Skip to content
Guide · AutomationVerified 2 Nov 2026

Copilot ecosystem & automation

Turn Copilot into a team of specialised, repeatable workers. Below: the building blocks, three end-to-end workflows, and a library of 18 ready-to-use templates.

PrimitiveFileUse it forTriggered by
Custom instructions.github/copilot-instructions.md · AGENTS.md · .github/instructions/*.instructions.mdAlways-on project rules: stack, conventions, how to validate.Automatic (applyTo glob or description match)
Prompt files.github/prompts/<name>.prompt.mdA repeatable task you run on demand.Type /<name> in chat
Custom agents.github/agents/<name>.agent.mdA persona with its own instructions, tool allow-list and handoffs.Agent picker, handoff, or as a subagent
Agent skills.github/skills/<name>/SKILL.md (+ scripts)A packaged capability with scripts/resources, loaded only when relevant.Auto-loaded by description, or /<name>
MCP servers.vscode/mcp.jsonConnect external systems: GitHub, databases, browsers, issue trackers.Agent calls the tools
HooksConfigured per harnessDeterministic commands at agent lifecycle events (e.g. format after edit).Lifecycle event
Copilot coding agent.github/workflows/copilot-setup-steps.ymlFully background work: issue → branch → draft PR on GitHub Actions.Assign an issue to Copilot

Agent workflow blueprints

Each workflow combines the templates below. Copy the files, then follow the steps.

Background task: issue → pull request

  1. 1Write a well-scoped issue: goal, acceptance criteria, files to touch, how to test.
  2. 2Assign the issue to Copilot (GitHub.com, GitHub Mobile, VS Code, or the Agents panel).
  3. 3Copilot coding agent starts a session on GitHub Actions using copilot-setup-steps.yml.
  4. 4It pushes commits to a copilot/* branch and keeps a draft PR updated with progress.
  5. 5Review the PR, @mention Copilot in comments to iterate, then approve and merge.

Automated test generation loop

  1. 1Commit tests.instructions.md so every generated test follows your conventions.
  2. 2Select the Test Generator custom agent (or run /write-tests).
  3. 3The agent runs coverage → writes tests → runs them → repeats until the target.
  4. 4Checkpoints let you roll back any iteration you dislike.
  5. 5Open a PR; CI re-runs the suite as an independent check.

Two-layer PR review

  1. 1Before pushing, run the read-only PR Reviewer agent on your branch diff.
  2. 2Fix high-severity findings locally with agent mode.
  3. 3Open the PR and request Copilot as a reviewer for an independent pass.
  4. 4Copilot code review follows .github/copilot-instructions.md, so encode review rules there.
  5. 5A human reviewer approves — AI review supports, not replaces, ownership.

Reusable template library

Agents, prompt files, skills, instructions and MCP config. Every file here is included in the downloadable ZIP.

Drop into .github/agents/. Each agent narrows tools to what its role needs. Tool names vary by harness — use editor completions in the .agent.md file to confirm.

.github/agents/planner.agent.md
---
name: Planner
description: Research a change and produce an implementation plan without editing code
tools: ['read', 'search', 'web']
handoffs:
  - label: Start implementation
    agent: agent
    prompt: Implement the plan above. Run tests after each step.
    send: false
---
You are a senior engineer who plans before anyone writes code.

1. Restate the goal and list explicit assumptions.
2. Find every file, symbol and test affected (use search; cite paths).
3. Produce a numbered plan: step, files, risk, how to verify.
4. Call out open questions. Never modify files.

Use it: Pick “Planner” in the agent dropdown, describe the change.

Terminal
# 1. Download the ZIP from the Download page, then:
unzip copilot-masterclass-templates.zip

# copy everything into your project
node copilot-masterclass-templates/scripts/install-templates.mjs ../my-app

# or only some groups, overwriting existing files
node copilot-masterclass-templates/scripts/install-templates.mjs ../my-app --only=agents,prompts --force

Then commit them

Grab the ZIP from Download templates. Once the files are in .github/, everyone on the team — and Copilot coding agent — gets the same agents, prompts and skills. Run /agents, /prompts or /skills in VS Code chat to confirm they were discovered.