Course
CCA-F Prep
Personal learning and development platform
Progress at a glance
- 0 mastered
- 0 in progress
- 41 not started
Per-section breakdown
- Section 1: Claude 1010/4
- Section 2: AI Fluency: Framework & Foundations0/5
- Section 3: Building with the Claude API0/6
- Section 4: Introduction to MCP0/6
- Section 5: Claude Code 1010/4
- Section 6: Claude Code in Action0/5
- Section 7: Introduction to Agent Skills0/4
- Section 8: Introduction to Subagents0/4
- Section 9: Introduction to Claude Cowork0/3
Time estimate is a credible minimum: ~17 min per concept (lesson + first quiz), 10 min per section test, 30 min per mock exam. Most learners end up at 1.5–2× this — budget headroom.
Modules
Mental model of Claude.ai products. Projects, artifacts, file attachments, model picker. The non-API surface.
- B1.1Project vs RAG corpusU
- B1.2Product-tier vs API-level limitA
- B1.3Artifact vs inline messageE
- B1.4Context cost of N file uploadsA
The 4D framework — Description, Delegation, Discernment, Diligence — for diagnosing AI failures by structural cause, not surface symptom.
- B2.1Classify failure into a 4DAn
- B2.2Identify mis-delegationAn
- B2.3Write a structured descriptionA
- B2.4Run description ↔ discernment loopA
- B2.5Name the diligence concernE
Message API, system vs user roles, tool use, structured output, prompt caching, few-shots as turns.
- B3.1Place instruction in system vs userE
- B3.2Few-shots as alternating turnsA
- B3.3Trace tool-use loop / stop_reasonAn
- B3.4Place cache_control markerE
- B3.5Tool-schema vs prompt-only structured outputE
- B3.6Detect system-field leakAn
Model Context Protocol primitives (tool / resource / prompt), transports (stdio / HTTP+SSE), and what auth boundary they imply.
- B4.1Classify tool / resource / promptAn
- B4.2stdio vs Streamable HTTPE
- B4.3Reject non-MCP transportsR
- B4.4Diagnose wrong-tool-pick to descriptionAn
- B4.5Server-per-capability vs monolithE
- B4.6Auth at transport, not in tool inputA
Claude Code anatomy: hooks, slash commands, permissions, the CLAUDE.md / settings.json split.
- B5.1Place hook on correct eventA
- B5.2Predict permission cascadeAn
- B5.3Route rule (CLAUDE.md / reference / skill / slash / hook)E
- B5.4Author slash commandA
Lifecycle events end-to-end, hook trust levels, subagents vs inline, CI surface, watermarks.
- B6.1Order all hook lifecycle eventsR
- B6.2Hook vs CLAUDE.md by trust levelE
- B6.3Predict no-watermark loop failureAn
- B6.4Subagent vs inlineE
- B6.5CI surface blast radiusAn
Skills as user-invoked capabilities. Routing between skill / slash / CLAUDE.md / hook by invocation model.
- B7.1Skill vs slash vs CLAUDE.md vs hook (by invocation model)An
- B7.2Critique skill description for discoverabilityE
- B7.3Place skill (project / user / plugin)A
- B7.4Detect kitchen-sink skillAn
Subagent triggers (parallelism, context isolation), the cost model, and what the parent does and does not see.
- B8.1Two valid spawn triggersR
- B8.2Critique underbriefed subagent promptE
- B8.3Cost of unnecessary subagent splitsAn
- B8.4What parent does NOT see from subagentU
Long-running agentic work, plugins vs skills, checkpoint cadence, the set-and-forget anti-pattern.
- B9.1Plugin vs skill (Cowork)U
- B9.2Add appropriate human-checkpoint cadenceE
- B9.3Detect set-and-forget antipatternAn
Practice exams
Independent of module progress. Use for calibration.
Diagnostic 01 — Mixed Domain
10 Q · 20m · 70% pass10 questions, ~20 min. Cold calibration across all five exam domains. Take it without notes — the score band tells you which domain to train first.