Library AI Workshop

A SvelteKit eLearning application that helps research librarians use graphical AI tools safely and critically. No coding is required. The June 2026 curriculum is product-neutral across ChatGPT, Claude, Gemini, and Microsoft 365 Copilot and follows four modules:

1. Safe Setup & the Reference Interview — Set a data boundary, scope a research request, and build a concept map
2. Search & Source Verification — Review research plans, inspect source sets, and audit claims against citations
3. Evidence Synthesis & Data — Build claim-evidence matrices, preserve disagreement, and verify calculations
4. Reproducible Research Support — Test database syntax, teach critical AI use, and package accountable handoffs

An optional fifth module sits outside that product-neutral core:

5. Bonus: The Skill Marketplace — Appraise, install, author, and govern AI skills as a collection, using Kanon and the agentic-skill-library

The bonus module needs no terminal, but it does need the plugin. Its default surface is Claude Cowork, where the library installs through menus (Customize → Plugins → Browse plugins → Add marketplace) and every step is a copy-paste prompt; Claude Code and Codex work too, and add the optional authoring and publishing steps. Because the module works against a real installed plugin, permission to install one is a prerequisite — confirm it with the cohort before the session.

It does not try to teach the CLI from scratch. The marketplace ships a kanon skill written for library staff — its authoring guide teaches artifact metadata through Dublin Core — and the module hands learners to it, then points anyone wanting depth at that skill's twenty-lesson tutorial, self-paced course, and curriculum guide.

The curriculum is grounded in the ACRL AI Competencies for Academic Library Workers and the ALA Guidance on the Use of Artificial Intelligence in Libraries. AI output is treated as draft material requiring meaningful human review.

Progress is tracked in AWS DynamoDB. A facilitator dashboard shows cohort progress, pacing alerts, and talking points keyed to the current exercise.

See FACILITATOR.md for the full run-of-show guide.

---

Prerequisites

• Node.js 20+
• An AWS account with DynamoDB access
• A ChatGPT, Claude, Gemini, or Microsoft 365 Copilot account that participants may use for the workshop
• File upload access and, for Module 2, web search or a longer-running research mode
• The src/content/library-context/ folder accessible to participants
• For the optional bonus module only: Claude Cowork (or Claude Code / Codex) plus permission to install a plugin on the machine participants are using. No terminal is required. The release package carries SKILL-MARKETPLACE-PROMPTS.md and SKILL-MARKETPLACE-HANDOUT.md for that module, but does not bundle Kanon or the Context Bazaar marketplace itself. See Distribution.

---

Setup

1. Install dependencies

npm install

2. Configure environment

cp .env.example .env

Edit .env with your role ARN and settings:

AWS_REGION=us-east-1
AWS_ROLE_ARN=arn:aws:iam::123456789012:role/LibraryWorkshopRole
AWS_ROLE_SESSION_NAME=library-workshop-session
DYNAMODB_TABLE=LibraryWorkshop
FACILITATOR_TOKEN=choose-a-secure-token
PUBLIC_WORKSHOP_TITLE=Library AI Workshop
PUBLIC_COHORT=spring2026

The app assumes the role specified in AWS_ROLE_ARN via STS on startup. The underlying credentials for the STS call come from the ambient AWS credential chain — EC2/ECS instance profile, EKS pod identity, an ~/.aws/credentials named profile, or the AWS_PROFILE environment variable. No static AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY are used or expected.

3. Create the DynamoDB table

Create a table named LibraryWorkshop (or whatever you set DYNAMODB_TABLE to) with:

• Partition key: pk (String)
• Sort key: sk (String)
• Billing mode: On-demand (PAY_PER_REQUEST)
• TTL attribute: expiresAt
• Global Secondary Index:
  • Name: cohort-lastSeen-index
  • Partition key: cohort (String)
  • Sort key: lastSeen (String)
  • Projection: All

You can create this via the AWS Console or AWS CLI:

aws dynamodb create-table \
  --table-name LibraryWorkshop \
  --attribute-definitions \
    AttributeName=pk,AttributeType=S \
    AttributeName=sk,AttributeType=S \
    AttributeName=cohort,AttributeType=S \
    AttributeName=lastSeen,AttributeType=S \
  --key-schema \
    AttributeName=pk,KeyType=HASH \
    AttributeName=sk,KeyType=RANGE \
  --billing-mode PAY_PER_REQUEST \
  --global-secondary-indexes '[
    {
      "IndexName": "cohort-lastSeen-index",
      "KeySchema": [
        {"AttributeName":"cohort","KeyType":"HASH"},
        {"AttributeName":"lastSeen","KeyType":"RANGE"}
      ],
      "Projection": {"ProjectionType":"ALL"}
    }
  ]'

4. Enable TTL

aws dynamodb update-time-to-live \
  --table-name LibraryWorkshop \
  --time-to-live-specification "Enabled=true, AttributeName=expiresAt"

---

Development

npm run dev          # Start dev server at http://localhost:5173
npm run check        # TypeScript + Svelte type check
npm run build        # Build for production
npm run preview      # Preview production build

Agent-Led Delivery

The repo includes an installable workshop Plugin at plugins/library-ai-workshop-facilitator/. It bundles four Skills:

• facilitate-library-ai-workshop coaches one learner through the curriculum;
• run-library-ai-workshop-cohort helps a human instructor prepare, teach, and debrief a live session;
• practice-library-reference-interview role-plays a fictional patron and gives a non-scored debrief;
• review-ai-research-output audits AI-assisted research work against evidence and release checks.

