GH-600 mock exam

Question 1, Prepare agent architecture and SDLC processesAn agent starts editing files before anyone has seen its approach, and rework is high. What do you change?

• AIncrease its memory so it can hold more of the codebase
• BRun it only at night so rework happens outside work hours
• CGive it more tools so it can check its own changes
• DHave it output a structured plan that is validated before it can act

Question 2, Implement tool use and environment interactionA team may only use approved MCP servers with its agents. What do you configure?

• AA longer system prompt
• BA branch naming rule
• CA larger model
• DAn MCP allow list

Question 3, Implement tool use and environment interactionAn agent should work on one repository and open pull requests, but never push to main. Which approach fits?

• AGive it an owner token and rely on the prompt not to push to main
• BLet it push anywhere and review its work in the audit log afterward
• CScope it to the repository and a working branch, with branch protection on main
• DRun it locally only, on a developer's own machine and credentials

Question 4, Manage memory, state, and executionAn agent working over several days starts repeating steps and contradicting earlier decisions. What helps most?

• AStart from nothing each time and re-read the whole repository first
• BCapture progress and decisions as durable artifacts it can resume from
• CRaise the model temperature so it considers more options
• DRemove its tools so it stops repeating steps

Question 5, Perform evaluation, error analysis, and tuningAn agent produced a wrong change. Logs show it called the right tools, but it misread the requirement. How do you classify this?

• AA network failure
• BA reasoning error
• CA tool misuse failure
• DAn environment failure

Question 6, Orchestrate multi-agent coordinationTwo agents run in parallel and keep changing the same files. What should you configure first?

• ALonger timeouts so each agent can finish
• BIsolation so each works in its own branch or workspace
• CA shared chat channel where they post their plans
• DFaster models so the overlap window is shorter

Question 7, Orchestrate multi-agent coordinationOne agent in a workflow stalls midway through its task. What is the best recovery pattern?

• ADetect the stall, roll back or retry the step, and escalate to a person if needed
• BLet the other agents continue on the partial output, and fix it in review
• CDelete all branches from the run and start the whole workflow again
• DLeave it running until a user or a reviewer complains about it

Question 8, Implement guardrails and accountabilityDevelopers complain that agents wait for approval on every small edit. What is the right change?

• AClassify actions by risk and require approval only where it reduces risk
• BRemove all approvals so agents can move at full speed
• CAdd a second approver so reviews are spread across the team
• DStop using agents for small edits and keep them for large ones