- For
- Anyone responsible for what an AI coding tool is allowed to reach, and for answering when someone asks whether it is safe.
- Lessons
- 13 existing Academy lessons
- Reading time
- about 3½ hours
- Prerequisites
- Familiarity with your organisation's existing code review and secrets handling.
Stage 1What is actually being sent
Every later control is easier to reason about once you know how context is assembled per request.
- What Is GitHub Copilot and How Does It Work?— completed
A precise definition of the product, how context shapes its output, what it can and cannot do, and where developer responsibility begins.
- GitHub Copilot Context: How to Give Copilot Better Information— completed
Relevant context beats maximum context. What Copilot can see, what it cannot, and what should never go into a prompt.
- What Is GitHub Copilot and How Does It Work?— completed
Stage 2The controls and their limits
Read the limitations first. Content exclusion not applying to Edit and Agent modes is the gap organisations most often miss.
- GitHub Copilot Security & Enterprise Guide— completed
The final pillar: how individuals, teams and enterprises use Copilot without giving up control of source code, secrets, policy boundaries or human accountability.
- GitHub Copilot Security Best Practices— completed
A working developer's security practice for AI-assisted code: what to check, which categories deserve a second pair of eyes, and where generated code fails quietly.
- Is GitHub Copilot Safe for Private Repositories?— completed
Neither yes nor no. Where requests actually go, what the plan commits to contractually, what content exclusion covers, and the surfaces it explicitly does not.
- GitHub Copilot and Secrets: What Developers Need to Know— completed
Why a credential in a prompt is different from a credential in a file, what secret scanning does and does not see, and how agents and MCP widen the surface.
- GitHub Copilot Security & Enterprise Guide— completed
Stage 3External tools
MCP is where an agent gains reach beyond the repository, which makes it the largest new surface to govern.
- GitHub Copilot MCP: Complete Guide— completed
Model Context Protocol as Copilot implements it: clients, servers, tools versus resources, transports, authentication, and the security model that matters most.
- GitHub Copilot MCP Security: Governing External Tools— completed
The MCP threat model at enterprise scale: allowlists in managed settings, why the private registry is not the stronger control, and the surface it does not reach.
- GitHub MCP Server Tutorial— completed
GitHub's own MCP server: setup, OAuth versus tokens, toolsets, read-only patterns, and the organization policy that gates all of it.
- GitHub Copilot MCP: Complete Guide— completed
Stage 4Review as a control
More generated code means more review. Knowing what automated review catches decides where human attention should go.
- GitHub Copilot Code Review: Complete Guide— completed
Requesting a review, the effort levels and what they cost, every documented customization mechanism, and the four things GitHub says the review does not do.
- How to Automatically Review Pull Requests with GitHub Copilot— completed
Automatic review is a branch ruleset rule, not a settings toggle — targeting, draft and new-push options, effort defaults, and how to pilot it without drowning a team in comments.
- GitHub Copilot Code Review: Complete Guide— completed
Stage 5Policy and enforcement
Content exclusion and policy management are organisation controls, so the plan boundary is part of the security design.
- GitHub Copilot Business: Complete Guide— completed
Operating Business day to day — seats, the policies that matter first, content exclusion, public-code blocking, code review and what the metrics will and will not show you.
- GitHub Copilot Governance and Policy Guide— completed
Governance as a set of deliberate decisions rather than a locked door — eleven domains, a precedence model that surprises people, and a quarterly review that takes an hour.
- GitHub Copilot Business: Complete Guide— completed
Put it into practice
Projects
- CLI permission matrix
Twenty-four permission configurations paired with the commands each should allow, refuse or confirm — every destructive probe in both direct and instrumental framing.
Check yourself
Reading time is computed from the prose of the lessons listed above at 200 words per minute, excluding code blocks. It is an estimate of reading, not of the practice that makes any of it stick.