hep-rumor-mill
Skill4 starsChanged 20 days agoMITView on GitHub
Analyze the HEP-theory postdoc rumor mill (the public Google Sheet at sites.google.com/site/postdocrumor) to see who got offers at which institutions, then pull each offer-holder's publication record from InspireHEP (and OpenAlex + Semantic Scholar for interdisciplinary people) to study what kind of profile lands where. Use when the user wants to study the HEP-theory / cosmology / hep-ph job market from the rumor mill: institute-by- institute cohort profiles (citations, papers, venues, subfield, PhD-age, named fellowships), benchmark their own InspireHEP profile against the accepted cohort, or generate a Korean job-market strategy report. Triggers on: rumor mill, postdoc rumor, HEP rumor mill, hep-th job market, theory postdoc offers, who got offers, offer holders, 루머밀, 포닥 루머, 잡마켓 분석, 오퍼 받은 사람, 합격자 실적 분석, 포닥 전략, InspireHEP 프로필 분석.
- Reads credentials
In this repository38 other files it gives its agents.
- academic-cvSkill
- academic-jobsSkill
- academic-proseSkill
- adversarial-reviewSkill
- bibtex-genSkill
- chrome-sessionSkill
- clickup-cliSkill
- clickup-todaySkill
- commit-triageSkill
- concept-explainerSkill
- contact-emailSkill
- cover-letterSkill
- dropboxSkill
- handdrawn-schematicSkill
- journal-club-reviewSkill
- magi-researchSkill
- md2pdf-typoraSkill
- morgenSkill
- overleaf-section-workflowSkill
- overleapSkill
- peer-agent-reviewSkill
- proton-mailSkill
- reference-searchSkill
- remarkable-syncSkill
- research-backupSkill
- research-boardSkill
- research-logSkill
- research-portalSkill
- research-proposalSkill
- research-reportSkill
- research-statementSkill
- scienceplot-pySkill
- vastaiSkill
- wide-slide-illustratorSkill
- workshop-paper-reviewSkill
- xkcd-pySkill
- zai-web-searchSkill
- zoom-summarySkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 26
Self benchmarking: place the user's own InspireHEP profile on the accepted cohort and surface gaps, overall and per target institute.
- Very specific: line 60
Every command takes --json for machine-readable output.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
HEP Rumor Mill Skill
Conversational interface over the prm CLI. It pulls the HEP-theory postdoc rumor mill (a
public Google Sheet) for a year, resolves each offer-holder's InspireHEP profile, and analyzes
their publication records so the user can study the job market and benchmark their own profile.
Two first-class uses:
- Institute intelligence: for an institution, what do the people it gave offers to look like (citations, papers, venues, subfield mix, PhD-age, named fellowships)?
- Self benchmarking: place the user's own InspireHEP profile on the accepted cohort and surface gaps, overall and per target institute.
Data is layered across three public, no-auth sources, keyed off the InspireHEP author id that the rumor sheet links for each person:
- InspireHEP: primary record (HEP papers, journals, citations, PhD year, advisors, ORCID).
- OpenAlex: cross-disciplinary augmentation for everyone with an ORCID, so ML / stats / CS venues (ICML, NeurIPS, journals) that InspireHEP misses are caught.
- ORCID public API: the author's own self-claimed work list, used to rescue people whose OpenAlex-by-ORCID cluster is contaminated by author-conflation (see below). Author-curated, so no conflation; citations are backfilled per DOI via OpenAlex.
- Semantic Scholar: name-based fallback (and h-index cross-check) when a person has no ORCID.
Read this before reporting: the data is self-reported and incomplete
The rumor mill is filled in by volunteers. Offers that were never reported are simply absent,
per-institute counts are usually single digits, and the same person appears multiple times
(one row per institution, and Offered then Accepted/Declined). Treat the output as a
qualitative signal, not a statistic, and always state N and how many were resolved. The
analysis is descriptive: it describes what kind of profile landed where; it never predicts
whether a given profile will or will not get an offer. See references/analysis.md.
