Context Engineering
Managing information flow and structure for optimal AI performance
Technique Taxonomy
192 techniques across 15 categories
Categories
Workflow Engineering
Orchestrating multi-step agent interactions and task coordination
Agentic Frameworks
Advanced autonomous agent architectures and multi-agent systems
Advanced Prompting Strategies
Sophisticated prompting techniques involving complex reasoning structures and meta-learning.
Basic Concepts
Fundamental prompting structures and conceptual frameworks
Reasoning Frameworks
Techniques that guide the model through explicit reasoning steps
Agent & Tool Use
Techniques that enable LLMs to interact with external tools and environments
Self-Improvement Techniques
Methods for the model to reflect on and improve its own outputs
Retrieval & Augmentation
Techniques that incorporate external knowledge into prompts
Prompt Optimization
Techniques to automate and improve prompt engineering
Multimodal Techniques
Techniques involving non-text modalities like images, audio, and video
Specialized Application Techniques
Techniques optimized for specific domains or applications
Multi-Agent Systems & Team Frameworks
Advanced techniques for organizing and coordinating multiple AI agents
Secure Agent Architectures
Architectural design patterns for building secure and resilient LLM agents against threats like prompt injection.
Prompt Structure & Engineering
Techniques for structuring prompts and engineering the context for optimal model performance.
All Techniques (192)
192 techniquesEfficient Attention Mechanisms
Context EngineeringFlash-Attention 3, sparse attention, and sub-quadratic scaling for processing massive contexts (200K+ tokens)
Context Compression
Context EngineeringAdvanced compression techniques for storing and processing large contexts efficiently
Sparse Attention for 1M+ Tokens
Context EngineeringAttention mechanisms that only compute on relevant tokens for ultra-long contexts
Dynamic Context Windowing
Context EngineeringVariable context size optimization based on task requirements and efficiency needs
Memory Management Strategies
Context EngineeringSystematic approaches to managing context memory across long interactions
Cross-Modal Context Fusion
Context EngineeringIntegrating information across text, image, audio, and action modalities
Context Layer Architecture
Context EngineeringOrganized context into persistent, session, immediate, and transient layers
Context Prioritization
Context EngineeringIntelligent ordering and filtering of context based on relevance and task requirements
World Model Forecasting
Context EngineeringPredictive context generation using world models for anticipatory reasoning
Continuous Representation
Context EngineeringMoving from discrete token generation to continuous representations
Context Compression Algorithms
Context EngineeringAlgorithmic approaches to compressing context while preserving meaning
Higher-Order Linear Attention
Context EngineeringAdvanced attention mechanisms that bridge linear and softmax attention efficiency
Agentic Environment Isolation
Context EngineeringBounded execution contexts for agents with specific tools and resources
Context Federation
Context EngineeringDistributing and coordinating context across multiple agents and systems
Video Keyframe Selection
Context EngineeringEfficient selection of important frames from long videos for multimodal processing
Memory-Efficient Transformers
Context EngineeringGradient checkpointing and activation recomputation for context optimization
Speculative Context Management
Context EngineeringAnticipatory context loading based on predicted information needs
Model Context Protocol Integration
Context EngineeringStandardized protocol for context delivery and tool interoperability
Context Caching
Context EngineeringCaching pre-computed states of the context window (prefix caching) to reduce latency and cost for repeated queries.
Agentic RAG
Context EngineeringAutonomous retrieval where an agent formulates queries, critiques results, and iteratively searches based on findings.
