A comprehensive collection of 100 AI katas progressing from simple fundamentals to complex, multi-agent and multi-modal systems. Each kata is designed as a hands-on task for practicing and mastering AI prompting and system design skills.
- Set 1: Core Fundamentals (Katas 1-20)
- Set 2: Advanced Techniques (Katas 21-40)
- Set 3: Specialized Skills (Katas 41-60)
- Set 4: Problem-Solving & Reasoning (Katas 61-80)
- Set 5: Robustness & Cognitive Skills (Katas 81-100)
-
Prompt Polishing Kata
Improve a poorly written prompt to make it precise, constrained, and testable. -
Instruction Following Kata
Give the model a multi-step instruction (e.g., "analyze → summarize → critique") and measure consistency. -
Few-Shot Prompting Kata
Create a task with examples and compare the model's output with and without the examples. -
Style Transfer Kata
Make the model rephrase text in a specific writing style (scientific, poetic, legal, etc.). -
Error Diagnosis Kata
Provide flawed model outputs and prompt the model to detect, classify, and fix the errors.
-
Chain-of-Thought Kata
Force the model to reason step-by-step and evaluate whether reasoning improves accuracy. -
Self-Consistency Kata
Ask the model to generate multiple reasoning paths and select the consensus answer. -
Role-Based Prompting Kata
Assign expert roles (e.g., "You are a senior data scientist…") and measure output changes. -
Schema-Constrained Output Kata
Make the model produce validated JSON, XML, or a custom schema; test error rates. -
Knowledge Distillation Kata
Have the model teach a concept in progressively simpler explanations (expert → child).
-
Task Decomposition Kata
Provide a complex goal and ask the model to break it into subtasks with dependencies. -
Reflection & Improvement Kata
Use a loop where the model critiques its own answer, improves it, and repeats. -
Retriever-Augmented Generation Kata
Simulate a RAG workflow: give a knowledge base and test retrieval + answer fusion. -
Code Generation Kata
Have the model write code, then request debugging and refactoring iterations. -
Tool-Use Simulation Kata
Have the model decide when to "call tools" (calculation, search, API) and justify the choice.
-
Multi-Agent Debate Kata
Simulate two AI agents debating a question; a third agent judges correctness. -
Long-Context Compression Kata
Give a long document and ask the model to produce a loss-graded compression (e.g., 10%, 5%, 1%). -
Plan-and-Execute Agent Kata
Model creates a plan (planner role) and another agent executes each step with validation. -
Multi-Modal Reasoning Kata
Provide image/text/table inputs (or describe them) and have the model integrate information. -
Autonomous Project Agent Kata
Give the model a multi-day project goal (e.g., "design a small curriculum"), require planning, progress tracking, reflection, and final deliverables.
-
Ambiguity Resolution Kata
Give the model a deliberately ambiguous question and prompt it to list possible interpretations before answering. -
Persona Switching Kata
Ask the model to respond in two different personas back-to-back and analyze consistency. -
Constraint Enforcement Kata
Create a prompt with strict rules (length, vocabulary limits, forbidden words) and evaluate compliance. -
Rewriting for Clarity Kata
Take a dense or confusing paragraph and ask the model to rewrite it for a specific reading level. -
Edge-Case Prompting Kata
Provide unusual or tricky inputs (empty fields, contradictory facts) and prompt the model to reason safely.
-
Rubric-Based Grading Kata
Provide a grading rubric and ask the model to score a piece of work using the rubric step-by-step. -
Theory-to-Example Kata
Take an abstract idea and ask the model to generate concrete examples, then generalize back. -
Multi-Perspective Summary Kata
Ask for summaries from different viewpoints (e.g., critic, supporter, economist, scientist). -
Hypothesis Testing Kata
Provide a claim and ask the model to generate hypotheses, evidence, counter-evidence, and a conclusion. -
Data Cleaning Kata
Give messy textual data and have the model normalize, categorize, and detect anomalies.
-
Prompt Optimization Kata
Have the model rewrite your prompt 3–5 times to maximize clarity, precision, or performance. -
Scenario Simulation Kata
Ask the model to simulate a dynamic system (e.g., supply chain, ecosystem, workplace conflict) with state updates. -
Algorithm Explanation Kata
Provide algorithm pseudocode and ask the model to walk through it line-by-line using an example input. -
Temporal Reasoning Kata
Give time-dependent events and ask the model to infer sequences, causality, or scheduling. -
Structured Comparison Kata
Have the model compare two or more items using a table with specific columns and constraints.
