Q33 of 60 — CCA Practice Exam

Code Generation with Claude Code

Your team has built a Claude Code environment for code generation with a suite of eight Skills — covering tasks like PDF extraction, test generation, linting, and dependency auditing. Developers report that on complex multi-step requests, Claude increasingly loses track of earlier instructions mid-task and produces truncated outputs. Examining the SKILL.md files, you find several contain lengthy preambles explaining fundamental concepts (e.g., the PDF Skill opens with three paragraphs defining what PDFs are and how text extraction libraries generally work). What is the most effective architectural fix for this degraded performance?

A. Increase the context window allocation for the session so all Skill metadata, SKILL.md content, and conversation history can coexist without competing for space.

B. Trim each SKILL.md to remove explanations of concepts Claude already knows — retaining only configuration values, environment-specific paths, and non-obvious constraints — so loaded Skill content consumes fewer tokens.

C. Split the eight Skills into two separate Claude Code instances — one for document-handling Skills and one for code-quality Skills — so each instance loads a smaller set of Skill metadata at startup.

D. Move the conceptual explanations from SKILL.md into the system prompt so they are always globally available rather than loaded conditionally per-Skill, reducing per-Skill token overhead.