Boomerang Coordination Pattern
Workflow EngineeringDistributed agent coordination with structured returns and validation
Multi-Agent State Management
Workflow EngineeringPersistent, shared state across multiple agents with isolation and validation
Hierarchical Task Planning
Workflow EngineeringSystematic decomposition across strategic, tactical, and operational layers
Graph-Based Workflow Optimization
Workflow EngineeringWorkflow execution optimization using graph theory and dependency analysis
Speculative Execution
Workflow EngineeringProactive task execution based on predicted needs with rollback capabilities
Mixture-of-Agents Architecture
Workflow EngineeringSpecialized agents with intelligent routing and consensus mechanisms
Agent Specialization
Workflow EngineeringDomain-specific agent development with capability-based routing
Multi-Agent Orchestration
Workflow EngineeringCoordination frameworks for managing multiple specialized agents
Inter-Agent Communication
Workflow EngineeringStructured communication protocols between agents with standardized formats
Adaptive Execution
Workflow EngineeringDynamic workflow modification based on intermediate results and environmental changes
Agentic RAG
Workflow EngineeringAutonomous retrieval-augmented generation with agent-driven retrieval strategies
RDMA Communication Infrastructure
Workflow EngineeringHigh-performance remote direct memory access for distributed LLM systems
Workflow Graph Analysis
Workflow EngineeringDependency analysis, critical path identification, and optimization using graph theory
Intelligent Task Decomposition
Workflow EngineeringAutomatic breaking down of complex tasks into manageable subtasks
Dynamic Resource Allocation
Workflow EngineeringIntelligent distribution of computational resources based on workflow needs
Systematic Failure Recovery
Workflow EngineeringStructured approaches to error handling and workflow recovery
Parallel Workflow Execution
Workflow EngineeringIdentifying and executing independent tasks simultaneously
Workflow Validation
Workflow EngineeringSystematic validation of workflow correctness and completeness
Workflow Execution Monitoring
Workflow EngineeringReal-time monitoring and optimization of workflow performance
Reflection & Feedback Loops
Workflow EngineeringExplicit workflow steps where the agent critiques its own output or intermediate state before proceeding.
Semantic/Dynamic Routing
Workflow EngineeringUsing a router (LLM or classifier) to dispatch queries to the most efficient model or specialized agent.
Vision-Language-Action Architectures
Agentic FrameworksUnified architectures for multimodal understanding and action execution
Dual-System VLA Designs
Agentic FrameworksTwo-system architectures separating perception from action planning
Graph-Based Chain-of-Thought
Agentic FrameworksGraphCoT-VLA for complex spatial reasoning and instruction following
Continuous Autoregressive Models
Agentic FrameworksParadigm shift from discrete to continuous token generation
Speculative Sparse Attention
Agentic FrameworksCombining speculative inference with sparse attention for efficiency
OpenHands Autonomous Learning
Agentic FrameworksFull autonomous coding with system interaction and continuous learning
Continue Agentic Workflows
Agentic FrameworksMulti-step reasoning with autonomous refactoring and project memory
Swarms Multi-Agent Coordination
Agentic FrameworksLarge-scale multi-agent orchestration with tree-of-thoughts integration
Open Interpreter Natural Language Computing
Agentic FrameworksNatural language control of computer systems with session-based context
Diffusion Policy Integration
Agentic FrameworksCombining diffusion models with autoregressive approaches for VLA
Embodied AI with Sim-to-Real Transfer
Agentic FrameworksAI systems designed for physical world interaction with transfer learning
Safety-Critical Adaptive Control
Agentic FrameworksAdaptive control systems with safety guarantees for critical applications
Autonomous Decision-Making
Agentic FrameworksAgent-driven decision systems with minimal human oversight
Mixture of Experts Architectures
Agentic FrameworksSpecialized expert networks with intelligent routing mechanisms
Agentic Synthetic Data Generation
Agentic FrameworksAutonomous generation of training data using agent-driven quality assessment
Real-Time Agent Inference
Agentic FrameworksLow-latency inference systems for real-time agent applications
Cross-Modal Attention Pooling
Agentic FrameworksAttention mechanisms across different modalities for unified understanding
Agent Memory Persistence
Agentic FrameworksLong-term memory systems for agents with learning and adaptation
Multimodal Fusion Architectures
Agentic FrameworksUnified architectures for processing and understanding multiple modalities
Agent Tool Composition
Agentic FrameworksDynamic composition and chaining of tools by autonomous agents
Speculative Inference Patterns
Agentic FrameworksAnticipatory execution for improved performance in agent workflows
Agent Consensus Mechanisms
Agentic FrameworksCoordination protocols for multiple agents reaching agreement
Contextual Agent Routing
Agentic FrameworksIntelligent routing of tasks to appropriate agents based on context
Agent Fallback Systems
Agentic FrameworksBackup agent systems for handling degraded performance scenarios
Agent Validation Protocols
Agentic FrameworksCross-specialist validation for ensuring output quality and consistency
Graph of Thoughts (GoT)
Advanced Prompting StrategiesModeling reasoning as a graph where thoughts are nodes, allowing for non-linear exploration and combination.
