시험 출제 기준 및 학습 가이드

The heaviest domain. Design agentic systems: the agentic loop, multi-agent topologies, subagent isolation, sessions, and when to let the model decide vs hard-code control flow.

Design tools the model can select reliably, handle structured errors, and integrate the Model Context Protocol end to end.

Configure and operate Claude Code: the CLAUDE.md hierarchy, slash commands and Skills, permission models, iterative refinement, and headless CI/CD usage.

Write effective prompts and produce reliable structured output with tool use, validation-retry loops, and review patterns.

Keep long-running systems reliable: manage the context window, set escalation rules, propagate errors, calibrate confidence with human review, and annotate provenance.