James opened his TutorClaw project from Module 9.3 and looked at the shim skill file. About fifty lines of tool declarations and a pointer to the MCP server. Then he looked at the server code: nine tools, pedagogical engine, assessment logic, code execution sandbox, and tier gating.
"If I improve the assessment logic," he said, "I change the server, push the update, and every learner gets it instantly. They wouldn't even know it happened."
Emma nodded. "But if you add a tenth tool, the learner's shim wouldn't know about it. Their agent wouldn't discover it until the shim was updated. That distinction is the entire versioning strategy."
James leaned back. "It's like parts catalogs at the warehouse. Most of the parts stay the same during a motor upgrade. The catalog only changes if the mounting brackets change. The motor is the intelligence; the shim is the catalog."
You shipped a product in Module 9.3. Now you face the question: what happens when you need to change it?
TutorClaw's three-component architecture defines its maintenance model:
The intelligence lives on your side. The shim is a thin connection layer. When the intelligence improves, only the server changes.
Most improvements you make are server-side changes that propagate instantly:
This is the 95% case. No update notifications. No version compatibility issues. No disruption. The server's internal logic can be rewritten from scratch, and the learner's experience simply improves.
Some changes alter the interface between the shim and the server. This requires a distributed update:
These change the Contract, not the Implementation.
Semantic versioning (MAJOR.MINOR.PATCH) applies strictly to the shim skill.
Asymmetry: The server might receive 50 improvements while the shim stays at 0.1.1. The user only cares about the shim version.
How do users know the product is better if they don't see an "Update" button?
James looked at his list. Ten of his twelve planned improvements were server-side. "Only two need shim updates. And even then, the old shim still works. They just miss out on the new tool."
Emma nodded. "The less the client knows, the less can go stale. That's the advantage of a thin client."
"But distribution is the next challenge," she added. "How do they find the catalog in the first place?"