Member-only story
Anthropic’s Claude Certified Architect Exam (CCA-F): The Coordinator That Shared Too Much, and the Loop That Never Stopped
CCA-F Part 2: Domain 1 is the largest slice of the CCA-F, and the one passers score highest on. Here is why agentic architecture is your floor, not your ceiling, and the six fixes that bank the points
Your agent does not finish when it says it is finished, it finishes when the model says so, and a coordinator that hands a subagent everything it knows is one leaked field away from a breach. Drive the loop off stop_reason and give each subagent only the context it needs, and the production system you ship stops guessing and stops oversharing.
In this article: You will learn the six repair patterns that Domain 1 of the CCA-F exam actually tests: stopping an agent loop on
stop_reasoninstead of guessing from text, scoping a coordinator so its subagents inherit nothing, moving state between agents on purpose, handling errors at the lowest capable level, enforcing hard guarantees in a hook rather than a prompt, and remembering that the Claude API is stateless by design. The throughline is one sentence: the right fix removes the architectural root cause, never bolts more infrastructure onto…
