fix(skills): browse shows full catalog, not first 5000 (#41413)

hermes skills browse capped the hermes-index source at 5000, so it
surfaced ~5.4k of the ~90.7k skills the index actually carries. Raise
the per-source ceiling above catalog size; browse already paginates
client-side and the index is disk-cached, so no extra fetch cost.
This commit is contained in:
Teknium 2026-06-07 10:15:31 -07:00 committed by GitHub
parent 16786f3bb3
commit 1892e22acb
No known key found for this signature in database
GPG key ID: B5690EEEBB952194

View file

@ -337,12 +337,15 @@ def do_browse(page: int = 1, page_size: int = 20, source: str = "all",
_TRUST_RANK = {"builtin": 3, "trusted": 2, "community": 1}
# NOTE: when the centralized index is available, parallel_search_sources
# skips the external API sources and serves everything from "hermes-index".
# That source MUST therefore carry a high limit, or browse silently caps
# the entire hub at the default (50) — it shipped that way and surfaced
# ~136 of 88k skills. The external-source limits below only apply when the
# index is unavailable (offline / first run before the cache populates).
# That source MUST therefore carry a limit large enough to cover the whole
# catalog, or browse silently caps the hub — it shipped at 50 (surfaced
# ~136 of 88k skills), then 5000 (surfaced ~5.4k of 90k). The index is
# disk-cached and browse paginates client-side, so a ceiling above the
# current catalog size is the right call. The external-source limits below
# only apply when the index is unavailable (offline / first run before the
# cache populates).
_PER_SOURCE_LIMIT = {
"hermes-index": 5000,
"hermes-index": 1000000,
"official": 200, "skills-sh": 200, "well-known": 50,
"github": 200, "clawhub": 500, "claude-marketplace": 100,
"lobehub": 500, "browse-sh": 500,