GH-600 practice test

Fitting agents into the development lifecycle, identifying the steps an agent should take, recognising anti-patterns, defining inputs, outputs and success criteria, keeping planning separate from execution, structured plans and validating them before anything runs, observability and control for autonomous agents, choosing a degree of autonomy, producing inspectable artefacts inside ordinary development tooling, and placing human intervention without slowing delivery.

Identifying and configuring the tools an agent needs and the permissions on them, MCP servers including the GitHub remote server, registries and allow lists, evaluating an agent's execution context, scoping an agent to a repository or a branch, invoking one from a CI workflow, letting an agent create branches and pull requests, handling environment-specific constraints, and the error handling, retries, rollbacks, escalation and traceability that make autonomous execution safe.

Choosing between short-term, long-term and external memory, scoping memory to what the task needs, expiry, pruning and reset rules, capturing progress and decisions as durable artefacts, resuming work without repeating steps or diverging from earlier decisions, detecting and correcting drift during long runs, and sharing state across tools and environments without conflicting or stale context.

Specifying expected outcomes and operational constraints, choosing qualitative and quantitative signals, aligning evaluation with development intent, generating signals with automated scanning, finding failures in logs, plans, traces, outputs and workflow artefacts, classifying root causes across reasoning errors, tool misuse and environment problems, and tuning instructions, workflows, constraints, memory and tool access in response.

Orchestration patterns, isolating agents for parallel execution, detecting and resolving overlapping changes, duplicated effort and contradictory outputs, observability across several agents through logs, artefacts and operational signals, artefacts suitable for review and audit, documenting decisions and handoffs, post-hoc analysis, identifying failed, partial and stalled executions, recovery patterns including rollback and human intervention, and adding, replacing and retiring agents without disrupting active work.

Classifying agent actions by operational, security and compliance risk, assigning autonomy levels that keep delivery fast and stay within security and responsible AI standards, identifying the actions that genuinely need human judgement, blocking what violates policy, scoping permissions and execution contexts to least privilege, requiring explicit authorisation for irreversible or compliance-sensitive changes, and removing approvals that do not reduce risk.