Prompt Smells and Optimization Dataset for Prompt Engineering

Abstract

Prompt Smells and Optimization Dataset & Local AI Agent is an open-access research dataset and autonomous evaluation toolkit for studying Prompt Smells (anti-patterns) in Large Language Models (LLMs) and Prompt Engineering education. 📊 Dataset Overview The dataset contains 435 curated English examples structured specifically for analyzing flawed user prompts, understanding root causes, and applying production-ready prompt engineering optimizations: User Wrong Prompts: Raw, imperfect prompts exhibiting common prompt engineering smells (e.g. ambiguity, small prompts / extreme brevity, missing context, lack of role, omitted output format). Pedagogical Analysis / Reason: In-depth explanations breaking down why the original prompt underperforms and what prompt engineering principles apply. Optimized Prompts: Production-ready engineered prompts incorporating explicit roles, context placeholders (e.g., {{variable}}), constraints, and structured directives. 🛠️ Local AI Agent & Pipeline Architecture This release includes a modular Python pipeline (prompt_smell_agent) and CLI tool (analyze_parquet_smells.py) capable of: Parquet Stream Parsing: Ingesting large Parquet files without memory overhead, auto-detecting flat columns and multi-turn nested conversational structures. High-Accuracy English Filtering: Dual-stage filtering combining Unicode script range exclusion with Latin-script stopword distribution checking. Prompt Smell Scoring Engine: Quantitatively scoring prompt quality (0–100) across 7 critical anti-pattern dimensions. Autonomous Prompt Optimization: Heuristic template synthesis and optional local LLM backend integration (Ollama / LM Studio). 📁 File Formats Included prompt_smells_and_corrections.xlsx: Formatted Excel spreadsheet with custom styling, wrapped text, and frozen headers. analyzed_prompt_smells.xlsx: Color-coded severity-badged analysis workbook. prompt_dataset.parquet: Apache Parquet formatted benchmark dataset. prompt_optimization_dataset.jsonl: Raw JSON Lines dataset. PROMPT_SMELL_ANALYSIS_REPORT.md: Executive summary report with quantitative empirical metrics. prompt_smell_agent/: Complete reusable Python package.