agentsclimarketplace

Paper extract

Skill lucaswychan/quant-paper-agent/quant_paper_skills/skills/paper-extract

A Claude Code plugin that turns a quantitative-finance arxiv paper into a production-ready event-driven strategy

Install
npx -y skills add lucaswychan/quant-paper-agent --skill paper-extract

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Inspect a quantitative-finance paper PDF with Codex's PDF capability and produce an evidence-backed synopsis, reported-metrics dataset, formulas, replication targets, and ambiguity report. Use when a paper.pdf exists and the user wants the contribution, methods, tables, formulas, data design, performance claims, or replication handoff. Requires visual verification of layout-sensitive tables and equations; do not use regex-only or static PDF parsing as the source of truth.

SKILL.md

3.7 KB, as published. Nobody here has run it

Paper extract

Turn papers/<versioned-id>/paper.pdf into a visually verified research record. Use the available PDF skill for all PDF reading and rendering. Read its SKILL.md before inspecting the paper and follow its render-and-verify workflow.

Required inputs and outputs

papers/<versioned-id>/
├── paper.pdf
├── meta.json
├── source_manifest.json
└── extraction/
    ├── note.md
    ├── reported.json
    └── extraction_report.md

Read references/extraction_schema.md before writing reported.json and references/note_template.md before writing note.md.

Workflow

  1. Verify the source. Match the title, authors, and arXiv version shown in the PDF to meta.json and source_manifest.json. Stop on a mismatch.
  2. Build a page map. Use PDF text/search only to navigate. Record the pages containing the abstract, data, methodology, results, appendices, formulas, and each relevant performance table.
  3. Inspect visually. Render and inspect the title page, data/method pages, every core equation, every table used for a reported metric, and ambiguous footnotes. Text extraction is not sufficient evidence for tables, multi-column layouts, signs, superscripts, or equations.
  4. Create an evidence ledger. Every metric and formula in reported.json must point to a page and a precise locator such as Table 4, row MSP2, column tau=0.95. Include a short source quote or transcription and set visually_verified: true only after inspecting the rendered page.
  5. Separate variants. Assign stable IDs to strategies and variants. Never collapse several lookbacks, quantiles, universes, or cost regimes into one metric.
  6. Choose replication targets. Define one or more target bundles that identify the exact strategy/variant and metric IDs a later run should reproduce. Do not let the replication stage choose the first metric with a matching name.
  7. Write the synopsis and ambiguity report. State what is new, what the evidence supports, what remains underspecified, and whether available public data could support exact replication or only reconstruction.
  8. Validate. Run:
python3 scripts/validate_output.py papers/<versioned-id>/extraction/reported.json --strict

Fix all errors before handoff.

Large papers

Inspect in page batches and keep a page ledger in working notes. Do not truncate a long Results section from the end. Prioritize section boundaries, tables, figure captions, appendices referenced by the method, and pages cited by the paper's own result discussion.

Hard rules

  • Do not run or recreate the removed regex/formula/text extraction pipeline.
  • Do not treat OCR or extracted text as authoritative for layout-sensitive content.
  • Do not invent missing values, units, annualization conventions, or formulas.
  • Keep reported facts separate from interpretation and replication assumptions.
  • Preserve metric signs as printed and record whether a risk metric is signed or reported as a positive magnitude.
  • A TL;DR must state the paper-specific contribution, mechanism, comparison, and practical replication risk; it must not paraphrase the abstract.
  • If a table or formula cannot be visually resolved, mark it unresolved rather than guessing.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.