Lab Guide

Two hands-on exercises covering RTCF rewrites and output format control.

Setup: Open lab.pdf in Jupyter. Make sure your .env file contains ANTHROPIC_API_KEY. Run the setup cell before starting any exercise.

📥 Lab Notebooks

⬇ lab.pdf ⬇ lab_answer.pdf

Exercise 1 — RTCF Rewrites

EX 1Rewrite poor prompts using the RTCF framework

Below are 5 poor prompts. For each one, write an improved version that includes all four RTCF components (Role, Task, Context, Format). Then use the compare_prompts() helper in the notebook to call Claude with both the original and improved version and compare the outputs side by side.

The 5 poor prompts

Prompt 1

"write something about climate"

Problems: No audience, no angle, no length, no format. Claude must invent everything.

Your goal: Produce a specific prompt that reliably generates a focused, usable piece of text.

Prompt 2

"summarize this"

Problems: No document is provided. No target length. No audience.

Your goal: Add a sample document (paste any ~200-word article) and specify summary format.

Prompt 3

"be helpful"

Problems: Not a task — it's a vague disposition. Claude has no idea what to do.

Your goal: Turn this into a specific task (e.g., a customer support agent instruction).

Prompt 4

"fix my code"

Problems: No code is provided. No language specified. No description of what's broken.

Your goal: Write a proper prompt that includes a code snippet, the error, and the desired fix format.

Prompt 5

"what do you think about AI?"

Problems: Invites a long, unfocused opinion essay. No angle, no constraint.

Your goal: Rewrite as a specific analytical task with a defined output format.

What to submit

Your 5 improved prompts (as strings in the notebook)
Side-by-side outputs from compare_prompts()
One-sentence note per prompt: what changed and why it matters

Self-check: Each improved prompt should have all 4 RTCF components. Read each one aloud — would a literal-minded human ask a clarifying question? If yes, add more specificity.

Exercise 2 — Format Control

EX 2Extract contact info in three different formats

Use the following block of text (provided in the notebook) as your input:

The project kickoff meeting was attended by Sarah Chen (sarah.chen@techcorp.io),
lead engineer Marcus Webb at m.webb@devstudio.com, and the client contact
Priya Nair — her email is priya.nair@clientco.org. Also cc'd were Jordan Ellis
(jellis@techcorp.io) and the PM, Dana Okafor, reachable at d.okafor@pm.io.

Write 3 separate prompts that extract the names and email addresses from this text and produce three different output formats:

Format A — JSON array

Expected output structure:

[
  {"name": "Sarah Chen", "email": "sarah.chen@techcorp.io"},
  ...
]

Your prompt must specify this exact JSON structure. The output should parse cleanly with json.loads().

Format B — Markdown table

Expected output structure:

| Name | Email |
|------|-------|
| Sarah Chen | sarah.chen@techcorp.io |
| ... | ... |

Your prompt must produce a clean markdown table with a header row.

Format C — Plain bullet list

Expected output structure:

- Sarah Chen: sarah.chen@techcorp.io
- Marcus Webb: m.webb@devstudio.com
...

Your prompt must produce a plain bullet list, no extra commentary.

What to compare

Did each prompt produce the correct format on the first try?
Did the JSON prompt produce parseable JSON (test with json.loads())?
Were all 5 contacts extracted in every format?
Which format required the most precise prompt wording?

Self-check: For the JSON format, add assert len(result) == 5 to verify all contacts were found. For the table, count the rows. For the list, count the bullet points.