Bonus: The Skill Marketplace

About This Module

Modules 1–4 treat the AI tool as a fixed product you work inside. This bonus module inverts that: it treats the instructions an AI tool follows as acquirable, describable, reviewable objects — and treats the place they come from as a collection you are responsible for.

The worked example is Kanon, a tool from the Johns Hopkins Digital Research and Curation Center, and the artifact library it distributes — published as the Context Bazaar marketplace. Kanon's premise is author once, compile to every harness: you write one canonical knowledge artifact, and Kanon compiles it into the native format each AI coding assistant expects.

That library already contains a library-ai-workshop collection — the four Skills from this repository, imported, described, and versioned by someone else. You are about to look at your own work as a catalog record.

The library comes with its own guide

You are not expected to memorise a command-line tool. The marketplace ships a skill called kanon whose entire job is to teach you Kanon — and once it is installed, you use it by asking questions in plain language.

kanon is written for library staff, and it practises what this module preaches: a short skill file loads first, and six longer references load only when they are needed.

Ask for	You get
the authoring guide	Nine steps from idea to compiled artifact, with a metadata-quality checklist and Dublin Core examples
the command reference	Every Kanon command, with flags and worked examples
the tutorial	Twenty sequential lessons, from "what is a coding agent" through publishing; each one skippable
the self-paced course	A three-to-four hour course on skill creation, with a practice artifact, assessments, and an answer key
the curriculum guide	Learning paths, facilitation notes, assessment strategy, and accessibility considerations for library staff
Souk Compass practice	An optional session on semantic search over the catalog

This module is the 90-minute library-facing pass. Those references are where you go afterwards, and Exercises 1 and 4 hand you to them directly.

What you need before you start

This module works against a real installed plugin. Before the session you need Claude Cowork or a coding agent, and permission to install a plugin on the machine you are using. If your institution has not decided who may install what, that decision has to happen first — which is itself a preview of Exercise 5.

Two paths are supported, and all five exercises work the same way on both:

Path	You need	How you work
Cowork (default)	Claude Cowork	Install through menus, then work by asking questions in plain language. No terminal.
Coding agent	Claude Code or Codex	The same, plus a terminal if you want to compile and publish your own artifact

To install in Cowork: open Customize in the sidebar, then Plugins, then Browse plugins, then Add marketplace. Enter jhu-sheridan-libraries/agentic-skill-library — the short owner/repo form is enough. Then install the plugin named context-bazaar.

In Claude Code: run /plugin marketplace add with the repository URL, then /plugin install context-bazaar.

Two companion files sit in the workshop folder beside WORKSPACE-BRIEF.md:

• SKILL-MARKETPLACE-PROMPTS.md — every prompt in this module, ready to copy and paste into Cowork.
• SKILL-MARKETPLACE-HANDOUT.md — the three working sheets: an appraisal crosswalk, a vetting checklist, and a local policy template.

You will not need a terminal for any exercise. You will need the plugin.

Vocabulary

Term	Plain meaning	Nearest library analogue
Knowledge artifact	A packaged unit of expertise an AI tool can load	An item
Skill	The most common artifact type: standing instructions the tool follows	A style manual on the ready-reference shelf
Harness	A specific AI coding assistant (Claude Code, Codex, Kiro, Copilot, Cursor, …)	A platform or reader
Collection	A named group of related artifacts	A collection
Catalog	catalog.json, the machine-readable index of every artifact	The catalog
Marketplace	A repository advertised as installable, plus its manifest	A vendor package or consortial profile
Trust lane	Declared oversight level: official, partner, community, experimental	Provenance / authority note
Maturity	Lifecycle state: experimental, beta, stable, deprecated	Edition status; deprecated is a weeding flag

Safety baseline for this module

Installing a skill means agreeing that an AI tool will follow instructions written by someone you have not met, in files you have not read. Everything from Module 1 still applies, plus:

• Do this in a practice folder or scratch repository, never in a production system or a repository holding patron data.
• Do not connect institutional email, cloud storage, or library systems to any tool during this module.
• Treat every artifact body as untrusted text until you have read it.
• If your institution requires review before installing software, that requirement covers this. Stopping to ask is the correct answer.

Going further

Ninety minutes is a survey. When someone wants depth, the kanon skill already carries it — ask for the reference by name:

• the tutorial — twenty sequential lessons, each self-contained, running from "what is a coding agent" through catalog, import, scaffold, validate, build, install, collections, evals, publish, upgrade, and team sync.
• the self-paced course — three to four hours on skill creation, with a safe practice artifact, assessments, an answer key, and a capstone review.
• the curriculum guide — written for a library that wants to teach this: learning paths, a curriculum map, facilitation notes, assessment strategy, accessibility considerations, and a production-readiness gate.

One honest caveat about all of it. The kanon skill's setup instructions still tell you to clone agentic-skill-forge, the repository's former name. The old name redirects, so nothing breaks — but you found a documentation drift in the guide you are being asked to trust, on your first read, without looking for it. Note where you found it. Exercise 3 is about exactly this habit.