Exhibit A: one skill, two copies

The thesis says skills are software packages, and that without a registry, a version and a single canonical copy, they will drift and one of them will end up wrong. Here is that happening, to our own skills, measured.

Why a word count is the least of it

A gap in length is only the symptom that is cheap to measure. The substantive failure is that the two copies disagree about something a caller depends on:

An auth claim about which header the proxy path accepts. The sg-compute pack found the '401 at the proxy' sentence in the __Send copy is wrong; the service copy does not carry it.

So an agent that loads one copy is told something the service does not do, and an agent that loads the other is not. Neither copy knows the other exists. Nothing in either repository records which one is canonical, because there is no field in which to record it, because there is no registry to hold the field.

What the thesis predicted, in order

1. A useful skill is needed in a second repository.
2. With no dependency mechanism, the only way to reuse it is to copy the file.
3. Both copies get edited, because both are now local files that work.
4. The copies drift. Nothing measures the drift, because nothing knows they are related.
5. One copy acquires a claim that is false. It stays false, because a review of one repository has no reason to look at the other.

Every step is the absence of a package manager, and every step happened. The thesis is on the next page; this is the evidence for it, and it costs the estate nothing to admit because the fix is cheap and named: one repository owns the skill, the other imports it. That is gap G1, and it is the estate's pip moment: the first dependency.