Lessons

• Lesson 1 Noun: name the artifact
  Most weak prompts never say what they want made.
• Lesson 2 Verb: task precision
  Looking is not doing.
• Lesson 3 Adjective: quality constraints
  The model will write something summary-shaped and stop.
• Lesson 4 Adverb: manner and degree
  "Carefully" tells the model to feel careful.
• Lesson 5 Preposition: scope and relation
  One boundary is not enough. Which problems, for whom, at what cost?
• Lesson 6 Composition: the first five together
  A composed prompt is not five sentences bolted together.
• Lesson 8 Pronoun: reference binding
  The model has to guess which thing you want changed.
• Lesson 9 Conjunction: conditional logic
  "And" bolts two directives together without saying how they relate, so the model has to invent both the branch and its outcome.
• Lesson 10 Determiner: definiteness and quantity
  A bare noun doesn't say how many: one error, a handful, or all of them?
• Lesson 11 Numeral: checkable budgets
  "Some options for your task" leaves both count and length open.
• Lesson 12 Interjection: attention markers
  Five instructions in a row read as five equal priorities, even when one matters far more than the rest.
• Lesson 13 Particle: phrasal precision
  Say "look over" when you meant "look up" and you get a skim where you wanted a lookup — wrong operation, nothing looked broken.
• Lesson 22 System prompts: separate what stays from what changes
  A prompt that repeats its rules every turn pays the same tax twice: once to write the rules, once to have the model re-read them alongside whatever's actually new.
• Lesson 23 Agent and tool prompting: tools, timing, and done
  "Use the tools to sort out your task" says neither which tool serves which purpose nor when the work should stop.
• Lesson 24 Self-critique loops: name a check that can fail
  "Review your answer before you finish" asks for a check with no test attached.
• Lesson 25 Writing evals: write criteria before you look
  "Make sure it is good" names a feeling, not a test.
• Lesson 26 Token economy: cut what doesn't earn its place
  Unused material doesn't just waste tokens; it dilutes the signal the model needs.
• Lesson 28 Prompt library: save the slots, not just the words
  Saving the words is not the same as saving the prompt.