GH-600 hands-on checklist
GH-600 rewards having actually used the features. Work through this list in a test repository. Anything you can't explain in a sentence or two is a gap.
GitHub Copilot Cloud Agent
- Explain how Copilot cloud agent researches, plans, changes code, and opens or updates PRs.
- Find where a cloud agent session log is visible.
- Explain why cloud agent PRs need human review before merge.
- Explain branch limits and why the agent should not mutate default branches directly.
- Explain how workflows are handled before running unreviewed agent code.
- Identify which controls are GitHub-built mitigations and which controls the organization must add.
Custom Instructions, Custom Agents, Skills, And Hooks
- Explain when repository custom instructions are better than a custom agent.
- Explain where repository-level custom agents live.
- Explain what an agent profile contains: description, prompt, tools, model, MCP server configuration.
- Explain how tool filtering changes an agent's capabilities.
- Explain how hooks can add validation, logging, or security scanning.
- Explain why custom-agent changes need testing and version awareness.
MCP And Tooling
- Explain what MCP is and why it matters for agentic workflows.
- Distinguish local MCP, remote MCP, GitHub MCP server, and Playwright MCP server.
- Identify who configures MCP servers for Copilot cloud agent.
- Explain why tool allow lists are safer than broad access.
- Explain why a third-party MCP server is a trust boundary.
- Explain what failure could happen if an MCP tool can write external state.
GitHub Actions And Execution Environment
- Explain workflow triggers, jobs, steps, contexts, secrets, variables, and job outputs.
- Explain when a workflow artifact, job log, or check result is the right evaluation evidence.
- Explain the difference between branch protection/rulesets and Actions environments.
- Explain how environment required reviewers protect deployments and secrets.
- Explain how runner environment constraints can cause agent task failure.
- Explain why retries and rollbacks should be explicit.
Repository Governance
- Explain CODEOWNERS and when code owners are requested.
- Explain branch protection and rulesets at a high level.
- Explain required checks and required reviews.
- Explain why draft PRs and human review matter for agent-generated code.
- Explain how to route review for security-sensitive files.
Memory, State, And Drift
- Explain short-term memory vs persistent memory vs external state.
- Explain when to prune, expire, or reset memory.
- Explain how stale context creates wrong agent behavior.
- Explain how to resume work from durable artifacts instead of replaying chat.
- Explain what state must be shared in a multi-agent workflow.
Evaluation And Tuning
- Given a failed agent run, classify the likely root cause before proposing a fix.
- Use logs, plans, traces, workflow artifacts, scans, and session logs as evidence.
- Explain when to tune instructions instead of granting more tools.
- Explain when to add a scan/check instead of requiring extra manual review.
- Explain how CodeQL, secret scanning, and dependency review can be evaluation signals.
Multi-Agent Coordination
- Choose single-agent vs sequential vs concurrent vs group chat vs handoff vs manager-led orchestration.
- Explain why parallel agents need disjoint scopes.
- Explain how to detect duplicated effort and contradictory outputs.
- Explain how to recover from a partial, failed, stalled, or degraded agent run.
- Explain how to retire or replace an agent while preserving auditability.
Guardrails And Accountability
- Classify actions by risk: low, medium, high.
- Identify which actions require human judgment.
- Identify irreversible or compliance-sensitive actions.
- Explain least privilege for repository, branch, workflow, tool, token, secret, environment, and network access.
- Explain how session logs, signed commits, audit events, PR review, and checks preserve accountability.
- Explain why approvals should be minimized when they do not materially reduce risk.