Domain 1 — Agentic Architecture & Orchestration

CCA Foundations course · Page 2 of 8 · ← Back to all courses · Weight: 27% (the heaviest domain — master this first). Mark this page complete at the bottom to advance your course progress.

1.1 — The Agentic Loop · Core

The loop: send request → inspect stop_reason → execute tool → append result → repeat.

• stop_reason values: "tool_use" (continue — run the tool), "end_turn" (done), "max_tokens" (truncated), "stop_sequence".
• Control flow is driven ONLY by stop_reason. Never parse the model's text to decide whether to continue.
• An iteration cap is a safety net only — never the primary stopping mechanism.
• Tool results are appended to the conversation so the model can reason about the next step.

1.2 — Multi-Agent Orchestration · Core

• Hub-and-spoke: one coordinator manages all inter-subagent communication, routing, and error handling.
• Parallel subagents: emit multiple Task calls in one response (not across separate turns).
• Zero context inheritance — subagents do not inherit the coordinator's history. Pass everything explicitly in the subagent prompt.

1.5 — Hooks: Pre vs Post ToolUse · Core

• PreToolUse = block/gate BEFORE execution (e.g. verify identity before a refund; block refunds over a threshold).
• PostToolUse = normalise/enrich results after execution, before the model sees them.
• Deterministic over probabilistic for financial / compliance / security — use hooks, not prompt instructions.

⚠ Often-missed — PostToolUse for Multi-Tool Normalisation · Gap

When multiple tools return different currencies / units / formats, a PostToolUse hook normalises ALL of them to a common format before Claude sees the results — so the model reasons over consistent data.

1.7 — Session Management · Core

• --resume — continue a session when prior context is still valid.
• fork_session — branch from a shared baseline to explore divergent approaches.
• Stale context → start fresh and inject a summary rather than resuming with stale tool results.
• Be aware of /compact risks (it can drop exact values — protect critical facts; see Domain 5).

⚠ Often-missed — Structured State Persistence / Crash Recovery · Gap

Periodically export a JSON manifest: goal + processed files + remaining files + key entities. On recovery, start a fresh agent and inject the manifest — far more reliable than resuming a long, stale session.

Exam reflexes for Domain 1

• "CI pipeline hangs" → not here, but remember -p flag (Domain 3).
• "Narrow results from multi-agent" → fix the coordinator's decomposition, not the subagents.
• "Parallel agents" → multiple Task calls in ONE response.
• "Must guarantee a step happens" (refund/identity) → programmatic gate (PreToolUse), not a prompt.
• "Agent crash recovery" → periodic JSON manifest + inject on resume.