Run2 advanced subject classifier
Advanced document classification using weighted keyword scores and expanded subject vocabularies.From its SKILL.md
npx -y skills add cxcscmu/SkillLearnBench --skill run2_advanced_subject_classifierAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
1.5 KB, 342 tokens by cl100k_base, as published. Nobody here has run it
Advanced Subject Classifier Skill
Expanded Keyword Sets
- LLM: Large Language Model, LLM, Transformer, GPT, BERT, Attention mechanism, Tokenization, Pre-training, Fine-tuning, RLHF, NLP, Generative AI.
- trapped_ion_and_qc: Trapped Ion, Quantum Computing, Qubit, Quantum Gate, Entanglement, Superposition, Ion Trap, Quantum Error Correction, Quantum Information, Paul Trap, Rydberg, NIST.
- black_hole: Black Hole, Event Horizon, Schwarzschild, Hawking Radiation, General Relativity, Singularity, Accretion Disk, Spacetime Curvature, LIGO, Gravitational Waves, Kerr metric.
- DNA: DNA, Genome, Genetic, Sequencing, Nucleotide, RNA, Protein Synthesis, Chromosome, CRISPR, Base Pair, Genomics, PCR, Polymerase.
- music_history: Music History, Composer, Symphony, Opera, Baroque, Classical, Romantic, Orchestra, Melody, Harmony, Polyphony, Sonata, Fugue, Jazz, Blues, Ethnomusicology.
Scoring Algorithm
Assign weights to keywords. If multiple subjects match, the one with the highest total score wins.
Example: "LLM" (3 points), "Quantum" (1 point) -> Subject is LLM.
Special case: If a file contains very little text (e.g. just a title), prioritize the title content.
If no matches, the file goes to the last subject music_history by default as per instructions (but only if it truly fits nowhere else).
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.