GitHub GH-600: Orchestrate multi-agent coordination Practice Questions

Correct (F): The security-review agent's own execution log is the primary source for diagnosing why it produced no output — it may show a silent crash, an unhandled exception, or an infinite wait. — Correct (D): Examining the handoff payload verifies whether the agent received malformed, empty, or unexpected input that caused it to short-circuit without error — a common cause of silent failures. — Correct (B): Runner-level diagnostic logs can reveal infrastructure-level causes such as out-of-memory kills, runner timeout signals, or network interruptions that would not appear in the agent's own logs. — Wrong (E): The style-check agent's output is upstream and does not control whether the security-review agent is invoked; the pipeline published incomplete output regardless, making this a lower-priority artifact for root-cause diagnosis. — Wrong (A): Branch protection rules govern merge requirements, not agent invocation or output routing; they would not suppress an agent's output during execution. — Wrong (C): Monthly billing minutes are an aggregate metric and cannot be correlated to a single agent's termination within a specific run; per-run job duration logs from option B are far more diagnostic.

Correct: Defining each agent as a separate GitHub Actions job allows them to run in parallel on independent runners with isolated workspaces and file systems, preventing read/write conflicts. Job-level concurrency groups scoped to each agent's function also prevent duplicate runs of the same agent type. — Wrong (A): Sequential execution in a single job eliminates parallelism and creates a bottleneck; environment variables between steps do not provide isolation. — Wrong (C): Background processes sharing a common workspace directory can cause race conditions and file-level conflicts, which is the opposite of isolation. — Wrong (B): While dedicated environments provide isolation, using needs only for dependency ordering still requires the parallel structure to be defined at the job level; this option conflates environment isolation with scheduling and doesn't describe a complete parallelism configuration.

Correct (D): Structured per-agent log entries with agent ID, input context, rationale, and output references create the precise, attributable audit trail required for compliance and post-hoc analysis. — Correct (E): Storing handoff payloads in an append-only artifact store with appropriate retention ensures that inter-agent communication is preserved for the required compliance period and cannot be retroactively altered. — Wrong (B): Global debug logging captures runner-level system noise rather than semantically meaningful agent decisions; it produces excessive volume with low signal and does not attribute decisions to specific agents. — Wrong (C): An in-memory cache with end-of-workflow flush risks data loss on partial or stalled executions, which is exactly when audit trails are most needed. — Wrong (A): PR cycle times measure workflow throughput, not agent-level decision attribution; they do not satisfy an audit trail requirement.

Correct: A fan-out/fan-in pattern uses a coordinator that dispatches work to agents in parallel but serializes the reconciliation of their outputs before any artifact is committed. This prevents contradictory outputs from coexisting by having the coordinator detect naming conflicts before they propagate. — Wrong (B): Writing to separate branches and relying on Git merge conflict resolution only detects the problem after the fact and requires manual resolution; it does not prevent contradictory outputs from being generated. — Wrong (C): Using timeouts to impose an implicit ordering is fragile and non-deterministic; it does not guarantee sequencing and introduces unnecessary latency without addressing the root cause. — Wrong (D): Post-run validation can detect the conflict but does not prevent it; the orchestration pattern should prevent contradictory states from being created in the first place, not just catch them afterward.

Correct: Tagging the final production version creates an immutable reference point; archiving the repository keeps it accessible without allowing new changes; retaining workflow run logs and artifacts ensures historical runs can be audited; and updating documentation provides context linking historical runs to the now-archived agent. This satisfies both auditability and continuity requirements. — Wrong (B): While Git history preserves code, simply deleting configuration without explicit archival and documentation makes it difficult for auditors unfamiliar with the repository to locate the correct version context. — Wrong (A): Retroactively modifying historical workflow YAML files corrupts the audit trail by misrepresenting what actually ran during historical executions — this is a serious compliance violation. — Wrong (C): Exporting only the last runtime state captures a single point-in-time snapshot and does not preserve the agent's full decision history or the ability to reproduce earlier runs.