CCA-F Study Map — YouTube Course ↔ Exam ↔ Repo
Course code (pick one path):
- Tracked in this repo: claude-certified-architect/ (repo root)
- Optional clone (often gitignored): Youtube-main/claude-certified-architect/
Exam hub (HTML): index.html · cca-f-prep.html
Use this map to study in exam-domain order while following the Peace of Code playlist in episode order.
Quick start
| Step | Action |
|---|---|
| 1 | Skim docs/02-domain-traps (five domains) |
| 2 | Watch Ep 01–08 with code open; run api_check.py → agent.py → capstone_project.py |
| 3 | Build p4-claude-code-workflow + p5-custom-mcp-server |
| 4 | Drill docs/03-exam-scenarios (pick 4 of 6 scenarios) |
| 5 | Practice exam ≥ 900/1000 → docs/04-practice-exam-tracker |
API setup (course): pip install anthropic · ANTHROPIC_API_KEY in env · model claude-haiku-4-5 for hands-on.
Episode ↔ domain ↔ files
| Ep | Title (playlist) | Primary domains | Code / config | Slides | Transcript |
|---|---|---|---|---|---|
| 01 | Agentic loops & stop_reason |
D1, D3 | agent.py, api_check.py |
slide decks/lesson 1.pptx |
✅ |
| 02 | Multi-agent & coordinator patterns | D1 | multi_agents.py |
docs/Claude_CertArch_Ep02_*.pptx |
✅ |
| 03 | Subagent context & sessions | D1, D5 | subagent_context.py, fork_session.py |
lesson 2.pdf, docs/Ep03_*.pptx |
✅ |
| 04 | Multi-agent Python + SDK | D1 | capstone_project.py (start) |
lesson 3.pdf, docs/Mayank_Chugh_Ep04_*.pptx |
✅ |
| 05 | PreToolUse / PostToolUse hooks | D2, D1 | capstone_project.py, dynamic_decomposition.py, prompt_chaining |
lesson 5.pptx |
✅ |
| 06 | Tool descriptions & misrouting | D3, D4 | tool_misrouting.py |
lesson 6.pptx |
✅ |
| 07 | Error handling & tool_choice |
D4, D5 | tool_choice.py, error schema in agent.py |
lesson 7.pptx |
✅ |
| 08 | MCP servers, config, Klein | D4, D2 | .mcp.json, MCP theory in video |
lesson_8.pptx |
✅ |
| 09 | Claude built-in tools (Grep, Glob, Read, …) | D2, D1 | inbuilt_tools.py, developer_productivity.py |
lesson_9.pptx |
⏳ placeholder |
| 10 | CLAUDE hierarchy & path rules |
D2, D3 | Align with p4-claude-code-workflow | lesson_10.pptx |
⏳ placeholder |
Transcripts: Youtube-main/claude-certified-architect/docs/Transcripts of youtube/
Later episodes (YouTube only — add locally as you go): Ep 11+ (e.g. Ep 15 few-shot → few_shot_examples.py). Playlist: Claude Certified Architect — Full Course.
Exam concepts by episode (high-signal)
Ep 01 — Agentic loop
stop_reason:tool_use→ execute tools, append results, continue;end_turn→ only valid primary exit.- Append assistant message (tool_use blocks) then user message (tool_result).
- Anti-patterns: parse natural language to exit; use
MAX_ITERATIONSas primary stop; inspect content types instead ofstop_reason. - Roles:
user,assistant,system; API is stateless — send fullmessageshistory.
Ep 02 — Coordinator
- Hub orchestrator + workers;
Tasktool;allowedToolsmust include"Task". - Decompose goals, not step-by-step scripts for workers.
Ep 03 — Context & sessions
- Pass structured findings (objects), not prose dumps.
client.beta.sessions.forkfor parallel exploration from one baseline.- On
--resume, tell the agent what changed on disk.
Ep 04–05 — Capstone & hooks
capstone_project.pyties Ep 01–05: coordinator, mock DB, PreToolUse block (e.g. high-value orders), PostToolUse normalize raw tool output.- Four error categories:
transient,permission,validation,internal.
Ep 06 — Tool routing
- Vague descriptions → wrong tool (misrouting); fix with specific descriptions and disambiguation (“USE INSTEAD of Grep when…”).
Ep 07 — tool_choice
- Force or restrict tool use when reliability matters (see
tool_choice.pysnippet).
Ep 08 — MCP
- Lifecycle: connect → initialize (session) → discover (
tools/list) → invoke → structured results. .mcp.json(project, shareable) vs~/.claude.json(personal secrets).- Env vars:
${GITHUB_TOKEN}in JSON, not hardcoded secrets. - stdio (local subprocess) vs HTTP/SSE (remote).
