agentleFS
lucaswychanlucaswychan/quant-paper-agent/.agents/skills/paper-extract/SKILL.md

paper-extract

Skill0 starsChanged 3 months agoMITView on GitHub

Inspect a quantitative-finance paper PDF with the available PDF capability and produce a checksum-bound, evidence-backed synopsis, metric dataset, formulas, exact replication targets, and ambiguity report. Use when paper.pdf exists and the user wants contributions, methods, tables, formulas, data design, performance claims, or a replication handoff. Requires visual verification of layout-sensitive tables and equations.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    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
  1. Very specific:

    Verify that the PDF title, authors, arXiv version, and checksums agree with meta.json and source_manifest.json.

    line 28
  2. Distinctive:

    Never use regex, OCR, or extracted text as authoritative evidence for layout-sensitive content.

    line 54
  3. Distinctive:

    Never invent values, units, signs, annualization, cost status, or formulas.

    line 55
  4. Distinctive:

    Keep reported facts separate from interpretation and replication assumptions.

    line 56

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Paper extract

Turn papers/<versioned-id>/paper.pdf into a visually verified, finalized research record. Read and follow the available visual PDF capability before inspecting the paper. If pages cannot be rendered and inspected, stop; extracted text alone cannot satisfy this skill.

Resolve paths and contracts

Resolve the absolute workspace root and this skill's directory. Read references/extraction_schema.md, references/reported.schema.json, and references/note_template.md before authoring artifacts.

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

Workflow

  1. Verify that the PDF title, authors, arXiv version, and checksums agree with meta.json and source_manifest.json. Stop on a mismatch.
  2. Build a complete page map using text/search only for navigation.
  3. Render and inspect the title, data/method pages, every core equation, every table used by a metric, and ambiguous footnotes.
  4. Record source.page_count, every rendered page in source.visually_inspected_pages, and the actual PDF, metadata, and source-manifest hashes.
  5. Attach page, table/equation, row/column, short transcription, and visual-verification status to every metric and formula.
  6. Assign stable strategy, variant, sample, and cost-regime IDs. Do not combine incompatible scopes in one replication target.
  7. Recompute evidence_coverage; state unresolved choices and likely replication classification.
  8. Validate and finalize:
QPA_EXTRACT_SKILL_DIR="/absolute/path/to/paper-extract"
QPA_WORKSPACE_ROOT="/absolute/path/to/workspace"
python3 "$QPA_EXTRACT_SKILL_DIR/scripts/validate_output.py" \
  "$QPA_WORKSPACE_ROOT/papers/<versioned-id>/extraction/reported.json" --strict
python3 "$QPA_EXTRACT_SKILL_DIR/scripts/finalize_extraction.py" \
  <versioned-id> --root "$QPA_WORKSPACE_ROOT/papers"

The validator enforces both reported.schema.json and cross-record/source invariants. The finalizer hashes the PDF, source metadata, note, reported dataset, and ambiguity report. Later changes invalidate the replication gate.

Large papers

Inspect in page batches and maintain a page ledger. Do not truncate long Results sections. Prioritize section boundaries, performance tables, captions, method-referenced appendices, and pages cited by the paper's result discussion.

Rules

  • Never use regex, OCR, or extracted text as authoritative evidence for layout-sensitive content.
  • Never invent values, units, signs, annualization, cost status, or formulas.
  • Keep reported facts separate from interpretation and replication assumptions.
  • Preserve metric signs and declare risk-metric polarity.
  • Make the TL;DR paper-specific: contribution, mechanism, fair comparison, evidence, and replication risk.
  • Mark unresolved material rather than guessing.
  • Author only the current reported.schema.json contract; reject noncurrent extraction artifacts and re-extract them from the bound PDF.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.