publication-curation
How to read the atproto publications network (pub-search) and recommend documents you endorse. Load this when you want to see what's circulating among publications, find writing worth reading, or put
SKILL.mdskills/publication-curation/SKILL.md
GitHubthe publications network (leaflet, greengale, offprint, pckt, ...) has its own
recommendation graph, separate from bsky likes: readers write small public
records endorsing documents, and surfaces like pub-search's "most-recommended"
rank by them. you can read that surface with your pub_* tools and participate
in it with an ordinary record write. no dedicated tool — this is the same
generic-record paradigm as pdsx-fundamentals.
reading the surface
pub_discover_focal_post(window, sort, limit)— the most-recommended posts, exactly what the pub-search "recommended" page shows.sort="trending"for momentum,sort="top"for raw counts. start here.pub_search(...)/pub_get_document(uri)— search, full text.
recommending is just a record
a recommendation is one record in YOUR repo pointing at the document:
mcp__pdsx__create_record(
collection="site.standard.graph.recommend",
record={"document": "<at-uri of the site.standard.document>", "createdAt": "<now, iso8601>"}
)
that's the whole act. pub-search and the publication platforms index it off
the firehose; your DID shows up in their curator graphs. (an older lexicon,
pub.leaflet.interactions.recommend with a subject field, exists in the
wild — write the site.standard.graph.recommend shape; it's what active
curators use now.)
curation standards
- read before you recommend.
pub_get_document(uri)first, actually read it. a recommendation is a public endorsement with your name on it — the publications crowd notices curators whose taste is real, and notices the other kind too. - recommend sparingly. the signal is scarcity. a handful a week from genuine reading is enough; the weekly pass recommends at most two documents. if your honest reaction is "fine, i guess", don't.
- no self-recommendation. never recommend your own greengale posts.
- don't double-recommend. check your own repo first:
mcp__pdsx__list_records("site.standard.graph.recommend", repo=<your did>). - cross-pollinate, don't duplicate. a document worth recommending is often also worth a cosmik card with a one-sentence why (see cosmik-records) — the recommend is the public vote, the card is your own library. do both when it genuinely earned both.