A Claude Code / agent skill that finds the real core people of any field — by mapping its endorsement graph, not by follower count.
Generalized from a GitHub-only tool (osint-field-graph) into a five-stage,
multi-platform method. The graph algorithm is one thing; the platform it runs on
is a swappable provider.
in-degree (followers/citations) = fame · cross-source = peer recognition in the niche
Method adapted from SimmerChan, 用 Agent 自动扒了一遍知乎推荐系统领域的大V (zhuanlan.zhihu.com/p/2052179994911171124).
The method (see SKILL.md)
- Define the field — and its disqualifier (the domain-filter yardstick).
- Route to a platform — where does this field leave a public endorsement
graph? → picks the provider. See
PLATFORMS.md. - Seed the top ~50 —
seed_bootstrap.sh(GitHub topic / OpenAlex field), or a platform 榜单 via the web-access skill. - Expand + rank —
field_graph.sh: forward BFS + cross-source ranking, then the manual domain filter. - Cross-verify by persona —
cross_verify.sh+SOURCES.md.
# GitHub (code fields) — bootstrap seeds, then rank
scripts/seed_bootstrap.sh --provider github --topic osint --n 40 --contributors > seeds.txt
scripts/field_graph.sh @seeds.txt --rounds 3 --promote 12 --top 40 --enrich
# OpenAlex (research fields)
scripts/seed_bootstrap.sh --provider openalex --field "natural language processing" --mailto you@x.com > seeds.txt
scripts/field_graph.sh @seeds.txt --provider openalex --mailto you@x.com --top 30 --enrich
# Any other platform: script one hop with cmd:, or harvest via web-access then --edges
scripts/field_graph.sh @seeds.txt --provider 'cmd:<command printing node {}\047s neighbors>'
scripts/field_graph.sh myfield --edges harvested_follows.tsv --top 40| provider | edge = | needs |
|---|---|---|
github (default) |
who a user follows | gh (authenticated) |
openalex |
a researcher's co-authors | curl, jq |
cmd:TPL |
stdout of TPL with {} = node |
anything scriptable |
--edges FILE |
pre-collected seed<TAB>neighbor TSV |
web-access harvest |
No Python, no runtime. Pure shell — coreutils + the chosen provider's tool.
Most tools rank by fame/volume (followers, engagement, contributions — 千瓜, 蝉妈妈, OSS Insight) or match a profile to a keyword/description (Juicebox / PeopleGPT, Exa Websets, Clay). This ranks by cross-source endorsement to surface the peer-recognized core — who is often not the most famous. The only method-cousins are academic citation-graph tools (ResearchRabbit, Connected Papers), and those only cover academia; this runs the same idea on any field.
Maps publicly-active contributors from public profiles, public follow/citation
graphs, and signed work. Not a deanonymization tool: public accounts as seeds
(never a phone/private email), respects deliberate anonymity, attaches no private
attributes. Read SKILL.md before using.
MIT