Run3 Classify Documents by Subject Content
[COLM'26] SkillLearnBench is the first benchmark for evaluating continual learning methods that automatically generate agent skills.
npx -y skills add cxcscmu/SkillLearnBench --skill run3_Classify-Documents-by-Subject-ContentAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
What its author says it does
Copied from the file, not written here
Use this skill to analyze extracted text content from documents and classify them into the correct subject folder based on keyword matching and content analysis.
SKILL.md
3.0 KB, 662 tokens by cl100k_base, as published. Nobody here has run it
When to use this skill
- After extracting content from PDF, PPTX, or DOCX files
- To determine which of the 5 subject folders a document belongs to
- Before moving files to their destination folders
Classification rules
Subject-specific keyword sets
LLM (Large Language Models)
- Keywords: "language model", "transformer", "neural network", "NLP", "large language model", "BERT", "GPT", "attention mechanism", "embedding", "tokenization", "natural language processing", "deep learning"
- Context: Papers about AI models trained on text data, language processing, machine learning models for text
Trapped ion and quantum computing
- Keywords: "trapped ion", "quantum computer", "qubit", "quantum gate", "quantum algorithm", "quantum circuit", "ion trap", "quantum computing", "quantum simulation", "quantum error correction"
- Context: Papers about quantum computing systems, quantum information processing, trapped ion technologies
Black hole
- Keywords: "black hole", "event horizon", "gravitational", "spacetime", "Hawking", "relativity", "general relativity", "singularity", "gravitational wave", "astrophysics", "cosmology"
- Context: Papers about gravitational physics, cosmology, astrophysical objects and phenomena
DNA
- Keywords: "DNA", "genetics", "gene", "molecular biology", "sequence", "protein", "genetic", "genomics", "mutation", "chromosome", "RNA", "amino acid"
- Context: Papers about biological molecules, genetics, molecular biology, genomics, biological sequences
Music history
- Keywords: "music", "composer", "symphony", "song", "musical", "historical", "concert", "notation", "harmony", "melody", "rhythm", "performance", "musician"
- Context: Papers or documents about music, composers, historical musical analysis, musical theory in historical context
Classification process
- Extract key phrases from the document's title, abstract, and first 500 words
- Count keyword matches for each of the 5 subject categories
- Identify dominant subject: The category with the most keyword matches
- Check context: Read surrounding sentences to confirm the subject matches
- Handle ambiguous cases:
- If two subjects match equally, read more content
- If still unclear, examine introduction/abstract more carefully
- Only assign to music_history as fallback if truly cannot match the other 4 categories
Confidence assessment
- High confidence: 5+ keywords match from one category, and no other category has more than 1-2 matches
- Medium confidence: 3-4 keywords match, but need to verify with context reading
- Low confidence: Only 1-2 keywords match; read full abstract/introduction before deciding
Decision output
For each document, record:
- Filename
- Detected subject
- Confidence level (high/medium/low)
- Key matching keywords
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.