|
| 1 | +[ |
| 2 | + { |
| 3 | + "id": "code_python", |
| 4 | + "name": "Python Code Generation", |
| 5 | + "description": "Writing, completing, or refactoring Python code including functions, classes, scripts, and modules.", |
| 6 | + "validation_examples": [ |
| 7 | + "Write a Python function to reverse a linked list", |
| 8 | + "Create a Python class for a binary search tree", |
| 9 | + "Implement quicksort in Python", |
| 10 | + "Write a decorator that caches function results in Python", |
| 11 | + "Refactor this Python code to use list comprehensions", |
| 12 | + "Write a Python script to parse CSV files and compute statistics", |
| 13 | + "Create a Python context manager for database connections", |
| 14 | + "Generate a Python dataclass for a user profile" |
| 15 | + ], |
| 16 | + "embedding_centroid": null, |
| 17 | + "preferred_adapters": [], |
| 18 | + "fallback_adapters": [] |
| 19 | + }, |
| 20 | + { |
| 21 | + "id": "code_sql", |
| 22 | + "name": "SQL Query Writing", |
| 23 | + "description": "Writing SQL queries including SELECT, JOIN, aggregations, CTEs, window functions, and schema design.", |
| 24 | + "validation_examples": [ |
| 25 | + "Write a SQL query to find duplicate rows in a table", |
| 26 | + "Create a SQL query with a LEFT JOIN between orders and customers", |
| 27 | + "Write a SQL window function to calculate running totals", |
| 28 | + "Generate a SQL CTE to rank employees by department salary", |
| 29 | + "Write a SQL query to find users who made purchases in the last 30 days", |
| 30 | + "Create a SQL stored procedure for bulk inserts", |
| 31 | + "Write a SQL query using GROUP BY and HAVING to filter aggregates", |
| 32 | + "Generate a SQL schema for an e-commerce database" |
| 33 | + ], |
| 34 | + "embedding_centroid": null, |
| 35 | + "preferred_adapters": [], |
| 36 | + "fallback_adapters": [] |
| 37 | + }, |
| 38 | + { |
| 39 | + "id": "code_javascript", |
| 40 | + "name": "JavaScript and TypeScript", |
| 41 | + "description": "Writing or reviewing JavaScript and TypeScript code for frontend, Node.js, or full-stack applications.", |
| 42 | + "validation_examples": [ |
| 43 | + "Write a React hook to fetch data from an API", |
| 44 | + "Create a TypeScript interface for a REST API response", |
| 45 | + "Implement a JavaScript debounce function", |
| 46 | + "Write an async function to handle file uploads in Node.js", |
| 47 | + "Create a TypeScript generic type for a paginated API response", |
| 48 | + "Write a JavaScript event listener for form validation", |
| 49 | + "Implement a Promise-based queue in JavaScript", |
| 50 | + "Write a Next.js API route that authenticates with JWT" |
| 51 | + ], |
| 52 | + "embedding_centroid": null, |
| 53 | + "preferred_adapters": [], |
| 54 | + "fallback_adapters": [] |
| 55 | + }, |
| 56 | + { |
| 57 | + "id": "code_debug", |
| 58 | + "name": "Debugging and Error Fixing", |
| 59 | + "description": "Diagnosing, explaining, and fixing bugs, exceptions, and unexpected behaviour in any programming language.", |
| 60 | + "validation_examples": [ |
| 61 | + "Fix the KeyError in this Python dictionary access code", |
| 62 | + "Why is this JavaScript async function returning undefined?", |
| 63 | + "Debug this SQL query that returns no results", |
| 64 | + "Find the memory leak in this C++ code", |
| 65 | + "Why does this React component re-render infinitely?", |
| 66 | + "Fix the NullPointerException in this Java method", |
| 67 | + "Explain and fix this Python TypeError: unsupported operand", |
| 68 | + "Debug why this API call returns 403 Forbidden", |
| 69 | + "Fix the off-by-one error in this loop" |
| 70 | + ], |
| 71 | + "embedding_centroid": null, |
| 72 | + "preferred_adapters": [], |
| 73 | + "fallback_adapters": [] |
| 74 | + }, |
| 75 | + { |
| 76 | + "id": "text_summarize", |
| 77 | + "name": "Text Summarization", |
| 78 | + "description": "Condensing long documents, articles, papers, or conversations into concise summaries.", |
| 79 | + "validation_examples": [ |
| 80 | + "Summarise this research paper in 3 bullet points", |
| 81 | + "Give me a TL;DR of this news article", |
| 82 | + "Write an executive summary of this annual report", |
| 83 | + "Condense this meeting transcript into action items", |
| 84 | + "Summarise the key findings from this dataset description", |
| 85 | + "Create a one-paragraph abstract for this technical documentation", |
| 86 | + "What are the main takeaways from this blog post?", |
| 87 | + "Summarise this legal contract in plain English" |
| 88 | + ], |
| 89 | + "embedding_centroid": null, |
| 90 | + "preferred_adapters": [], |
| 91 | + "fallback_adapters": [] |
| 92 | + }, |
| 93 | + { |
| 94 | + "id": "text_classify", |
