Skip to content
agent-skills-toolkit

The family registry (a dated collection report)

Reference: the family registry

Section titled “Reference: the family registry”

This tier reports structural conformance to a written Standard - deterministic and reproducible; it is not a content review, a safety audit, or a statement that the skills work. See what a tier does not certify.

This is the first artifact this project has published that grades a whole portfolio at once rather than one plugin at a time. It is produced by marketplace scope, which is deterministic: every number below comes from running one command, and running it again on the same commits produces the same numbers.

The live measurement is generated; this page is what it means. Since cut 2 the numbers are re-measured on every deploy and published at the generated registry, whose measurement date IS its deploy date and therefore cannot drift from its own numbers. Read that page for the current verdict, and this one for what the verdict means: what the columns are, why the collection reads as it does, what would turn it green, and the record of the episode below in which an earlier hand-run version of this page carried claims that were wrong. None of that is derivable from a run, so no generator writes it and none can overwrite it.

The table further down is the last HAND-RUN measurement, kept as the record of the regeneration that ended the staleness episode. It is not refreshed automatically, it does not report what installers receive, and it does not certify anything. The reproduction command is given so you never have to take either page’s word for any of it.

What the columns mean, for a non-engineer

Section titled “What the columns mean, for a non-engineer”
  • Declares is the quality tier the plugin’s own manifest claims: Bronze (Universal), Silver (Convergent), or Gold (Advanced).
  • Earns is the tier the deterministic checks actually award it.
  • A member is judged against its own claim, never against its siblings’. Declaring Bronze and earning Bronze is a pass. Declaring Gold and earning Silver is a failure, for that member, of the promise it made.
  • Pin is the commit the catalogue advertises. Graded is the commit on this machine that was actually read. When they differ, the catalogue is simply between releases, which is normal; the columns are shown either way so agreement is never assumed from silence.
  • Standard debt counts the checks that are only warnings because the member pins an older version of the Standard. They become failures the moment it adopts the current one.

The snapshot

Section titled “The snapshot”

Measured 2026-09-01, against the registry fetched fresh at 81dbbde (catalogue v1.72.0), with every member checked out at the sha the catalogue pins:

Terminal window
# one checkout per member, each at the sha marketplace.json pins, under <members>/
npx agent-skills-toolkit evaluate ../agent-plugins --members <members> --format md

Collection verdict: RED. Graded 6 of 6 members. 0 collection errors, 0 collection warnings. Two members fail their own declared claim.

Member Status Declares Earns Errors Warns Standard debt Entry version Pin Graded sha Divergence
pm-skills FAILS OWN CLAIM none declared none 234 40 0 2.32.0 e8a641c e8a641c in sync
thinking-framework-skills FAILS OWN CLAIM Advanced Convergent 1 137 130 0.13.0 9aab9f3 9aab9f3 in sync
writing-style-catalog OK Universal Universal 0 3 0 0.13.0 00e4884 00e4884 in sync
agent-skills-toolkit OK Advanced Advanced 0 0 0 1.17.0 fd5286b fd5286b in sync
critique-skills OK Convergent Convergent 0 7 7 0.1.6 30ec617 30ec617 in sync
product-lifecycle-templates OK Advanced Advanced 0 1 1 0.4.0 e501e04 e501e04 in sync

Tier distribution across graded members: Advanced 2, Convergent 2, Universal 1, none 1.

Every row reads in sync, and that is new. Earlier snapshots of this page graded whatever local checkout happened to be on the machine, so three of six rows read diverged and the numbers described trees the catalogue does not pin. This run checks each member out AT its pinned sha first, so Pin and Graded sha agree by construction and the verdicts are reproducible by anyone from the two commands above.

What was wrong with the previous snapshot, and for how long

Section titled “What was wrong with the previous snapshot, and for how long”

This page carried a 2026-08-12 measurement until 2026-09-01 - twenty days. Two things in it were wrong, and they are different kinds of wrong:

  • It described a toolkit six releases old. It listed agent-skills-toolkit at v1.11.1 / f57aa3f. The catalogue now pins v1.17.0 / fd5286b.
  • It graded drifted checkouts rather than the pins. Three rows showed diverged, meaning the numbers beside them came from trees the catalogue does not pin. A reader had no way to reproduce them.

A correction to the record, since a repaired page should say what was actually true. The 2026-08-28 internal audit recorded this page as carrying “a false FAILS OWN CLAIM for a member fixed on 2026-08-15”. Checked at regeneration time, that is not right, and the difference matters. thinking-framework-skills did fix its workflow declarations in fd343dd on 2026-08-15 - but the catalogue still pins 9aab9f3, which is v0.13.0 from 2026-06-25 and is not a descendant of that fix. Graded at the sha the catalogue pins, the member genuinely fails its own claim. What is stale is the registry PIN, not this page’s verdict. The remedy is a re-pin in the catalogue, which is that repository’s call, not a correction here.

Why the collection is red, said plainly

Section titled “Why the collection is red, said plainly”

It is red on purpose, and the two reasons are different.

thinking-framework-skills declares Advanced (Gold) and earns Convergent (Silver) at its pinned sha. Its single error is G4 (index drift), and the cause is this toolkit’s own change: v1.10.0 fixed the index generator, which puts any consuming plugin whose committed INDEX.md predates that fix into drift until it regenerates. Its 130 findings of Standard debt come from pinning Standard 0.8: the components mirror and its workflow half are held at warn for it and become errors when it adopts a current pin. Both are migration consequences of shipped fixes, surfaced in one place instead of six separate runs. The findings report for that member is filed separately; this page does not edit anyone else’s repository.

pm-skills declares no tier at all and carries 234 errors. An undeclared plugin is graded by the same default the gate applies to any undeclared plugin, with no tier ceiling, so every error counts. It is not being held to a bar it never claimed; it is being shown what it currently is.

The remaining four members satisfy their own claims outright, including two carrying Standard debt (critique-skills 7, product-lifecycle-templates 1) - findings held below their severity by an older pin, which is exactly what makes green-by-an-old-pin visible rather than flattering.

What would turn it green

Section titled “What would turn it green”

Nothing in this toolkit. The collection turns green when each member satisfies its own declared claim - either by clearing its findings, or by declaring the tier it actually earns. Both are that repository’s maintainer’s call. For thinking-framework-skills there is a third path that belongs to neither: the catalogue re-pinning it to a commit at or after its 2026-08-15 fix. Tuning the aggregation rule until the number looks better is the one repair the design explicitly forbids: a threshold a maintainer can move is not a gate.

Section titled “Related”