Prompt Engineering for Agents

Hands-on lab with starter code you can copy + run, and guided steps to produce the artifact for this module.

Setup

Open a fresh terminal in a new folder, e.g. mkdir lab-4 && cd lab-4.
Create a Python 3.11 venv: python -m venv .venv && source .venv/bin/activate.
Install the dependencies noted on each starter snippet (the small pip install … line above the code).
Copy each starter into a file with the suggested name, then run it and capture the output.

Starter code

01 · Structured system prompt for an agent

save as: 04_01_lab.py

python

SYSTEM = """You are GACS-Agent v1.
ROLE: financial-fraud investigator.
TOOLS: search_web(q), check_blacklist(name), report(json).
RULES:
1. Never fabricate evidence.
2. Always cite a tool observation before claiming a fact.
3. If unsure, ask the user one clarifying question, then continue.
OUTPUT: JSON {thought, action, action_input}."""

def call(user_msg):
    return client.chat.completions.create(
        model="gpt-5-mini",
        response_format={"type":"json_object"},
        messages=[{"role":"system","content":SYSTEM},
                  {"role":"user","content":user_msg}],
    ).choices[0].message.content

Run with: python 04_01_lab.py

Guided steps

Work through each step in order. Stop when your output matches the expected artifact — then push your work to your lab folder.

1
Structured system prompt
Do — Write a system prompt that enforces JSON output with fields {thought, action, action_input}. Validate 20 outputs.
Expected artifact — 20/20 outputs parse as valid JSON with all 3 required fields.
Run free:
2
Persona ablation
Do — Run the same 10 tasks with and without a domain-expert persona. Compare answer quality.
Expected artifact — persona.md table with subjective 1–5 scores and an aggregate delta.
Run free:
3
Few-shot vs zero-shot
Do — Build a tool-calling agent with 0, 2, and 5 in-context examples. Test on 30 prompts.
Expected artifact — Tool-call accuracy curve; 5-shot should outperform 0-shot by ≥10 points.
Run free:

Authoritative sources

Independent primary sources used to check and corroborate the guidance on this page.

NIST National Initiative for Cybersecurity Education (NICE)
US national framework for cybersecurity work roles and competencies.
Financial Action Task Force (FATF)
Global AML/CFT standards referenced by financial-crime curricula.
FBI Internet Crime Complaint Center — Annual Reports
Primary loss and typology data used in fraud-investigation training.
US Federal Trade Commission — Consumer Sentinel Data Book
Official complaint statistics for teaching fraud trend analysis.