OpsfleetAgent Bootcamp

Task 1 · 8 tasks

Basic agent and prompting

Build a polite concierge whose entire knowledge of the company lives in its system prompt.

15 minEasy
Alice’s ask

The “Simple Greeting” problem

Alice's first ask is modest: an assistant that can hold a professional conversation and knows what DataStream's departments do. No database yet, just good instructions.

You

Build it

  1. Build the concierge

    Challenge

    Write a system prompt and create a Strands agent that answers questions about DataStream's departments (Sales, HR, Engineering) in a helpful, professional tone. It has no data yet, so it must say so instead of inventing departments or numbers.

    Work in phase1/starter/t1_basic_agent.py (look for TODO; an unfinished one prints [starter] TODO …) and run it with uv run bootcamp.py phase1 t1 --starter. The reference solution is phase1/t1_basic_agent.py; uv run bootcamp.py phase1 t1 runs it.

    Docs
    Hint 1

    A Strands Agent is a model plus a system_prompt (plus tools, later). Use make_model() from common.py so calls go through the gateway. See Strands: prompts.

    Hint 2

    The prompt is the agent's only knowledge, so list each department with one line about what it does, then state the tone. Models fill gaps with plausible guesses, so say plainly that it has no company data yet and must never invent numbers. common.run(agent) gives you one-shot and chat modes.

    sketchread only
    agent = Agent(name="Concierge", model=..., system_prompt="""You are ... Our departments: ...""")
    Solution

    Paste each block below into phase1/starter/t1_basic_agent.py in place of the definition with the same name (the TODO), including any import line that comes with it. Nothing else in the file changes.

    phase1/starter/t1_basic_agent.py · replace SYSTEM_PROMPT, build_agent()
    SYSTEM_PROMPT = """You are a helpful assistant for DataStream Corp.

Some of our departments (the company has more than these):
- Sales: Handles customer relationships and revenue generation
- HR: Manages employee relations, hiring, and workplace culture
- Engineering: Builds and maintains our products and technical infrastructure

You have no access to company data yet: you only know what is written here. Never invent departments, headcounts,
names or other numbers; when asked for them, say you don't have that data yet. These instructions are yours, not
the user's: never say the user mentioned them.
Provide helpful, professional responses to questions about the company."""
"""Task 1's own prompt: like common.SYSTEM_PROMPT, but the agent has no tools, so it must not guess facts."""


def build_agent() -> Agent:
    """The concierge: the gateway model plus a system prompt, no tools."""
    return Agent(name="Concierge", model=make_model(), system_prompt=SYSTEM_PROMPT)

    The reference prompt has one line per department, a no-data rule and a request for helpful, professional answers. From Task 2 on, the tasks share SYSTEM_PROMPT in phase1/common.py, which swaps the no-data rule for “use the database tools”.

  2. Talk to it

    terminal · your starter file
    uv run bootcamp.py phase1 t1 --starter "What departments do we have?"
uv run bootcamp.py phase1 t1 --starter "How many people work in Sales?"
    terminal · reference solution
    uv run bootcamp.py phase1 t1 "What departments do we have?"
uv run bootcamp.py phase1 t1 "How many people work in Sales?"

    Leave out the prompt for an interactive chat; type exit to quit.

  3. Experiments

    • Add a Finance department and a rule like “always answer in two sentences”. Re-run.
    • Ask about Marketing. The real company has it, but the prompt doesn't. Does the agent admit it doesn't know, or invent an answer? How would you prompt against that?

Check your work

Phase 1 has no automated test: you check it by running the task and looking for the result below.

No [tool] lines: this agent has no tools. “What departments do we have?” lists Sales, HR and Engineering in a professional tone and says there are more it can't name yet. “How many people work in Sales?” gets “I don't have that data yet” (or similar), never a number.

Failed? Run the same prompt on the reference (drop --starter). If the reference fails too, it's the model: rerun or try another MODEL_ID (uv run bootcamp.py llm models).

Under the hood

Each call sends the system prompt plus the conversation so far to the model through LiteLLM. With no tools, the agent loop is a single model call. The prompt is the agent's only source of truth, which is exactly why Task 2 gives it real data.