path uses private Corp networking and pre-provisioned platform services, not a new public standalone environment.

We start with the files and decisions that shape the work. Native Plan mode makes the proposed change visible before implementation. Reviewed files, instructions, skills, and MCP lookups focus the agent. Squad adds roles, routing, ownership, handoffs, and repository-backed decisions. Three narrow agent profiles in the repository are a coder, a validator, and a separate reviewer. The validator has no edit tool, but its shell can still write files; instructions and native approval prompts enforce its command list. They narrow lanes without pretending they are sandboxes. Parallel work stays bounded; humans review changes.

Code qualification comes first. The live session shows genuine new Copilot CLI execution with Squad selected from a disclosed clean checkpoint, not the first-ever implementation. Both speakers narrate the decisions live. We'll inspect the plan, follow a handoff, read a meaningful failed check, and discuss the repair. We'll also use diff and review tools, resume with decisions intact, and explain model, context, and usage controls. Worktrees, recovery, bounded automation, and cloud delegation stay in a concise reference section.

You'll leave able to choose an appropriate CLI or Squad workflow, define a module boundary, and connect an agent's claim to evidence. You'll also know why passing local tests, a reviewed Terraform plan, and observed Azure behavior are different milestones.

Public code and examples stay separate from private inputs, identities, backend/state, and secrets. The October 5 Azure validation is cited only as sanitized, revision-bound evidence, not reusable environment data.