Meta-Prompting
Advanced Prompting StrategiesUsing a 'meta-model' or higher-level prompt to orchestrate multiple sub-models or generate task-specific prompts.
Recursive Self-Refinement
Advanced Prompting StrategiesIteratively improving an output by feeding it back into the model with specific critique instructions.
Basic Prompting
Basic ConceptsThe simplest form of prompting, usually consisting of an instruction and input, without exemplars or complex reasoning steps.
Also known as: Standard Prompting, Vanilla Prompting
Few-Shot Learning/Prompting
Basic ConceptsProviding K > 1 demonstrations in the prompt to help the model understand patterns.
Zero-Shot Learning/Prompting
Basic ConceptsPrompting with instruction only, without any demonstrations or examples.
One-Shot Learning/Prompting
Basic ConceptsProviding exactly one demonstration in the prompt to help the model understand patterns.
In-Context Learning (ICL)
Basic ConceptsThe model's ability to learn from demonstrations/instructions within the prompt at inference time, without updating weights.
Cloze Prompts
Basic ConceptsPrompts with masked slots for prediction, often in the middle of the text.
Prefix Prompts
Basic ConceptsStandard prompt format where the prediction follows the input.
Templating (Prompting)
Basic ConceptsUsing functions with variable slots to construct prompts in a systematic way.
Instructed Prompting
Basic ConceptsExplicitly instructing the LLM with clear directions about the task.
Role Prompting
Basic ConceptsAssigning a specific role or persona to the model.
Chain-of-Thought (CoT) Prompting
Reasoning FrameworksEliciting step-by-step reasoning before the final answer, usually via few-shot exemplars.
Zero-Shot CoT
Reasoning FrameworksAppending a thought-inducing phrase without CoT exemplars, like 'Let's think step by step'.
Few-Shot CoT
Reasoning FrameworksCoT prompting using multiple CoT exemplars to demonstrate the reasoning process.
Tree-of-Thoughts (ToT)
Reasoning FrameworksExploring multiple reasoning paths in a tree structure using generate, evaluate, and search methods.
Skeleton-of-Thought (SoT)
Reasoning FrameworksA two-stage approach: first generating a skeleton (outline) and then expanding points in parallel.
Graph-of-Thoughts (GoT)
Reasoning FrameworksExtending Tree-of-Thoughts with more flexible graph structures for complex reasoning.
Least-to-Most Prompting
Reasoning FrameworksBreaking down complex problems into simpler subproblems and solving them sequentially.
Recursion-of-Thought (RoT)
Reasoning FrameworksUsing recursive problem-solving approaches in prompting.
Plan-and-Solve Prompting
Reasoning FrameworksFirst devising a plan to solve the problem, then executing the plan step by step.
Step-Back Prompting
Reasoning FrameworksTaking a step back to ask higher-level questions before solving specific problems.
Program-of-Thoughts (PoT)
Reasoning FrameworksExpressing reasoning as executable programs rather than natural language.
Maieutic Prompting
Reasoning FrameworksUsing a question-driven approach to guide reasoning through self-questioning.
Chain-of-Verification (CoVe)
Reasoning FrameworksGenerating initial responses, then creating and answering verification questions to improve accuracy.
Agent-Based Prompting
Agent & Tool UseAssigning an agent role to the LLM that can use tools, make decisions, and interact with the environment.
ReAct (Reasoning + Acting)
Agent & Tool UseCombining reasoning traces and task-specific actions in an interleaved manner.
MRKL System
Agent & Tool UseModular Reasoning, Knowledge and Language system combining neural language models with symbolic tools.
Program-Aided Language Models (PAL)
Agent & Tool UseReading natural language problems and generating programs as intermediate reasoning steps.
CRITIC
Agent & Tool UseCorrecting with Retrieval and Iterative Tool Interaction and Critique.
TaskWeaver
Agent & Tool UseA code-first agent framework for seamlessly planning and executing data analytics tasks.