| 95 | + "name": "Text Classification and Labeling", |
| 96 | + "description": "Classifying, categorising, or labelling text — sentiment, intent, topic, or custom taxonomy.", |
| 97 | + "validation_examples": [ |
| 98 | + "Classify this customer review as positive, negative, or neutral", |
| 99 | + "Label these support tickets by urgency level", |
| 100 | + "Categorise these product descriptions by department", |
| 101 | + "Detect the intent behind this user message: book a flight", |
| 102 | + "Tag these news headlines with their topic", |
| 103 | + "Classify this email as spam or legitimate", |
| 104 | + "Label this text for named entity recognition", |
| 105 | + "Determine the sentiment of these tweets", |
| 106 | + "Categorise these bug reports by severity" |
| 107 | + ], |
| 108 | + "embedding_centroid": null, |
| 109 | + "preferred_adapters": [], |
| 110 | + "fallback_adapters": [] |
| 111 | + }, |
| 112 | + { |
| 113 | + "id": "text_translate", |
| 114 | + "name": "Language Translation", |
| 115 | + "description": "Translating text between languages, including idioms, technical content, and cultural adaptation.", |
| 116 | + "validation_examples": [ |
| 117 | + "Translate this paragraph from English to French", |
| 118 | + "Translate this technical document to Spanish", |
| 119 | + "Convert this German medical text to English", |
| 120 | + "Translate this Japanese product description to English", |
| 121 | + "Localise this marketing copy from English to Portuguese", |
| 122 | + "Translate this legal clause from Latin to English", |
| 123 | + "Convert this Chinese business email to English", |
| 124 | + "Translate this Python error message to plain English" |
| 125 | + ], |
| 126 | + "embedding_centroid": null, |
| 127 | + "preferred_adapters": [], |
| 128 | + "fallback_adapters": [] |
| 129 | + }, |
| 130 | + { |
| 131 | + "id": "math_reasoning", |
| 132 | + "name": "Math and Quantitative Reasoning", |
| 133 | + "description": "Solving arithmetic, algebra, calculus, statistics, logic puzzles, and word problems step by step.", |
| 134 | + "validation_examples": [ |
| 135 | + "Solve this system of linear equations", |
| 136 | + "Calculate the derivative of this function", |
| 137 | + "What is the probability of drawing two aces from a deck?", |
| 138 | + "Solve this word problem: if a train leaves at 3 PM at 60 mph...", |
| 139 | + "Compute the integral of x^2 from 0 to 5", |
| 140 | + "Find the eigenvalues of this matrix", |
| 141 | + "How many ways can 5 people be seated in a row?", |
| 142 | + "What is 15% compound interest on $10,000 over 3 years?", |
| 143 | + "Prove that the square root of 2 is irrational" |
| 144 | + ], |
| 145 | + "embedding_centroid": null, |
| 146 | + "preferred_adapters": [], |
| 147 | + "fallback_adapters": [] |
| 148 | + }, |
| 149 | + { |
| 150 | + "id": "qa_factual", |
| 151 | + "name": "Factual Q&A", |
| 152 | + "description": "Answering factual questions about history, science, geography, technology, or general knowledge.", |
| 153 | + "validation_examples": [ |
| 154 | + "What is the capital of Australia?", |
| 155 | + "Who invented the telephone?", |
| 156 | + "What year did World War II end?", |
| 157 | + "What is the speed of light in a vacuum?", |
| 158 | + "Explain how HTTPS encryption works", |
| 159 | + "What is the difference between RAM and ROM?", |
| 160 | + "What causes the aurora borealis?", |
| 161 | + "How many bones are in the human body?", |
| 162 | + "What is the GDP of Germany?" |
| 163 | + ], |
| 164 | + "embedding_centroid": null, |
| 165 | + "preferred_adapters": [], |
| 166 | + "fallback_adapters": [] |
| 167 | + }, |
| 168 | + { |
| 169 | + "id": "agent_planning", |
| 170 | + "name": "Task Decomposition and Agent Planning", |
| 171 | + "description": "Breaking down complex goals into step-by-step plans, sub-tasks, or agent workflows.", |
| 172 | + "validation_examples": [ |
| 173 | + "Create a step-by-step plan to build a REST API from scratch", |
| 174 | + "Decompose the task of launching a product into weekly milestones", |
| 175 | + "Write a plan for migrating a database to the cloud", |
| 176 | + "Break down the goal 'build a recommendation system' into sub-tasks", |
| 177 | + "Create an agent workflow to automate data ingestion and reporting", |
| 178 | + "Plan the steps to refactor a monolith into microservices", |
| 179 | + "What are the subtasks needed to onboard a new engineer?", |
| 180 | + "Outline a research plan for evaluating three competing ML models", |
| 181 | + "Create a checklist for deploying a web app to production" |
| 182 | + ], |
| 183 | + "embedding_centroid": null, |
| 184 | + "preferred_adapters": [], |
| 185 | + "fallback_adapters": [] |
| 186 | + } |
| 187 | +] |
0 commit comments