Classroom exercises built on TopMVA

1. Give students one synthetic MVA intake narrative (invented names, plausible crash, labeled synthetic).
2. Round 1: students prompt a general AI assistant freehand: "summarize this intake for a supervising attorney."
3. Round 2: students install the intake skill and run the same narrative.
4. Compare on four axes: structure; what got invented or assumed; what got flagged for review; what a supervising attorney would still have to check.
5. Discussion: the controlled skill is not "smarter" — it is scoped. What does the scoping buy, and what does it still not buy?

1. Take the limitations-watch section of any skill output (or the propositions in mo-authority.md).
2. Students verify each proposition against the primary source: the subsection on revisor.mo.gov, the opinion on CourtListener — including later history.
3. Students classify each item using the definitions on the research page: verified, open question, unverified.
4. Debrief with the verification protocol — especially the eleven published errors, which make an honest syllabus unit on how verification actually fails.

Have students run the demand builder on a synthetic file, then rewrite the draft as their own: tighten the liability narrative, re-argue general damages from the treatment story, and justify every number against the synthetic record. Grade the rewrite, not the draft — the pedagogical point is that the draft is raw material, and editing it well is a distinct, teachable writing skill.

Use the skills as a live case study for the duties framework: competence (what must the lawyer independently verify?), confidentiality (what could permissibly enter the tool, under which model rules and comments?), supervision (how does responsibility for nonlawyer assistance map onto AI output?), and candor (what happens when an unverified citation reaches a tribunal?). The "Items for attorney review" block that ends each demand draft is a ready-made discussion artifact: who is responsible for each item on it?

• Classroom use on synthetic facts: appropriate.
• Clinic use on real client matters: a supervising-attorney decision that must be made deliberately — evaluating the tool's confidentiality posture, client consent policy, and the clinic's own AI-use rules — before any real fact enters any tool.
• Student output on a real matter is never filed, sent, or shown to a client without supervising-attorney review and signature.
• These materials are drafting aids, not legal advice, and do not change who is responsible for clinic work product.

• Every skill is plain text — assign the SKILL.md itself as reading. The instructions are the curriculum: scoping, refusal rules, flagged assumptions.
• Run the demos yourself first; model outputs vary between runs and models, and the variance is itself teachable.
• Pair each generation exercise with a verification exercise. Generation without verification teaches the wrong habit.
• Installation takes about two minutes per skill — see the get-started steps.

• Verification rigor (40%) — every citation and figure in the student's final product traced to a primary source or explicitly flagged as unverified.
• Critical editing (30%) — substantive improvement over the raw draft: argument structure, accuracy, tone, jurisdiction fit.
• Assumption handling (20%) — flagged assumptions resolved or escalated, never silently accepted.
• Professional-responsibility analysis (10%) — accurate mapping of the duties implicated by the workflow used.

Course materials that adapt these skills should cite the project (formats at /cite/), identify modifications, and preserve the canonical source URL. Skills and written materials are under the TopMVA Skills License; classroom and clinic use is squarely within the permitted internal use. Questions: GitHub issues.