Tool-Use Agents
Agent & Tool UseAgents specifically designed to interact with and use external tools effectively.
Code-Based Agents
Agent & Tool UseAgents that primarily operate through code generation and execution.
Generate, Implement, Test, and Modify (GITM)
Agent & Tool UseAn iterative framework for code generation involving generation, implementation, testing, and modification.
Reflexion
Agent & Tool UseLearning from self-reflection and environmental feedback to improve performance on subsequent attempts.
Voyager
Agent & Tool UseA lifelong learning agent with a growing skill library for open-ended exploration.
ToRA (Tool-integrated Reasoning Agent)
Agent & Tool UseIntegrating multiple tools into reasoning processes for mathematical problem solving.
Self-Consistency
Self-Improvement TechniquesGenerating multiple reasoning paths and selecting the most consistent answer.
Self-Correction
Self-Improvement TechniquesModel reviews and revises its own output.
Self-Refine
Self-Improvement TechniquesIteratively refining outputs through self-feedback without additional training.
Self-Verification
Self-Improvement TechniquesHaving the model verify the correctness of its own answers.
Self-Calibration
Self-Improvement TechniquesAdjusting confidence estimates to better match actual accuracy.
Reverse Chain-of-Thought
Self-Improvement TechniquesWorking backwards from conclusions to verify reasoning paths.
Self-Ask
Self-Improvement TechniquesModel asks itself follow-up questions to improve reasoning.
Universal Self-Consistency
Self-Improvement TechniquesApplying self-consistency across different reasoning formats and approaches.
Metacognitive Prompting
Self-Improvement TechniquesEncouraging the model to think about its own thinking processes.
Self-Generated In-Context Learning
Self-Improvement TechniquesModel generates its own examples for in-context learning.
Retrieval-Augmented Generation (RAG)
Retrieval & AugmentationEnhancing LLM responses by retrieving relevant information from external sources.
Demonstration-Search-Predict (DSP)
Retrieval & AugmentationA retrieval technique that searches for demonstrations relevant to the input query.
Iterative Retrieval Augmentation
Retrieval & AugmentationMultiple rounds of retrieval and generation for complex tasks.
Interleaved Retrieval-Guided Chain-of-Thought
Retrieval & AugmentationCombining retrieval with chain-of-thought reasoning in an interleaved manner.
Implicit RAG
Retrieval & AugmentationRetrieval-augmented generation where the retrieval process is implicit and automatic.
Verify-and-Edit
Retrieval & AugmentationUsing retrieval to verify and correct generated content.
Cross-File Code Completion Prompting
Retrieval & AugmentationUsing information from multiple files for code completion and generation.
Retrieved Cross-File Context
Retrieval & AugmentationRetrieving relevant context from multiple files to inform code generation.
Automated Prompt Optimization
Prompt OptimizationUsing algorithms to automatically improve prompt effectiveness.
Automatic Prompt Engineer (APE)
Prompt OptimizationAutomatically generates and optimizes prompts for a given task.
GRIPS
Prompt OptimizationGradient-based prompt search for optimization.
Continuous Prompt Optimization
Prompt OptimizationOptimizing prompts in continuous vector spaces rather than discrete text.
Discrete Prompt Optimization
Prompt OptimizationOptimizing prompts at the discrete token level.
Hybrid Prompt Optimization
Prompt OptimizationCombining continuous and discrete optimization approaches for prompts.
Soft Prompt Tuning
Prompt OptimizationLearning continuous prompt embeddings while keeping the model frozen.
RLPrompt
Prompt OptimizationUsing reinforcement learning to optimize prompts based on task performance.
Foundation Model-Based Optimization
Prompt OptimizationUsing large language models themselves to optimize prompts.
Genetic Algorithm Optimization
Prompt OptimizationApplying genetic algorithms to evolve better prompts.
Gradient-Based Optimization
Prompt OptimizationUsing gradient information to optimize prompt effectiveness.
3D Prompting
Multimodal TechniquesIncorporating 3D spatial information and models into prompts.
Audio Prompting
Multimodal TechniquesUsing audio inputs as part of the prompt context.
