Description-Tuning Drill

Module 7, Lesson 7.3 — the description is the classifier. Find out where five candidate descriptions fail (or don't).
The headline insight. A skill description is not marketing copy. It is the classifier the agent reads to decide whether to invoke the skill. Three failure modes dominate — vague, overfit, triggerless — and one target shape wins: descriptive + inclusive + exclusive (D+I+E).
Audience = only you. All five skills in this drill are authored for the student's own use. No sharing.

How the drill works

  1. Five skill descriptions follow. Each is shown as it would appear in the skill's frontmatter — exactly the text the agent's classifier would read.
  2. Each description is paired with five test requests. For every request, predict whether the classifier should fire this skill on that request (YES) or should not (NO).
  3. After five predictions, pick the description's failure mode:
    Vague Description is under-specified. Fires on many requests that have nothing to do with the skill's actual job. High false-positive rate. Overfit Description is tied to a one-time, hyper-specific context (named class, named teacher, dated assignment). Fails to fire on valid adjacent requests. High false-negative rate. Triggerless Description has no descriptive clauses — nothing the classifier can match against. Rarely fires, and the reasons are opaque. Well-tuned D+I+E shape: descriptive about the job, inclusive of valid invocation phrases, exclusive of what it is not for. Low both-kinds-of-error rate.
  4. Submit to see which predictions matched the reference, the correct failure mode, and the one-sentence fix (or, if already well-tuned, the reason nothing needs to change).

There is no perfect score. The drill is calibration. The descriptions you disagreed with matter more than the ones you nailed — those are the failure modes most likely to recur in your own 7.3 authoring work.