Extract structured metamaterial knowledge from paper abstracts. Use when you have retrieved papers and need to classify each paper's geometry family, mechanisms, and claims into structured records.
You are a metamaterial design expert. Extract structured knowledge from paper abstracts into PaperRecord objects.
When this skill is invoked (typically after literature retrieval):
Read the retrieved papers — either from a previous retrieval step or from files in D:/Claude/target_to_hypothesis/tests/intermediates/.
For each paper abstract, classify it using your domain expertise:
geometry_family: one of the canonical families from D:/Claude/target_to_hypothesis/config/family_ontology.yaml:
square_ring, split_ring, slotted_square_ring, patch_with_slots, cross_resonator, mirrored_disconnected, nested_multi_resonator, supercell, resistive_film, multilayermechanisms: list from D:/Claude/target_to_hypothesis/config/mechanism_ontology.yaml:
LC_resonance, magnetic_resonance, dipole_resonance, multi_resonance, impedance_matching, near_field_coupling, slot_resonance, capacitive_loading, Fabry_Perot, ohmic_loss, symmetry_breaking, current_perturbationrelevance_to_target: 0.0–1.0 score for how relevant this paper is to the current targetconfidence: 0.0–1.0 your confidence in the classificationclaims: key findings (e.g., "achieved 95% absorption at 10 GHz using nested SRR on FR4")Filter out irrelevant papers (relevance < 0.1): vibration dampers, acoustic metamaterials, heat exchangers, optical-only devices when target is microwave, etc.
Output structured records as JSON or pass to the next pipeline step.
cd D:/Claude && python -c "
from target_to_hypothesis.skills.paper_reader import extract_batch
# paper_records = extract_batch(paper_candidates, target, llm_fn=my_llm_fn)
"
D:/Claude/target_to_hypothesis/skills/paper_reader.py
D:/Claude/target_to_hypothesis/models/literature.py