read-publication

name

read-publication

description

Use when the user gives you a PDF path, arXiv ID (`2401.12345` or `arXiv:2401.12345`), INSPIRE record id (`recid:1234567` or `inspire-hep:1234567`), DOI (`10.1103/...`), or HEP paper URL and wants extracted content — abstract, measured cross-section / branching ratio / mass, conditions, cuts, fit method. Routes to the right retrieval (`pdftotext` for local PDFs; INSPIRE + arXiv APIs via `WebFetch` for IDs and URLs). Always cites per AGENTS.md rule 5. Does NOT cover PDG canonical particle constants (use `pdg-lookup`), HEPData numerical tables for re-fitting (use `hepdata`), CERN service operator docs (use `cern-docs`), or Open Data dataset records (use `cern-opendata` or `atlas-opendata`). Disambiguator phrase: published-HEP-result extractor.

read-publication — extract published HEP results from a paper

This skill turns a paper reference (PDF, arXiv ID, INSPIRE record, DOI, URL) into extracted content the user can use — abstract, measured values, experimental conditions — with a citation per AGENTS.md rule 5.

Scope

Load this skill when the user provides any of:

• A local PDF path (typically EOS, AFS, lxplus home, or SWAN workspace)
• An arXiv ID (2401.12345, with or without arXiv: prefix; old-style hep-ex/0501001 also valid)
• An INSPIRE record id (recid:1234567, inspire-hep:1234567, or a inspirehep.net/literature/1234567 URL)
• A DOI (10.1103/PhysRevLett.130.041801, with or without https://doi.org/)
• A journal or preprint URL (link.aps.org/doi/..., arxiv.org/abs/..., link.springer.com/article/..., etc.)

Do NOT load this skill for:

• PDG canonical constants (electron mass, Z width, B branching ratios from PDG average) → pdg-lookup.
• Numerical tables for re-fitting (HEPData-attached YAML / ROOT / CSV) → hepdata.
• CERN service operator docs (SWAN, HTCondor, Athena how-tos) → cern-docs.
• Open Data primary dataset records (CMS / ATLAS / LHCb releases on the portal) → cern-opendata or atlas-opendata.

Input-type detection

Before retrieving, identify which input you have:

Input shape	Type	Retrieval path
Path starts with / and ends .pdf	Local PDF	pdftotext via Bash (per AGENTS.md PDF guideline)
\d{4}\.\d{4,5} or arXiv:...	arXiv ID	INSPIRE API by arxiv eprint, then arXiv abstract API
recid:\d+ / inspire-hep:\d+ / inspirehep.net/literature/\d+	INSPIRE recid	INSPIRE API by id
10\.\d{4,}/... (DOI)	DOI	INSPIRE API by doi query
https?://... (other)	Paper URL	WebFetch directly, then INSPIRE lookup if a metadata anchor is needed

If multiple inputs are given (e.g. arXiv ID + a path to a local PDF copy), prefer the local PDF for content + INSPIRE for metadata + citation.

Retrieval recipes

Local PDF

Follow the AGENTS.md PDF text extraction guideline (pdftotext first, pypdf fallback). Capture stdout into your reasoning context. Do not paste the full extracted text into the user reply unless asked — extract the requested content (abstract, table value, conditions) and cite the file path.

arXiv ID

1. INSPIRE: WebFetch https://inspirehep.net/api/literature?q=arxiv:<id>&fields=titles,authors,abstracts,arxiv_eprints,dois,inspire_id
2. If the user wants the abstract, INSPIRE's abstracts[].value is enough.
3. If the user wants a measured value or table, fetch the arXiv PDF: WebFetch https://arxiv.org/pdf/<id> — but note that WebFetch returns markdown-rendered text, not always reliable for tables. For tables, ask the user whether to download the PDF locally and run pdftotext.

INSPIRE record id

WebFetch https://inspirehep.net/api/literature/<recid>?fields=titles,authors,abstracts,dois,arxiv_eprints,publication_info

DOI

1. INSPIRE: WebFetch https://inspirehep.net/api/literature?q=doi:<doi>&fields=...
2. If INSPIRE has no record (rare for HEP), WebFetch https://doi.org/<doi> directly.

Paper URL

WebFetch <url> for content. Resolve to INSPIRE recid in parallel for citation: WebFetch https://inspirehep.net/api/literature?q=url:<url>.

HEP extraction heuristics

When the user asks "what cross-section did they measure":

• Search the extracted text for Table headings — published cross-sections are almost always tabulated.
• Search for fiducial / total / differential cross-section keywords plus ± stat ± syst patterns.
• Search for √s = N TeV or at √s = to confirm the centre-of-mass energy.
• Search for the analysis luminosity (L = N fb⁻¹ or integrated luminosity of N fb⁻¹).
• If the paper has a HEPData submission attached (search INSPIRE response for external_system_identifiers with schema: HEPData), tell the user HEPData has the table programmatically and offer to switch to the hepdata skill.

Citation

Per AGENTS.md rule 5, every quoted value must come with a source. Format:

The fiducial cross-section is σ_fid = 2.85 ± 0.21 pb (Table 4), measured at √s = 13 TeV with 139 fb⁻¹. [arXiv:2401.12345 / INSPIRE recid 1234567 / DOI 10.1103/...]

Pick whichever identifier the user provided as primary; include the INSPIRE recid as secondary if available.

Failure recovery

• INSPIRE returns 0 hits for an arxiv id → arXiv abstract API directly, warn the user that INSPIRE has no record (uncommon for HEP, may be a preprint not yet ingested).
• pdftotext fails on a local PDF → fall through per AGENTS.md guideline.
• DOI 404s → likely the DOI is mis-typed; show the user what you tried.

Never invent a value. If extraction fails, report the failure and ask the user to provide the value or a different input.