Coverage

Which repositories were searched for a skill, in which task classes, and what was found. A blank cell would say nothing; every cell here says which of five things happened.

16 repositories by 7 task classes. Rendered from the selection record and the run.
RepositorySkill authoringAccessibility auditSpec writingOn-page auditVoiceCode reviewTest writing
obra/superpowersPinned, not runNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillDeclared empty
affaan-m/ECCNo qualifying skillTestedTestedTestedTestedNo qualifying skillDeclared empty
anthropics/skillsTestedNo qualifying skillPinned, not runNo qualifying skillNo qualifying skillNo qualifying skillDeclared empty
rampstackco/claude-skills operator of this siteTestedTestedTestedTestedTestedPinned, not runNo qualifying skill
wshobson/agentsNo qualifying skillPinned, not runNo qualifying skillNo qualifying skillNo qualifying skillPinned, not runNot verified
alirezarezvani/claude-skillsNo qualifying skillPinned, not runPinned, not runPinned, not runNo qualifying skillPinned, not runNot verified
phuryn/pm-skillsNo qualifying skillNo qualifying skillPinned, not runNo qualifying skillNo qualifying skillNo qualifying skillNot verified
samber/cc-skillsNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillPinned, not runNo qualifying skillNot verified
openclaw/openclawPinned, not runNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNot verified
addyosmani/agent-skillsNo qualifying skillNo qualifying skillPinned, not runNo qualifying skillNo qualifying skillPinned, not runNot verified
mattpocock/skillsNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillPinned, not runNot verified
Jeffallan/claude-skillsNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillPinned, not runNot verified
github/awesome-copilotNo qualifying skillNo qualifying skillPinned, not runNo qualifying skillNo qualifying skillNo qualifying skillNot verified
EveryInc/compound-engineering-pluginNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillPinned, not runNot verified
open-gsd/gsd-coreNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillPinned, not runNot verified
yusufkaraaslan/Skill_SeekersPinned, not runNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNo qualifying skillNot verified

What each cell means

Tested
A skill from this repository is pinned in this class and has been run, with the skill loaded and without it, on the same task pairs.
Pinned, not run
A skill from this repository qualifies in this class under the published selection rule, and its bytes are committed here, but no model has been run against it. Either it holds a slot the run has not reached, or it qualified and lost the output-format tiebreak while the cohort still capped a class at one external skill.
No qualifying skill
The selection rule was run over this repository for this class and found nothing that qualifies. The candidates it weighed, and the sentence each was weighed on, are in the cohort record.
Declared empty
This class has declared its external slot empty and said so. The rule has not been run over this repository for it, and no candidate has been weighed.
Not verified
Nothing has been recorded either way. The rule has not been run over this repository for this class, and this site is not claiming it has.

Where the rows come from

THE REPOSITORIES THE COHORT DRAWS FROM, AND WHERE THIS LIST CAME FROM. Replaced 2026-08-31 by the second recon pass, which IS committed to this repository at docs/recon/recon-skill-source-expansion-2026-08-31.md. The previous note said this list was not the output of a recon pass because no recon pass was committed; that is no longer true, and the replacement carries the discovery routes with it exactly as that note said it would. The list is no longer a bare array of names: each entry records when it entered the cohort, which route surfaced it, the HEAD SHA observed during the recon, and where a licence was read. THE HEAD SHA IS PROVENANCE FOR THE SEARCH AND IS NOT A PIN, which is a distinction this record keeps everywhere: a pin names the commit an arm's bytes came from and lives on the slot. The gate that reads this list is unchanged and still refuses an entry that appears in no pin, no selection candidate and not in the selection rule, so a repository still cannot be added by typing it here. `agentskills/agentskills` remains deliberately ABSENT: it is Key A, the document the skill-authoring class is scored against, and a key is not a source repository.

How these repositories were found, and what the routes missed

THE ENUMERATION BEHIND THIS FIELD IS A SAMPLE, NOT A CENSUS, AND THE ROUTES SAY SO THEMSELVES. Two independent discovery routes were run, a GitHub topic search over five topics ranked by stars and a code search for `filename:SKILL.md path:skills` resolved and ranked client-side, and NEITHER returned `obra/superpowers`, which is the second-largest catalog in the pool at 279851 stars and a repo the published selection rule already names. It had to be added by hand. A route set that misses a repo the cohort already draws from cannot be assumed to have found the ones it does not draw from, so the 52 repositories the recon verified are a sample of the field, and the absence of a mechanism from this list is absence of evidence rather than evidence of absence. THE THIRD ROUTE FOR THE NEXT PASS is the plugin marketplace manifests, `.claude-plugin/`: both routes here key on repository topics and on a `skills/` path convention, and a repo that publishes its skills through a marketplace manifest satisfies neither. Following those manifests is what would have found `obra/superpowers` without a hand addition, and it is the route to add before the next expansion is read as complete.

Amended 2026-08-31 from the recon report committed at docs/recon/recon-skill-source-expansion-2026-08-31.md.

The cap on external skills per class, removed

Amended 2026-08-31, R23. The per-class cap on external slots is removed. No pin was added, removed or changed, and no selection was re-run.

THE CAP IS GONE AND THE FLOOR IS NOT. Until this amendment the cohort admitted exactly one external skill per class, and `loadCohort` enforced it by refusing any class whose filled and declared-empty external slots did not total one. That single comparison was the cap and the floor at once. Under R23 every skill that qualifies under the published selection rule runs in its class, and cohort size per class is a property of the field rather than a number this project chose, so the comparison is now a floor only: a class still accounts for exactly one own slot, and still accounts for AT LEAST one external slot, filled with a pin and a selection record or declared empty and said to be empty. Removing an upper bound must not become a licence to ship a class with one arm and no record of why, which is what R2 exists to prevent on a filled slot and must not be reachable by leaving one empty instead. THE CONSEQUENCE WAS ALREADY IN THIS FILE. Two candidates QUALIFIED under the rule and lost only the output-format tiebreak: obra/superpowers skills/writing-skills in skill-authoring, and anthropics/skills skills/doc-coauthoring in spec-writing. Under the cap those were rejections, because the one external slot was taken. Under this amendment they are qualifying skills whose bytes are committed and which no model has been run against, and the coverage matrix says that rather than continuing to call them rejected. THEY ARE NOT PROMOTED TO SLOTS BY THIS AMENDMENT: a slot carries a pin whose bytes an arm was given, and no arm has been run against either of them. Promoting them is its own change, with its own run, and it is gated on the licence rule recorded in the amendment above, which anthropics/skills skills/doc-coauthoring already satisfies by carrying a licence record that says it has no licence file at all. Nothing in the selection rule moved, and no pin, hash or cell is touched here.

Every cell is derived at build time from harness/claims/skill-cohort.json and from the run records, by the same modules the cohort index is built from. No state on this page is written by hand, and there is nowhere to write one.