Agentic Workflow Orchestration and Deployment: free practice questions

Option B is correct because Codespaces are designed as interactive cloud development environments and implement automatic suspension after an idle timeout (default 30 minutes, configurable up to a maximum). An always-on agent backend requires continuous uptime; a Codespace that suspends when not interactively used would silently stop processing events, making it unreliable for production agentic workloads. Option A is wrong because Codespaces do have full outbound internet access and can call GitHub APIs and external services — network isolation is not a Codespaces limitation. Option C is wrong because Codespaces fully support custom dev container configurations via `devcontainer.json`, including custom Docker images, Dockerfiles, and feature sets. Option D is wrong because Codespaces and GitHub-hosted Actions runners use entirely different billing models: Codespaces bill by core-hours and storage, while Actions bill by minute of job execution; the rates and cost structures are not equivalent.

Option C is correct because the GitHub Copilot coding agent applies its own default branch naming logic unless the repository is configured with a custom branch naming convention or the workflow/issue template explicitly instructs the agent on naming. Without explicit configuration, Copilot uses its built-in heuristics, which may not match the team's desired `agent/<issue-number>-<slug>` convention. Option A is wrong because branch protection rules would prevent branch creation entirely and result in an error, not a branch with a different name. Option B is wrong because assignment vs. mention may affect whether the agent triggers, but once triggered, the naming behavior is governed by configuration, not the trigger method. Option D is wrong because a `contents: write` permission failure would prevent any branch from being created and would surface as a permission error, not as a branch with a different name; the symptom describes incorrect naming, not creation failure.

Option B is correct because dedicated retry actions (such as `nick-fields/retry` or similar community actions) wrap a failing step and re-execute it with configurable delays and exponential backoff — exactly the pattern needed for transient rate-limit errors — without requiring changes to the underlying action's source code. Option A is wrong because `continue-on-error: true` suppresses the error and moves on, meaning subsequent steps that depend on the LLM output will receive no valid data, leading to downstream failures or incorrect behavior; it does not retry. Option C is wrong because running a duplicate job on failure re-runs the entire job, not just the failing step, which wastes resources and resets all prior successful steps unnecessarily; it is also not configurable with backoff. Option D is wrong because `timeout-minutes` controls how long a step is allowed to run before being killed; increasing it does not cause a retry on failure and has no effect on an HTTP 429 error that returns immediately.

Option B is correct because GitHub Actions Environments support required reviewer approvals: when a job targets a protected environment, the workflow pauses and waits for a designated human reviewer to approve before the job executes. This is the canonical, built-in human-in-the-loop gate for deployment workflows. Option A is wrong because `timeout-minutes: 0` is not valid syntax for pausing a workflow; the timeout setting cancels a job after a duration, not suspends it for review. Option C is wrong because skipping the merge step entirely when the actor is an AI agent means the workflow never merges, which defeats the purpose — the goal is to merge after human approval, not to block merging unconditionally. Option D is wrong because inserting a `sleep` command is a fragile anti-pattern: it does not provide a structured approval interface, burns billable runner minutes during the wait, and does not prevent the workflow from eventually proceeding without actual human confirmation.

Option A is correct because `strategy: matrix` allows the worker jobs to run in parallel across multiple configurations simultaneously, and listing all worker jobs in the aggregator's `needs` array creates the fan-in gate — execution only continues after all parallel workers succeed. This is the canonical GitHub Actions fan-out/fan-in pattern. Option B is wrong because sequential steps negate parallelism; workers run one after another, dramatically increasing total runtime and defeating the purpose of parallel agent decomposition. Option C is wrong because `workflow_call` used in sequence within a single job is still serial; reusable workflows do not automatically execute in parallel when called from the same job step. Option D is wrong because while `repository_dispatch` can trigger separate workflows, orchestrating aggregation across asynchronously triggered workflows requires complex external state tracking and is not natively supported as a built-in fan-in mechanism.