eda-pdf-local-summary

Summarize local paper PDFs into structured, citation-grounded evidence notes. Use when the user provides local PDF paths and asks for methods/assumptions/results/limitations extraction or wants evidence mapped to current EDA hypotheses.

EDA PDF Local Summary

Overview

Use this skill to produce reusable local summary notes tied to page-level evidence.

Workflow

  1. Validate input PDF path.
  • require absolute or repo-relative local path.
  • expected location: docs/papers/pdf/.
  1. Extract text to local artifact.
  • Run scripts/extract_pdf_text.sh <pdf_path> <out_dir>.
  • default output dir: docs/papers/summaries/raw/.
  • default extractor priority is GROBID (via grobid_client + GROBID_SERVER), then falls back to pdftotext/mutool/pypdf if unavailable.
  • to disable GROBID for one run: set DISABLE_GROBID=1.
  • quality guard: if GROBID TEI result is too sparse, extractor auto-falls back to local text extraction.
  • when GROBID succeeds, keep TEI artifact alongside text (<pdf_stem>.tei.xml) for traceability.
  1. Build structured summary.
  • Load references/summary-template.md.
  • Fill sections with explicit evidence snippets.
  • Prefer page-cited claims (p.xx) when recoverable from extracted text.
  • write output to: docs/papers/summaries/<pdf_stem>.summary.md.

3.5 Update paper manifest status.

  • locate corresponding row in docs/papers/manifests/*.paper_manifest.tsv.
  • update local_pdf_path and set status=summarized.
  1. Sync landscapes after summary write.
  • run python3 scripts/common/paper_landscape_sync.py build.
  • ensure both landscape docs are updated:
    • docs/knowledge_base/95_PROBLEM_LANDSCAPE.md
    • docs/knowledge_base/96_METHOD_LANDSCAPE.md
  1. Link to active hypothesis.
  • State whether paper supports, contradicts, or is neutral to current model/flow hypothesis.
  • add one "actionable next experiment" sentence.

Output Contract

Always output:

  1. extracted text artifact path,
  2. summary markdown path (docs/papers/summaries/<pdf_stem>.summary.md),
  3. 3-5 bullet findings with evidence strength (strong/medium/weak),
  4. unresolved ambiguities requiring manual read,
  5. updated manifest path (or explicit note if no manifest row matched).
  6. landscape sync artifact paths (95/96 docs or explicit failure reason).

References

Load:

  1. references/summary-template.md
  2. references/evidence-grading.md