AI Agents35% of the exam
Agent Script building blocks and the agent graph
What Agent Script is and which blocks define an agent (system, config, variables, start_agent, subagents), how the script compiles into an Agent Graph that the Atlas Reasoning Engine runs, and what belongs in before_reasoning and after_reasoning.
Explain how an agent works and its basic building blocks of agent script.
~8 min read · 12 practice questions ·
What the exam expects
- Describe Agent Script and name the building blocks it defines: subagents, instructions, variables, actions and transitions.
- Explain how an agent turn runs, from the starting subagent to the LLM reasoning loop.
- Trace the three-stage pipeline from authored script to Agent Graph to the Atlas Reasoning Engine, and say which layer you author.
- Choose between agent-level (global) instructions, per-subagent instructions and a fallback subagent.
- Place logic correctly in before_reasoning, reasoning or after_reasoning, given how often each runs.
- Key terms: Agent Script (the language for building agents in Agentforce Builder); subagent (formerly topic: one job the agent can do); Agent Graph (the compiled execution plan); Atlas Reasoning Engine (the runtime that traverses the graph); parse (one full cycle through a subagent's lifecycle blocks).
Key facts
Compilation
- Detail
- Agent Script compiles when you save a version of the agent
System messages
- Detail
welcomeanderrorare required in the system block
Entry point
- Detail
- start_agent is the entry block (Agent Router in Canvas view);
reset_to_initial_nodeTrue restarts there every turn, and it is off by default
Parse triggers
- Detail
- First entry, after every tool call whose action completes and returns a result, every new user turn in the subagent
Displayable output
- Detail
- Per the architecture guide,
is_displayable: Trueexits the loop at once and after_reasoning never runs
Mid-subagent transition
- Detail
- The original subagent's after_reasoning isn't run
Lifecycle block limits
- Detail
- No
|command; not allowed in a subagent that uses the HyperClassifier model
- Why can one customer message run a subagent's before_reasoning block more than once?
- When is a subagent's after_reasoning block not run?
- Which layer of the Agentforce pipeline does the team author and debug?
- Why put tone and escalation rules in agent-level instructions rather than each subagent?
The questions are free to read. Answers with their sources and self-grading are included with the pack.
Keep reading with the Agentforce Specialist pack
- Core concepts
- Exam traps
- Worked scenario
The pack includes:
- All 26 lessons of the Agentforce Specialist guide, with reading progress
- Interactive decision walks and step-by-step sequence steppers
- Inline practice questions and one-click topic practice (12 questions for this lesson)
- An official source cited for every claim
- An AI tutor that answers from the guide
- Unlimited timed practice exams