Q36 of 60 — CCA Practice Exam

Code Generation with Claude Code

Your team uses Claude Code's Agent SDK to power a code generation service that handles long-running refactoring sessions. Developers report that after roughly 45–60 minutes of work, Claude begins ignoring the project's coding standards — specifically, it stops enforcing the required error handling pattern that was explained at the start of the session. Logs confirm that automatic compaction events (type: "system", subtype: "compact_boundary") are occurring during these sessions. What is the most effective architectural fix?

A. Increase the agent's token budget so the context window limit is reached less frequently, keeping the original instructions in scope for longer sessions.

B. Move the error handling pattern requirements into CLAUDE.md rather than the initial prompt, so they are re-injected on every request and survive compaction boundaries.

C. Add a stronger system prompt instruction at session start telling Claude to never forget the error handling pattern, even after long periods of inactivity.

D. Trigger manual /compact calls at regular intervals so the compactor runs on a predictable schedule rather than automatically, giving Claude a chance to re-read the initial instructions before each compaction.