Quick Reference
| Intent | Command | Reference |
| Pull a year's rumor sheet | prm fetch --year YEAR [--sheet-id ID] | references/sheets.md |
| Resolve offer-holders' records | prm enrich [--year Y] [--status Accepted] [--limit N] [--no-cross-disc] | references/schema.md |
| Show one person's record | prm profile {recid | --name NAME} | references/metrics.md |
| Institute cohort profile | prm institute "NAME" [--status Accepted] [--year Y] | references/metrics.md |
| Benchmark your own profile | prm analyze --me RECID [--institutes "A,B,C"] [--year Y] [--refresh] | references/analysis.md |
| Emit Korean report skeleton | prm report --year Y [--me RECID] [--out PATH] | references/analysis.md |
Every command takes --json for machine-readable output. Run via uv run prm ... from the
skill directory. The local store lives at ~/.local/share/hep-rumor-mill/rumor.db
(override with PRM_DATA_DIR).
Typical workflow
prm fetch --year 2026: download the sheet, tag each institution's country/region, store entries. Reports total rows, unique people, institutions, and theOffered/Accepted/Declinedbreakdown.prm enrich --year 2026 --status Accepted: resolve each accepted person across sources and compute metrics. Network-heavy and polite; it is resumable (--limit Ncaps a run, re-run to continue) and reports failures and any unresolved people. Start with--status Acceptedsince that is the cohort you usually benchmark against.- Explore:
prm institute "Perimeter Institute" --status Accepted,prm profile 1804701,prm analyze --me {your recid} --institutes "Perimeter Institute,CERN,DESY". prm report --year 2026 --me {your recid}: emit a Korean markdown skeleton with the data-backed tables filled and clearly-marked blanks for the qualitative judgment, which you complete by reading the offer-holders' actual papers (seereferences/analysis.md).
Data-quality flags surfaced by the tool
These are computed automatically; surface them when you report, do not hide them.
n_large_collab: InspireHEP papers with > 50 authors. Big experimental / collaboration papers inflate a person's paper and citation counts; a high value means the headline numbers do not reflect lead-author output. Pair it withn_first_author.- OpenAlex author-conflation: OpenAlex sometimes merges several same-named people (and a
shared ORCID) into one author entity, pulling in unrelated disciplines (crystallography,
neuroscience, economics).
enrichdrops clearly distant-field works, and when contamination is heavy it discards the whole OpenAlex cluster and rescues the real list from the ORCID public API instead (author-claimed, conflation-free), saying so in the run notes. interdisciplinary: set when the author's InspireHEParxiv_categoriesinclude a non-physics category (cs, eess, stat, q-bio, econ, q-fin), or when a material share (>= 15%, >= 2 papers) of the record is in adjacent fields (CS / ML / stats / data science) or ML venues. Mathematics and mathematical physics are excluded on purpose (normal for formal hep-th). Physics-family fields, including OpenAlex's coarse "Physics and Astronomy", never trigger it on their own.- Citation undercount for arXiv-only preprints: a work that exists only as an arXiv preprint
(no DOI, not in InspireHEP) has its citations backfilled by a Semantic Scholar title match,
which is a conservative floor (lower than Google Scholar); works Semantic Scholar does not
index stay at 0. So citation totals for interdisciplinary candidates with many ML preprints
still read low. Lean on the paper list and venues, not the citation total, for these people.
Without an
S2_API_KEYthe in-pipeline backfill is best-effort (Semantic Scholar rate-limits hard); set the key for reliable backfill.
Notes
- The rumor sheet is a public Google Sheet exported as CSV; no login. Year-to-sheet-id mapping
lives in
prm/sheet.pyandreferences/sheets.md. Only 2026 ships known; add older years by extracting the embedded sheet id from eachpostdocrumor/{year}-rumorspage. - Journal prestige tiers (Tier A / B / C / preprint) are an editable opinion seeded with hep-th/ph
convention. They live in
prm/metrics.pyand are documented inreferences/metrics.md. - Be considerate with the APIs:
enrichpaces InspireHEP requests and OpenAlex paging. Semantic Scholar rate-limits hard and is used for the no-ORCID author fallback and for arXiv-only citation backfill; setS2_API_KEY(the same env varreference-searchuses) to raise the limit and make the backfill reliable. Without it the pipeline still works, backfill is just best-effort. prm analyze --meuses the cached enriched record when one exists; pass--refreshto force a fresh re-enrich (otherwise a manual citation correction is preserved across runs).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