-
Truthfulness vs. Plausibility Kata
Give the model a mix of real and fake claims and ask it to distinguish "sounds plausible" from "actually true," with justification. -
Model Critique Kata
Provide a model answer (from another AI or older version) and ask the current model to give a detailed critique and rewrite. -
Self-Generated Dataset Kata
Ask the model to create a synthetic dataset, define schema, include edge cases, and validate it. -
Cross-Domain Transfer Kata
Ask the model to apply concepts from one domain (e.g., biology) to another (e.g., business strategy). -
Modular Agent Design Kata
Have the model design a multi-agent architecture with explicit modules (planner, researcher, executor, critic, memory), including message-passing protocols.
-
Ambiguity Annotation Kata
Provide the model a paragraph and ask it to highlight ambiguous phrases and rewrite them unambiguously. -
Terminology Extraction Kata
Give a block of text and ask the model to extract domain-specific terms and define them succinctly. -
Minimal Prompt Kata
Challenge the model to complete a task with the shortest possible prompt while maintaining quality. -
Prompt Anti-Pattern Identification Kata
Give examples of bad prompts and ask the model to diagnose what makes them poor and rewrite them. -
Response Classification Kata
Provide several model outputs and ask the model to classify them (correct, incorrect, hallucinated, irrelevant).
-
Reverse Engineering Intent Kata
Give the model an answer and ask it to generate the most likely original question or prompt. -
Logical Fallacy Detection Kata
Present arguments containing logical fallacies and have the model identify and explain them. -
Conditional Logic Tree Kata
Create nested if-then-else scenarios and test the model's ability to follow branching logic. -
Information Hierarchy Kata
Have the model organize information into a clear hierarchy (main points → sub-points → details). -
Assumption vs. Fact Separation Kata
Give mixed content and ask the model to separate verifiable facts from assumptions or opinions.
-
Recursive Summarization Kata
Summarize a document, then summarize the summary, continuing until reaching a single sentence. -
Counterfactual Reasoning Kata
Ask "what if" questions and have the model reason through alternate scenarios systematically. -
Bias Detection & Mitigation Kata
Analyze text for various biases and propose rewrites that minimize them. -
Emergent Pattern Recognition Kata
Provide data points and ask the model to identify patterns not explicitly stated. -
Meta-Learning Documentation Kata
Have the model document its own learning process while solving a problem.
-
Pipeline Architecture Kata
Design a complete data processing pipeline with error handling and validation at each stage. -
Feedback Loop Design Kata
Create a system where outputs inform future inputs with continuous improvement mechanisms. -
Resource Optimization Kata
Given constraints (time, tokens, API calls), optimize a workflow for efficiency. -
Failure Mode Analysis Kata
Identify all possible failure points in a system and design mitigations. -
Compositional Task Building Kata
Build complex tasks by composing simpler subtasks with clear interfaces.
-
Keyword-Driven Generation Kata
Force the model to produce an output that must include (or exclude) a specific list of keywords. -
Intent Clarification Kata
Give the model a vague user message and ask it to produce clarifying questions and rationale before answering. -
Redundant Text Removal Kata
Ask the model to remove repetition from a text while preserving meaning. -
Prompt Completion Kata
Provide the first half of a prompt and ask the model to guess the intended task and complete it logically. -
Microformat Rewriting Kata
Ask the model to rewrite text into a strict microformat (bullet rules, tweet threads, glossary entries, etc.).
-
Contradiction Detection Kata
Provide two paragraphs and have the model detect contradictions, ambiguities, or conflicts in logic. -
Sparse Clue Reasoning Kata
Give minimal information and ask the model to make logical inferences while stating assumptions. -
Precedence Rule Application Kata
Define a set of rules with priorities and test the model's ability to apply them correctly. -
Evidence Weighting Kata
Present multiple pieces of evidence of varying quality and have the model weight them appropriately. -
Diagnostic Tree Navigation Kata
Create a decision tree and have the model navigate it based on given symptoms or conditions.
-
Strategic Planning Kata
Given goals and constraints, develop a multi-phase strategy with contingency plans. -
Trade-off Analysis Kata
Present scenarios requiring trade-offs and have the model analyze pros/cons systematically. -
System Bottleneck Identification Kata
Describe a complex system and have the model identify performance bottlenecks. -
Emergent Behavior Prediction Kata
Given simple rules, predict complex emergent behaviors in a system. -
Optimization Under Uncertainty Kata
Make decisions with incomplete information while quantifying uncertainty.
-
Distributed Decision Making Kata
Design how multiple agents coordinate decisions without central control. -
Adversarial Scenario Planning Kata
Plan strategies assuming an intelligent adversary trying to counter your moves. -
Knowledge Graph Construction Kata
Build a knowledge graph from unstructured text with entities, relationships, and confidence scores. -
Cognitive Load Balancing Kata
Distribute complex tasks across agents based on their capabilities and current load. -
Self-Modifying Workflow Kata
Create workflows that can modify themselves based on performance metrics.
