Exercise: An Executive Summary Bot with the Anthropic SDK

After the Brightline kickoff, the sponsor (Dana, VP Operations) asks for a short summary she can forward to her CEO. You'll write it many times over the next months, so you decide to automate a first draft with Claude.

Your task

The editor already creates the client and defines MODEL. Complete three things.

1. SYSTEM_PROMPT: a system prompt string that tells Claude its role. It must mention that the audience is an executive (the word "executive" must appear) and ask for a short summary. Write it the way you'd brief a sharp human assistant.

2. summarize_for_exec(client, notes):

• Call client.messages.create with:
  • model=MODEL
  • max_tokens of at least 1024
  • system=SYSTEM_PROMPT
  • messages containing one user message whose content includes the full notes text
• If stop_reason is "refusal", return None.
• Otherwise return the text of the response: join every block whose type is "text", then strip surrounding whitespace.

3. estimate_cost(usage, input_price=4.0, output_price=20.0): given a response's usage object, return the cost of the call in US dollars, rounded to 6 decimal places. Prices are per million tokens (defaults are Claude Opus 5.5 pricing).

Things the tests check

• You passed the right parameters (model, max_tokens, system prompt, a single user message containing the notes).
• You don't assume the first content block is text. The simulated response, like a real Opus 5.5 response, starts with a thinking block.
• You handle a refusal without crashing.
• Your cost math is right.

Press Run to see your summary and its cost, then Submit.

FDE tip: Treat LLM output as a draft for a human, especially anything going to an executive. A good pattern is "Claude drafts, FDE edits, FDE sends." You'll learn to measure and improve draft quality with evals in Module 8.