Context Management and Reliability
Free Claude Certified Architect - Foundations practice — 6 questions on Context Management and Reliability, with explanations. No sign-up. Full 12-question mixed test →
A coding agent is running a long autonomous session refactoring a service. After several hours, its context window is nearly full with raw tool outputs (file contents, command logs) from sub-tasks that have already been completed and verified. Which approach BEST maintains reliability as the session continues?
Active compaction — summarizing finished work and dropping stale raw outputs while preserving task state — is the standard pattern for sustaining reliable long-running agent sessions without losing critical information or exceeding the context window.
In a multi-agent pipeline, an orchestrating agent delegates to Sub-agent A, which hallucinates an incorrect file path, and that path is passed directly to Sub-agent B, which then writes code against the nonexistent file, cascading the error further downstream. Which architectural change would BEST prevent this class of cascading failure?
Inserting explicit validation/verification gates at agent handoff boundaries is the core reliability pattern for stopping error propagation, since it catches invalid outputs before they can corrupt downstream agent state.
An agent must answer 'Where is the authentication logic implemented, and how does it interact with session management?' against a 2-million-line enterprise codebase. Which approach reflects the best-practice architecture for large-codebase exploration?
Agentic, tool-based exploration (grep/search plus targeted file reads) keeps context focused on relevant code, scales to arbitrarily large codebases, controls cost, and grounds the answer in the actual repository rather than a lossy or oversized representation.
A team is configuring escalation thresholds for a claims-processing agent and wants to balance reliability with efficiency. Which configuration approach reflects correct confidence-calibration design?
Tiered, risk-aware thresholds that combine confidence signals with business-risk criteria route only genuinely uncertain or high-stakes cases to humans, preserving both reliability and operational efficiency.
In a Claude-based retrieval-augmented agent answering questions from a document corpus, what does 'source provenance' primarily refer to?
Provenance refers to the ability to trace and attribute generated claims to their originating source material, which is essential for verification and trust in an agent's outputs.
A coding agent is asked to 'fix the bug in the payment module.' While investigating, it identifies three plausible candidate bugs, and the ticket does not specify which one the user means. What is the architecturally correct behavior?
Correct ambiguity handling means surfacing the uncertainty with the specific findings and requesting clarification, which avoids hidden assumptions while still leveraging the agent's partial investigation to speed up human decision-making.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →