CUHK-Business-School-AI-Hub
OrganizationA change of perspective on your own teaching. Audits a lecture you already teach from the student's point of view, then rebuilds what you accept: activities, rotatable datasets, productive-failure cases, and edits to the slide file itself. One Claude skill.
Categories
Indexed Skills (6)
ai-teaching-guardrails
How a teacher should work with an AI assistant on academic work, and how to check what it produces. Use whenever helping with teaching material, course design, marking schemes, assessment policy, or any academic writing where the person will put their name on the result. Covers what to tell the model before it starts, what to ask for rather than assume, how to keep it honest, how to check output too long to read, when to write plainly and when to keep the subject's technical vocabulary, how to check material from a student's point of view, how to design for students who all have AI, which tasks a teacher must never hand over, and how to set an AI policy for staff rather than students. Also use when someone asks whether they can trust an AI answer, how to check it, or what to disclose.
case-authoring
How to author a productive-failure teaching case, where students critique a deliberately weak AI's answer before instruction and then repair it. Use when designing a new case, a hands-on exercise, a lab or a notebook students work through, when turning a lecture into something students can run, or when a case demos badly, feels overloaded, or lets students score well by flagging everything. Covers one concept per case, authoring the data so untaught errors cannot happen, putting the failure in the conclusion rather than the arithmetic, and keeping the marking scheme out of reach. See BUILDING.md for the folder that makes it run.
course-audit
Work out what a piece of teaching material is for, what is really in it, and where it falls short from a student's point of view. Use when someone shares a lecture in PowerPoint, PDF or LaTeX, a notebook, a syllabus, an assignment or a whole course folder and wants to know what to improve, what is worth keeping, or which parts an AI tool has made easy for students. Establishes the topic and the core objective first, then measures the file, then reads it as somebody who does not yet know the answer. Hands over a short written verdict, the evidence behind it, numbered findings the lecturer marks accept, reject or change, and last a table of what had to be assumed, which is what course-rebuild then works from. Also use for "is my course still fit for purpose", "what should I update", or "where is AI a problem in this course".
course-rebuild
Answer the findings from a course audit with concrete proposals, then build what the lecturer chooses. Use after course-audit, or when a teacher wants to make a lecture more interactive, turn slide questions into activities students actually answer, build practice datasets that rotate each year, write speaker notes another teacher could use, rewrite exam questions AI can now answer in a minute, or change the slide file itself. Offers two or three formats for each problem and lets the lecturer pick rather than choosing silently. Also use for "make this more engaging", "I need a new version of this dataset", or "how do I assess this now that ChatGPT exists".
safe-to-publish
Decide whether teaching material is safe to put in front of people, and stop the marking scheme escaping. Use when a case, an answer key, a dataset, a handout or a whole site is about to become visible to students or colleagues, when someone asks whether a solution file could leak, or when a check "keeps getting ignored". Covers putting a deterministic gate between authored and published that refuses rather than warns, exporting by whitelist rather than filter, planting sentinels in everything that must not escape, and the version of both patterns for material that is just files rather than an application.
verify-claims
Check every citation and number in a draft using a verifier that never sees the draft. Use when someone says "verify my citations", "check these numbers against the source", "fact-check this section", or before publishing anything that quotes a source or states a figure it did not just compute. Extracts the checkable claims, asks an independent fresh-context check on each, and reports PASS, PARTIAL or FAIL with a severity. Also use before handing out teaching material that cites research, quotes a filing, or prints a number a student could look up.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.