Case study companion · Creating AI agents for course production

From storyboard to Canvas

Explore how I used five Claude Agent Skills to support course production, from checking the course setup to reviewing the completed build. This demonstration uses fictional content to illustrate the workflow.

Fictional module: FCA_DA_101 · Module 3 — Communicating with data
Stage 1 of 5

Setup review

Before any content is built, an agent reviews the Canvas sub-account and course shell against the programme’s requirements, so problems are caught while they are still cheap to fix.

Where judgement staysThe agent flags deviations; a person decides which are genuine problems and which are intentional, like an extended navigation on an orientation course.
Input — the course setup
FCA_DA_101 · course configuration
Theme
FCA_DA CSS v2.3 · JS v2.3
Course navigation
11 items visible
Grading scheme
FCA standard (pass / not yet)
Term dates
2 Nov 2026 – 29 May 2027
Module locks
Not set
Template pages
6 pages imported
The task

The setup-review agent reads the live configuration and the template pages, and compares both against the programme’s documented standards: the approved theme version, the six-item navigation, the grading scheme, and the markup the template pages should use.

It changes nothing — the output is a findings list, ordered by how expensive each issue becomes if it reaches the build stage.

Sample output — the review
  • Theme current. CSS and JS both v2.3 — matches the programme standard.
  • Grading scheme matches. FCA standard scheme applied at course level.
  • Two heading styles in use. Three template pages carry manually-styled H2s instead of the themed class — these would propagate into every built page.
  • Navigation shows 11 items; standard is 6. Files, Outcomes and three tool links are visible to learners.
  • Module locks not set. All modules currently open; the programme releases Module 1 only at launch.