Before deploying a new agent to production, a platform engineer wants to evaluate the execution context to ensure the agent will have the correct permissions, environment variables, and secrets available when it runs. Which approach best accomplishes this evaluation?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Evaluating an agent's execution context is like doing a pre-flight checklist before takeoff — you need to actually run through the systems, not just read the manual. A dry-run or sandbox invocation generates runtime logs showing the real resolved state: which secrets were actually injected (not just expected), which permissions were granted, and which MCP tools successfully initialized. Reading docs or counting YAML keys tells you what's intended, not what actually happened.
Full explanation below image
Full Explanation
The execution context for an agent includes the runtime environment it operates within: injected secrets and environment variables, granted permissions (repository access, API scopes), initialized tool connections (MCP servers, GitHub API), and the identity/token under which it runs. Evaluating this context before production deployment is a critical quality step.
Option A is incorrect. GitHub Marketplace model cards may document what an agent is designed to do, but they do not reflect the runtime state of your specific deployment environment. Documentation tells you what should exist, not what does exist in your configuration.
Option B is correct. A dry-run or sandbox execution is the only way to observe the actual resolved execution context. Runtime logs produced during this test reveal: whether secrets were successfully injected from the secrets store, what permission scopes the agent token actually has, whether each MCP server connected and completed its handshake, and what tool definitions were discovered. Discrepancies between expected and actual context surface as errors, missing tools, or permission-denied messages in these logs — allowing issues to be caught before they affect production.
Option C is incorrect. A README lists what the agent's author says is required. It does not validate that your environment satisfies those requirements. The README might be outdated, incomplete, or describe a different deployment scenario. Only runtime verification confirms the actual state.
Option D is incorrect. Counting env: blocks in a workflow file tells you how many environment variables are declared, but not whether those variables resolve to valid values, whether referenced secrets exist in the secrets store, or whether the values are correct for the target environment. A variable can be declared in YAML and still be empty or undefined at runtime.
Key concept: execution context evaluation requires runtime observation, not static analysis of configuration files.