Reference
The parts we actually grade
We do not hand you a finished prompt to copy. Every graded step in this course is judging one or more of the parts below, by name, against the message you actually wrote. Here is the whole list, in the words the course itself uses, each one linked to where it is actually graded.
02
the deal’s facts
states concrete deal specifics, or supplies the document that carries them — pasted into the message or attached to it
Graded in Build a Complete Request04
a sharper verb
directs the assistant with a specific action word or phrase (compare, flag, calculate, recommend — not a vague "look into")
Graded in Choose Better Words05
a guardrail: one line on what not to do
explicitly says what NOT to do or how to handle unknowns (e.g. don’t invent numbers; if unsure, say so)
Graded in Choose Better Words07
the example put to work
includes a pasted example and explicitly tells the assistant to match its style on the new content
Graded in Choose Better Words08
a brief add-on — not a restart
a brief follow-up that adds or changes one thing, building on the conversation instead of restarting it
Graded in Fix an Answer Without Starting Over09
a rule that travels
a durable, reusable instruction phrased to apply to every future answer (a preference or guardrail), not a one-off task
Graded in Know What AI RemembersTwo more, once you are grounding real documents
10
the names taken out
keeps the operative deal terms but carries no personal identifiers — no individual’s name, home address, account or policy number, employer, or family detail
Graded in Share Only What the Job Needs11
the line you are rejecting, and why
rejects a SPECIFIC figure or claim the assistant just produced — names the row, unit or line it is wrong on, states what the value should be or why it cannot stand, and points at what in the source file establishes that. A vague "check your work", "are you sure?" or "that looks off" is NOT an overrule
Graded in Build a Reusable Claude Code Skill12
your decision, and the rule that makes it
commits to a decision state in the learner's own voice — names the call the criteria settle and the rule, evidence or source that produces it. A message that asks the assistant to choose, hedges the call ("likely", "conditionally"), or lays out options without deciding is NOT a decision
Graded in Decide Whether a Site QualifiesNothing on this page is a prompt to paste. Every one of these is a rubric a real exercise checks your own words against.
Back to the course →