The repo ships an agent skill at .claude/skills/authoring-scalpal-cases. Claude Code picks it up automatically when you ask for a new case, or you can call it with /authoring-scalpal-cases. It includes a command-line tool that imports the preop service’s own validators, so a package that passes check passes the same rules the app enforces.

Ask an agent

The agent follows the skill: it reads the chart, starts from a template, writes the package, runs check until it’s clean, and reports what it verified.

Or do it by hand

1

Set up

2

Pick a patient

3

Start from a template

Exports a complete, passing case as a package. The skill also bundles one at assets/example-priya/.
4

Write the package

Edit case.json (plan and encounter) and the three files in content/. See package format and writing the interview.
5

Check it

Runs the catalog validator, builds the case from the chart, validates the interview, checks chart-risk coverage, persona against chart, diagnosis leaks, voices and more. Fix every error. Read every warning.
6

Trial it

Installs the package into a throwaway git worktree, runs the whole preop test suite, and lists anything that breaks. Re-planning an existing patient often breaks a test written for their old case; update that test as part of your change.
7

Install it

8

Run the gates

Catalog validation, typecheck, the full test suite, Unity id generation, the offline bundle export, and the C# playthrough when the .NET SDK is installed.

Tool reference

The tool never calls FinchNode or any model provider. Charts come from the recorded fixtures in services/preop/test/fixtures.