Claude Certified Architect – Foundations (CCA-F)

Question 1 of 30

Question Navigator

Agentic Architecture & Orchestration

Context: An architect is designing a long-running autonomous agent that relies on a continuous ReAct loop to interact with multiple external APIs. Problem: During complex multi-step operations, the agent occasionally stops generating output midway through a large JSON tool input payload, causing the orchestration parsing logic to crash. The API response indicates a specific stop reason rather than a network error. Question: How should the orchestration layer be designed to robustly handle this specific truncation failure mode?

Detect stop_reason: "max_tokens", append the truncated assistant message to the context, and prompt the user to manually increase the token limit.

Detect stop_reason: "max_tokens", append the partial assistant message to the context, and send a new request with the exact same messages array to allow Claude to continue generation.

Detect stop_reason: "length", and automatically inject a system message instructing the model to summarize its previous thoughts before continuing.

Detect stop_reason: "tool_use", extract the partial JSON, and use a secondary lightweight model to auto-complete the missing syntax.