The Plugin includes the course materials, simulated data, practice scenarios, and review rubric it needs at runtime.

After changing course content or FACILITATOR.md, refresh the bundled references:

npm run sync:facilitator-plugin

The repo-local marketplace entry is .agents/plugins/marketplace.json. See FACILITATOR.md for the agent teaching protocol, validation commands, installation steps, and test scenarios.

---

Distribution

Two separate things ship from this repository, and the bonus module deliberately depends on neither.

The workshop materials release. .github/workflows/release-materials.yml packages src/content/library-context/ as library-context.zip on the workshop-materials release whenever that folder changes. It contains the standing brief, the simulated data used by Modules 1–4, and two companion documents for the bonus module: SKILL-MARKETPLACE-PROMPTS.md, every prompt in that module ready to paste into Cowork, and SKILL-MARKETPLACE-HANDOUT.md, its appraisal crosswalk, vetting checklist, and local policy template. Both are our own material under MPL-2.0, and between them a participant needs no terminal at any point.

The facilitator plugin. plugins/library-ai-workshop-facilitator/, installed from the repo-local marketplace at .agents/plugins/marketplace.json, carrying its own copy of the course.

Kanon and the Context Bazaar marketplace are not bundled into either, by design. Module 5 works against the upstream jhu-sheridan-libraries/agentic-skill-library at whatever version a learner finds there. Vendoring a copy into our release would mean redistributing third-party code under a different licence (BSL-1.0 plugin, MIT repository) inside an MPL-2.0 package, pinning a snapshot that immediately begins to drift, and implicitly vouching for software that Exercise 3 exists to teach learners to vet for themselves.

Ways to install

Module 5 depends on the plugin being installed. Where the GUI route is unavailable, manual installation is a first-class upstream path, not a workaround. In rough order of effort:

1. Add the marketplace in Cowork. Customize → Plugins → Browse plugins → Add marketplace, entering jhu-sheridan-libraries/agentic-skill-library, then install context-bazaar. No terminal. This is the route the module assumes.
2. Copy a single skill. kanon/skills/<name>/ in the upstream repository is a plain SKILL.md plus a references/ folder. Copy the directory into .claude/skills/ in a project, or ~/.claude/skills/ for personal use. This is all that is needed for the kanon skill itself, which is what Module 5 leans on most.
3. Install a pinned artifact. kanon install <artifact> --harness <harness> --from-release <tag> pulls from an upstream tagged release. Upstream publishes per-harness dist-<harness>.tar.gz assets and a release manifest, so an install can be pinned and reviewed rather than tracking main.
4. Clone and build locally. git clone, then bun run dev build --harness <harness> and bun run dev install <artifact> --harness <harness> --source ..

Whichever route a library takes, the review in Module 5's Exercise 3 should happen before the install, not after.

---

Facilitator Access

The facilitator dashboard is at:

http://[your-url]/facilitator?token=<FACILITATOR_TOKEN>

The token is checked against the FACILITATOR_TOKEN environment variable. The dashboard refreshes every 30 seconds.

---

Production Deployment

This app uses @sveltejs/adapter-node. Build and run:

npm run build
node build/index.js

Set environment variables in your deployment environment (not in .env).

For a workshop, a simple option is to run the app on a local machine on the same network as participants. Participants still need internet access for their AI tool and current-source exercises.

---

Adding Content

To add a new module, create a directory under src/content/modules/<id>/ with:

• module.md — module metadata in frontmatter + overview body
• 01-<name>.md through N-<name>.md — exercise files

No code changes needed. The content loader discovers modules automatically at server startup.

Exercise frontmatter schema is documented in src/lib/content/types.ts.

---

Teardown

After the workshop, delete the DynamoDB table (all data has a 48-hour TTL anyway):

aws dynamodb delete-table --table-name LibraryWorkshop

---

License

Copyright (c) 2026 Steven J. Miklovic. Licensed under the Mozilla Public License 2.0.

---

Project Structure

src/
├── lib/db/           # DynamoDB client and queries
├── lib/content/      # Markdown loader and TypeScript types
├── lib/components/   # Svelte components
├── content/
│   ├── library-context/   # WORKSPACE-BRIEF.md + simulated sample data
│   └── modules/           # Workshop exercise markdown files
└── routes/           # SvelteKit pages and API endpoints