-
Context Boundary Kata
Give the model a long conversation or document and ask it to identify what is and is not relevant to a specific question. -
Perspective Flip Kata
Provide an argument and ask the model to reframe it from the opposite viewpoint with equal coherence. -
Controlled Randomness Kata
Have the model generate an idea or example with a specific "creativity level" (e.g., 0–10 scale). -
Simple Fact Distillation Kata
Provide a dense paragraph and have the model distill just the atomic facts with no interpretation. -
Keyword-to-Paragraph Kata
Give only a list of 5–8 keywords and ask the model to construct a coherent, factual paragraph connecting them.
-
Cause–Effect Mapping Kata
Provide an event and ask the model to produce upstream causes and downstream effects in a structured tree. -
Hypothetical Rewrite Kata
Ask the model to rewrite a scenario under a new assumption (e.g., "What if gravity were half as strong?"). -
Constraint-First Planning Kata
Provide hard constraints first and ask the model to develop a plan that satisfies all of them. -
Error Taxonomy Creation Kata
Give the model several flawed responses and ask it to invent a taxonomy of error types, then categorize each item. -
Progressive Elaboration Kata
Have the model take a one-sentence idea and elaborate it into:- a paragraph
- an outline
- a full plan
- a refined final version
-
Adversarial Prompt Defense Kata
Ask the model to identify possible adversarial attacks hidden in a prompt and rewrite the prompt safely. -
Meta-Prompt Interpretation Kata
Give a prompt about how to construct prompts and ask the model to analyze and optimize it. -
Nested Reasoning Kata
Instruct the model to reason at three levels:- immediate step reasoning
- meta-reasoning (why that step?)
- meta-meta reasoning (how to improve reasoning itself)
-
Scenario Divergence Kata
Ask the model to simulate two alternate timelines from a single starting point and show how they diverge. -
Implicit Assumption Extraction Kata
Provide a text and have the model extract every hidden assumption the writer is making.
-
Spec-to-Implementation Kata
Give the model a specification and ask it to describe how different agents or modules would implement it. -
Self-Healing Workflow Kata
Ask the model to create a workflow where each step has built-in mechanisms to detect and repair errors from previous steps. -
Role-Conflict Simulation Kata
Have the model simulate two agents with conflicting goals negotiating toward a resolution, with reasoning traces. -
Cross-Format Translation Kata
Ask the model to convert information across formats (e.g., map → story → checklist → SOP) while preserving essential data. -
Concept Spine Building Kata
Have the model extract the "spine" (core conceptual skeleton) of a topic and then rebuild the full knowledge structure from it.
- 🟢 Beginner (25 katas): Focus on basic prompting, control, and understanding
- 🟡 Intermediate (25 katas): Emphasize reasoning, structure, and multi-step processes
- 🔴 Advanced (25 katas): Cover planning, optimization, and complex systems
- ⚫ Expert (25 katas): Address multi-agent systems, meta-cognition, and high-complexity tasks
- Prompt Engineering: Crafting, optimizing, and debugging prompts
- Reasoning & Logic: Chain-of-thought, consistency, logical inference
- Structure & Format: Schema compliance, hierarchies, conversions
- Planning & Strategy: Decomposition, optimization, contingency planning
- Multi-Agent Systems: Coordination, debate, distributed decision-making
- Error Handling: Detection, classification, self-correction
- Meta-Cognition: Reflection, self-improvement, learning documentation
- Robustness: Edge cases, adversarial inputs, failure modes
- Knowledge Management: Extraction, distillation, graph construction
- Creative Problem-Solving: Cross-domain transfer, emergent patterns, divergent thinking
- Sequential Practice: Work through katas in order, building skills progressively
- Targeted Training: Focus on specific difficulty levels or skill categories
- Daily Practice: Select one kata per day for consistent improvement
- Team Challenges: Use katas for group training or competitive practice
- Skill Assessment: Use subsets as benchmarks for evaluating AI capabilities
- Curriculum Building: Combine katas into structured learning paths
Katas: 1, 3, 21, 23, 31, 43, 44, 61, 62, 91, 92
Katas: 2, 6, 7, 29, 47, 48, 66, 69, 93, 95
Katas: 11, 16, 18, 40, 56, 76, 96, 97, 98, 100
Katas: 5, 25, 45, 54, 59, 89, 91, 97
Katas: 4, 27, 39, 82, 83, 87, 94, 99
This collection represents a comprehensive training resource for developing AI interaction skills from foundational to expert level. Regular practice with these katas will build proficiency in prompt engineering, system design, and advanced AI orchestration.