“Skills are a software package, that is the best analogy… the same way we have npm and pip for managing dependencies, we need a way to manage skills like that.” This site publishes both halves honestly: eight skills shipped, generated from their own frontmatter, and 25 briefs, 60,782 words of theory that runs a long way ahead of them — including the theory failing, live, on the estate’s own skills.
The work exists in two very different states, and every page on this site says which one it is in. S shipped and real · D designed, unbuilt · E an economic or strategic position.
| Half | Size | What it is |
|---|---|---|
| S Shipped | 8 skills · 16,338 words of SKILL.md · 2 repos | Real code, running today, with five authoring conventions nobody had written down until this site did |
| D / E Thought | 25 briefs · 60,782 words | Almost all written in one four-day burst, 1–4 June 2026 — the most complete theory of agent skills in the estate, and almost none of it built |
Scored against its own checklist — version control, dependency management, vulnerability management, code review, CI, distribution, documentation — the estate’s skills get one green cell out of seven. The full scorecard →
“Skills are code and need to be version-controlled, and need to be treated like software… the same way we have npm and pip for managing dependencies, we need a way to manage skills like that.”
Not a metaphor — a checklist: “vulnerabilities management, documentation, integration, wrappers, code reviews, deployment, CI pipelines, distribution reviews — every analogy we have with code, we need with skills.” And the sharper claim that makes it more than an analogy: skills describe intent, in English, not just capability — which is why they are proposed as the successor to code packages, not a variant of them.
The full checklist, the one-green-cell scorecard, and why versioning intent is harder than versioning capability.
Read → Design, not shipped“Today’s skills are static photographs of what they should be.” Typed primitives, and the forking ecosystem that keeps a fork reconcilable.
Read → The most consequential, least builtA skill must come with the permission set it needs to run — and the OAuth critique: it “forces you to grant a hundred percent when you need ten percent.”
Read → Who gets paidBase vaults, the branded/certified/customised cascade, and the scoring position: provenance and evals, “rather than gameable popularity.”
Read →The clearest argument for the thesis is not in the briefs. It is in the estate’s own two repositories, and it is measured on every build of this site.
use-sg-playwright exists in both skill repositories — and one copy carries an auth claim the service code contradicts. No registry, no single source, no version pin, so the copies drifted, and one of them is now wrong.
measured at build time · the full exhibit →
Every shipped skill, generated from frontmatter: trigger summary, do-NOT-use clause, and the version/permissions fields none of them declare.
Read → Usable the same dayFive conventions extracted from eight skills, plus the three governance rules from the estate’s best worked example of when not to make a skill.
Read → Published unresolvedWhat is a skill version, when the interpreter changes underneath it? How do you test intent? Two questions with no answers anywhere in the corpus.
Read → If you are an agentllms.txt, llms-full.txt, and the catalogue as JSON — because this site’s subject is the artefact type agents load, so it owes agents its own duty of self-sufficiency.
Read →