D One-page cheat sheet · CCAR-F

Review this the evening before the exam.

Files and scopes

What Shared (project) Personal (user)

Instructions CLAUDE.md or .claude/CLAUDE.md; subdirectory CLAUDE.md

~/.claude/CLAUDE.md
Topic and path
rules
.claude/rules/*.md with paths: [...] n/a
Slash commands .claude/commands/ ~/.claude/commands/

Skills .claude/skills/<name>/SKILL.md ~/.claude/skills/<name>/SKILL.md (use a different name)

MCP servers .mcp.json with ${ENV_VAR} ~/.claude.json

Skill frontmatter

context: fork: run in an isolated subagent context; only the result returns.

allowed-tools: restrict tools during the skill.

argument-hint: prompt for missing arguments.

Commands and flags

Need Use
Non-interactive CI run claude -p "..."
Parseable output --output-format json plus --json-schema
Continue a named session --resume <name>
Branch from a baseline fork_session
See loaded memory files /memory
Shrink a bloated context /compact
Big or architectural change Plan mode, then direct execution
Verbose discovery Explore subagent

API and SDK

Need Use
Keep looping stop_reason == "tool_use"
Finished stop_reason == "end_turn"
Model may answer in text tool_choice: {"type": "auto"}
Must call some tool tool_choice: {"type": "any"}

Must call this tool tool_choice: {"type": "tool", "name": "..."} Spawn subagents Task tool; "Task" in allowedTools; several calls in one response for parallelism Subagent config AgentDefinition: description, prompt, tools

Block a call Pre-tool hook
Transform a result PostToolUse hook
MCP failure isError: true + errorCategory, isRetryable, message

Cheap, slow work Message Batches: 50% off, up to 24 h, custom_id, no mid-request tools

Decision rules

If the question mentions... Lean towards...

"must always", money, identity, compli‐
ance
Hooks and programmatic gates

Tool confusion Better descriptions first; rename or split if needed "First step" The cheapest fix aimed at the root cause

Too many tools Scope per agent; 4 to 5 tools each
Subagent missing information Put it in the subagent's prompt
Report missing whole topics Coordinator decomposition

Timeout or failure Structured error, local retry, partial results, coverage notes Empty results Distinguish from failure

Noisy reviews Explicit criteria, few-shot pairs, disable noisy category temporarily Invented values Nullable fields

Values do not add up Semantic validation, retry with errors
Data not in the document Stop retrying
Uneven review of big PRs Per-file passes plus integration pass
Self-review Independent instance
Long conversations Case facts block, trimmed tool results

Escalation Explicit request, policy gap, no progress; not sentiment, not self-confidence

Several customer matches Ask for another identifier

High aggregate accuracy Segment, calibrate, stratified sampling

Appendix d: one-page cheat sheet

If the question mentions... Lean towards...

Conflicting sources Keep both, with sources and dates

Sources

All opened on 29 October 2026.

Claude Certified Architect – Foundations Exam Guide, v1.0, July 2026 https://everpath-coursecontent.s3-accelerate.amazonaws.com/instructor%2F6nizmqk8tpzpfjvt6qmmav7rh%2Fpublic%2F1783542750%2FClaude+Certified+Architect+%E2%80%93+Foundations+Exam+Guide.pdf, Anthropic. The authoritative scope document.

Partner certifications page https://anthropic-partners.skilljar.com/page/partner-certifications, Anthropic Partner Academy.

Certification FAQ https://anthropic-partners.skilljar.com/page/faq-certifications, Anthropic Partner Academy.

Anthropic Claude Certification Program https://www.pearsonvue.com/us/en/anthropic.html, Pearson VUE. Four role-based Claude certifications https://claude.com/blog/four-role-based-claude-certifications, Claude blog, 23 July 2026.

Claude certifications guide https://blog.udemy.com/claude-certifications-guide/, Udemy blog (third party; used for the program overview and CCAR-P price).

Claude Certification Guide https://claudecertificationguide.com/, independent site not affiliated with Anthropic (cross-check only).