12 Task-statement review · CCAR-F

The exam guide lists 30 task statements, and every item is written against one of them. This chapter gives each a final review in the same format: the one idea to remember, what a correct answer usually looks like, and the trap to avoid. Use it in the last week to find gaps; the chapter in brackets is where to go back.

Domain 1: Agentic Architecture & Orchestration

1.2 Coordinatorsubagent (Ch.

and context (Ch.

2.3)

Subagents inherit nothing Pass full findings with structured metadata; parallel Task calls in one response; "Task"

1.5 Hooks (Ch.

2.5)

Pre-call hooks block; postcall hooks transformBlock over-limit refunds and redirect; normalise heterogeneous data

1.7 Sessions (Ch.

Domain 2: Tool Design & MCP Integration

tion and tool_choice (Ch.

Domain 3: Claude Code Configuration & Workflows

Task Remember Correct answers usually... Trap
3.1
CLAUDE.md

hierarchy (Ch.

Team conventions in ~/.claude/CLAUDE.md

context: fork for verbose skills; allowedtools to restrict; argument-hint; personal variants under new names

Invented config files; editing
team skills for personal use
3.3 Path-spe‐

cific rules (Ch.

4.3)
paths globs load rules
only for matching files

Glob rules for scattered files such as tests Subdirectory CLAUDE.md for cross-cutting conven‐

Examples beat prose 2 or 3 input/output examples; tests first; interview pattern; batch interacting fixes

More adjectives in the de‐

scription

3.6 CI/CD (Ch.

Domain 4: Prompt Engineering & Structured Output

Task Remember Correct answers usually... Trap
4.1 Explicit criter‐
ia (Ch. 5.1)

Categories beat caution Report/skip lists; severity examples; temporarily disable noisy categories

Task-statement review

Task Remember Correct answers usually... Trap
4.4 Validation and
retry (Ch. 5.4)
Retry with specific er‐
rors; know when not to

Document plus failed output plus errors; calculated_total; conflict_detected; detected_pattern

Domain 5: Context Management & Reliability

propagation (Ch.

How to use this chapter

Read each row and try to produce, without looking, one realistic symptom and its fix. If you cannot, that task statement is a gap. Most candidates find their gaps cluster in one domain; spend your last study session there, using that chapter's worked examples and the drills tagged with that domain.