- Resources = catalog upfront (lower discovery token burn).
- toolOverrides to beat default Grep when semantic search is intended.
- No lazy-load of tools — limit tools per agent; hooks for visibility.
Ep 09 — Built-in tools
- Grep (content) vs Glob (paths); Read / Write / Edit / Bash.
- Incremental exploration: Glob → Grep → Read (don’t bulk-read huge repos).
- Exam: prefer targeted discovery over loading entire codebase.
Ep 10 — CLAUDE.md hierarchy
- Three levels: user / project / directory-specific rules.
.claude/rules/+ YAML front matter + glob paths for conditional rules (e.g.src/**/*.tsvsterraform/**/*.tf)./memoryto debug what loaded.- Sample exam pattern: different rules per file type → path-scoped rules, not one giant file.
Domain ↔ course ↔ this repo
| Domain | Weight | YouTube focus | Tracked practice |
|---|---|---|---|
| D1 Agentic architecture | 27% | Ep 01–04, 09 (exploration) | capstone_project.py, Task-Manager-App, scenarios 1–2, 6 |
| D2 Claude Code | 20% | Ep 05, 08–10 | p4-claude-code-workflow, course.html #claude-md #hooks |
| D3 Prompt engineering | 20% | Ep 01, 06, 10, 15+ | few_shot_examples.py, tool_misrouting.py, scenario 4 |
| D4 MCP & tools | 18% | Ep 06–08 | p5-custom-mcp-server, .mcp.json in course repo |
| D5 Context & reliability | 15% | Ep 03, 07 | subagent_context.py, fork_session.py, scenario 2 |
Six exam scenarios ↔ your anchors
| Scenario | Course episodes | Repo anchor |
|---|---|---|
| 1 Customer support | 01, 06, 07 | Task-Manager-App |
| 2 Research pipeline | 02–04 | CLAUDE-MD/AgentTeamsDemo |
| 3 Claude Code workflow | 05, 08–10 | p4-claude-code-workflow |
| 4 Enterprise RAG | 03, D3 Academy | DailyFlow |
| 5 MCP integration | 08 | p5-custom-mcp-server |
| 6 Automation pipeline | 02, 04–05 | Capstone + optional P3 build in 03-exam-scenarios |
Slide decks (local paths)
Under Youtube-main/claude-certified-architect/:
| File | Notes |
|---|---|
slide decks/lesson 1.pptx |
Ep 01 |
slide decks/lesson 2.pdf, lesson 3.pdf |
Image PDFs — text extract: scripts/pdf_extract_output/lesson_2.json, lesson_3.json |
slide decks/lesson 5.pptx … lesson 7.pptx |
Ep 05–07 |
slide decks/lesson_8.pptx, lesson_9.pptx, lesson_10.pptx |
Ep 08–10 |
docs/Claude_CertArch_Ep01_*.pptx … Ep04_*.pptx |
Alternate / earlier decks |
Missing numbering: no lesson 4.pptx in slide decks/ — use docs/Mayank_Chugh_Ep04_MultiAgent.pptx.
Runnable lab order (Python SDK)
# Use repo-root copy, or Youtube-main\claude-certified-architect if you have that clone
cd claude-certified-architect
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install anthropic
$env:ANTHROPIC_API_KEY = "sk-ant-..."
python api_check.py
python agent.py
python capstone_project.py
python research_summarizer.py
Snippets (multi_agents.py, tool_misrouting.py, …): read alongside matching episode; do not expect python file.py to run end-to-end.
Claude Code projects (tracked)
cd examples\CCA-F-PREP\p5-custom-mcp-server
pip install -r requirements.txt
$env:MCP_API_KEY = "dev-local-key"
python verify_setup.py
cd ..\p4-claude-code-workflow
$env:MCP_API_KEY = "dev-local-key"
python verify_setup.py
Reference docs in course repo
| File | Purpose |
|---|---|
CLAUDE |
Setup, conventions, file map |
docs/agent-guide |
Single-agent loop + coordinator pattern |
.claude/skills/create-agent/ |
Scaffold agents in Claude Code |
Maintenance
| Task | How |
|---|---|
| Add Ep 09–10 transcript | Replace placeholders in docs/Transcripts of youtube/9-*.txt, 10-*.txt (Merlin or manual) |
| Regenerate PDF notes | python scripts/extract_pdf_content.py from repo root |
| Sync new episodes | Extend table above + 01-skilljar-academy-checklist |
Last aligned to local folder scan: Ep 01–08 complete, slides through lesson 10, code through Ep 08 + snippets for 09.