← AI Resources

Colophon

How this site is written, hosted, published, and kept current — and which parts a machine drafted.

Written with AI, and saying so

This portal tells faculty that transparency is the right default, and that if AI contributed meaningfully to a piece of work you should say so. It would be strange not to apply that here.

Most of these pages were drafted in Claude Code and then edited, cut, and corrected by hand. The model does not write unsupervised: it works from a standing instruction file in the repository that sets voice and house rules, and from a set of reusable skills. Every name, date, statistic, and citation is checked against a source before it ships — the tools invent plausible ones otherwise, which is the whole reason the Accuracy guidance exists.

What that looks like in practice

Colleagues have asked to see the actual prompts rather than a description of them. Here is a real one, and what it produced.

That produced an audit of every teaching-related section here, which turned up five skills described in full on both this portal and the pedagogy portal — same tools, two sets of prose, and display names that had already drifted apart. The fix replaced the duplicated cards with a single handoff, and shipped as a pull request against a public repository. The prompt is unremarkable, and that is rather the point: it names a goal and a constraint, and leaves the method open.

The reusable pieces are public. The skill files are plain-English instructions, not code — reading one is the fastest way to see how these tools are actually steered.

Hosting

Publication pipeline

There is no build step and no framework. The site is hand-written HTML with its CSS and JavaScript inline, which means a page can be opened straight from disk in a browser and looks exactly as it will once published.

• Substantive edits go on a branch and through a pull request; main is kept deployable.
• Merging to main triggers the Pages build. Changes are live in a minute or two.
• A Jekyll _config.yml exclusion list keeps repository notes — instructions, status files, working decks — out of the published tree, so internal scaffolding never ships.
• No GitHub Actions. The whole pipeline is git plus Pages.

Kept current by a scheduled agent

AI tooling changes faster than anyone can re-read a reference site, so a scheduled Claude Code routine re-checks this one every two weeks: link targets, version claims, pricing, and which tools Penn currently licenses. It opens ordinary commits, which are visible in the repository history alongside everything else.

It has one blind spot worth naming. PennChat sits behind Penn's network and a login, so no unattended agent can reach it; those claims are verified by hand.

What stays human

The judgment. What belongs on this site and what belongs on the pedagogy portal, which recommendations are worth making to colleagues, what is too uncertain to assert, and everything written in the first person. Errors here are mine.

Typography and colour

Headings are set in EB Garamond and body text in DM Sans, both served from Google Fonts. The palette is Penn navy #011F5B, red #990000, cream #FAF8F3, and gold #756000. The four-pixel navy-and-red bar at the top of every page is shared with pennai.law, so the two sites read as siblings.

Source and licence

The repository is public: github.com/pennai-law/penn-law-ai-resources. Original prose, editorial content, graphics, and visual design are CC BY 4.0; HTML structure, CSS rules, and JavaScript are Apache 2.0. Files may contain both kinds of material. See the licence page for the license map and complete terms.