What's changed: New GH-300 Chapter 2 (IDE & CLI = enable, inline/Chat/Plan Mode, Copilot CLI, content exclusion; advanced = Edit Mode/Agent Mode/MCP, agent sessions/sub-agent delegation, code review, instructions/prompt files, PR summaries/Spaces/Spark, Chat commands/limits; org settings = policy mgmt/Code Review policy/feature availability (IDE+github.com)/audit logs/REST API subscriptions)
2.2Agent Mode, Edit Mode, MCP, and Advanced Features
Understand advanced development via Agent Mode, Edit Mode, and MCP (Model Context Protocol); managing agent sessions and delegating to sub-agents; code-review assistance; and consistent help via Spaces, Spark, Pull Request summaries, and instructions/prompt files.
Copilot goes beyond completion with more autonomous, workflow-oriented features. GH-300 tests their differences and when to use each. Central are Edit Mode (instruct multi-file edits via chat), Agent Mode (give a goal and it autonomously plans, edits, and verifies iteratively), and MCP (Model Context Protocol) for connecting to external tools/data.
2.2.1Edit Mode, Agent Mode, and MCP
Edit Mode proposes and applies multi-file changes from natural-language instructions (you indicate the scope). Agent Mode is more autonomous: given a goal, it iterates plan → edit → verify via build/test → fix on its own, deciding which files and commands it needs. MCP (Model Context Protocol) connects Copilot to external tools/data sources (issue trackers, databases, docs) in a standardized way, broadening context for more accurate help. Summarize: "scoped editing = Edit, autonomous execution = Agent, external connection = MCP."
2.2.2Agent sessions and sub-agents
Agent Mode work is managed as agent sessions—you can check progress, pause, and resume. For large tasks or growing context, delegate subtasks to sub-agents to optimize each agent’s context usage (the parent doesn’t hold everything; it carves out subproblems). This keeps long, complex work from straining the context limit and makes it more tractable.
2.2.3Code review and consistency features
Copilot also assists code review, returning comments and improvement suggestions on changes. Organizations can supply review standards and coding conventions via instructions files for consistent review criteria. Pull Request summaries auto-summarize changes to speed up review. Spaces bundle related code/docs/instructions as shared context, and Spark quickly scaffolds an app from an idea. In Chat, reuse a prompt file to get consistent responses from standard instructions. Chat has commands (e.g., slash commands) and limits, and you can send feedback to help improve it.
| Feature | Essence |
|---|---|
| Edit Mode | Multi-file edits from instructions (you scope) |
| Agent Mode | Autonomous plan/edit/verify from a goal |
| MCP | Standard connection to external tools/data |
| Instructions files | Consistency via conventions/review standards |
| PR summaries / Spaces / Spark | Summarize changes / bundle context / scaffold |
Common: (1) Edit Mode = multi-file edits from instructions (you scope); Agent Mode = autonomous plan/edit/verify from a goal; MCP = standard connection to external tools/data. (2) Manage large work as agent sessions and delegate to sub-agents to optimize context. (3) Consistent review/responses = instructions files / prompt files. (4) Know the uses of PR summaries, Spaces, Spark.
Watch out: (1) Don’t confuse Edit Mode (you scope the edit) with Agent Mode (it decides and iterates autonomously). (2) Even in Agent Mode, validation is required (Chapter 1)—autonomy is not "no review." (3) MCP is a standard for external connections—mind permissions/secrets of what it connects to. (4) Instructions files (always-applied conventions) and prompt files (reusable instructions) are related but distinct.
2.2.4Section summary
- Edit Mode = multi-file edits from instructions; Agent Mode = autonomous plan/edit/verify; MCP = standard external connection
- Manage as agent sessions; delegate to sub-agents to optimize context
- Use instructions/prompt files for consistent review/responses; leverage PR summaries, Spaces, Spark
- Even autonomous features need output validation; mind permissions/secrets of MCP targets
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. Which feature has Copilot, given a goal, autonomously plan, edit multiple files, and iterate while verifying via build/test?
Q2. You want to connect Copilot to external tools/data (issue tracker, DB) in a standardized way to broaden context. Which mechanism?
Q3. For a large task with growing context, you want to optimize context usage. What is an appropriate Agent Mode practice?
Q4. You want Copilot’s code review to apply consistent coding conventions and review standards org-wide. What do you use?
Q5. Which feature auto-summarizes changes to speed up the start of PR review?
Q6. Which correctly states the essential difference between Edit Mode and Agent Mode?