Image Prompting
Multimodal TechniquesIncorporating images as part of the prompt to guide model outputs.
Video Prompting
Multimodal TechniquesUsing video content as context for generating responses.
Chain-of-Images
Multimodal TechniquesUsing sequences of images to guide reasoning processes.
Multimodal Chain-of-Thought
Multimodal TechniquesCombining reasoning over text and images in a step-by-step manner.
Multimodal Graph-of-Thought
Multimodal TechniquesExtending graph-of-thought reasoning to multimodal inputs.
Multimodal In-Context Learning
Multimodal TechniquesLearning from multimodal examples provided in context.
Image-as-Text Prompting
Multimodal TechniquesConverting images to textual descriptions for text-based models.
Negative Prompting for Images
Multimodal TechniquesSpecifying what should not appear in generated images.
Paired Image Prompting
Multimodal TechniquesUsing pairs of related images to guide reasoning or generation.
AlphaCodeium
Specialized Application TechniquesAdvanced code generation system using iterative refinement and testing.
Code Generation Agents
Specialized Application TechniquesAgents specialized for generating and refining code.
Structured Chain-of-Thought (SCoT)
Specialized Application TechniquesApplying structured reasoning to specific domains like mathematics.
Tab-CoT
Specialized Application TechniquesChain-of-thought reasoning for tabular data analysis.
Chain-of-Table
Specialized Application TechniquesStructured reasoning over tabular data with explicit table operations.
DATER
Specialized Application TechniquesDate and time reasoning for temporal question answering.
LogiCoT
Specialized Application TechniquesLogic-focused chain-of-thought for logical reasoning tasks.
MathPrompter
Specialized Application TechniquesPrompting techniques specialized for mathematical problem solving.
Chain-of-Code
Specialized Application TechniquesCombining natural language reasoning with code execution for problem solving.
Modular Code Generation
Specialized Application TechniquesBreaking down code generation into modular components.
Flow Engineering
Specialized Application TechniquesDesigning structured workflows for complex task completion.
Test-Based Iterative Flow
Specialized Application TechniquesUsing testing to guide iterative improvement in workflows.
Boomerang Task Delegation
Multi-Agent Systems & Team FrameworksA hierarchical task decomposition pattern where complex requests are broken into subtasks, delegated to specialized modes, and their results 'boomerang' back for integration.
Mode-Based Agent Specialization
Multi-Agent Systems & Team FrameworksOrganizing AI systems into specialized operational modes, each with distinct capabilities, roles, and system prompts optimized for specific types of tasks.
Semantic Guardrails
Multi-Agent Systems & Team FrameworksMode-specific validation mechanisms that monitor AI outputs for semantic drift, ensuring responses align with expected behavior and role-appropriate content.
Task Boundary Enforcement
Multi-Agent Systems & Team FrameworksImplementing strict schemas and validation to prevent errors from propagating between tasks in multi-agent systems through immutable inputs and sanitized outputs.
Error Pattern Libraries
Multi-Agent Systems & Team FrameworksCommunity-maintained repositories of common AI system errors, their causes, reproduction steps, and correction strategies to enable systematic learning from failures.
Workflow Template Prompting (.mdc Pattern)
Multi-Agent Systems & Team FrameworksUsing structured markdown templates with YAML frontmatter to create reusable, configurable AI assistant workflows that work across different AI platforms.
AI Assistant Rule Systems
Multi-Agent Systems & Team FrameworksImplementing structured rule hierarchies with global and project-specific configurations to guide AI assistant behavior consistently.
Automated Development Workflows
Multi-Agent Systems & Team FrameworksStructured prompting patterns for common development tasks like commits, PR reviews, issue analysis, and code quality checks.
MCP Server Integration Patterns
Multi-Agent Systems & Team FrameworksPrompting techniques for integrating and orchestrating Model Context Protocol servers to extend AI capabilities with external tools and services.
GitHub Integration Prompting
Multi-Agent Systems & Team FrameworksStructured approaches for AI assistants to interact with GitHub repositories, issues, PRs, and project management through systematic research and action patterns.
Agent Configuration Management
Multi-Agent Systems & Team FrameworksSystematic approaches to managing AI agent configurations, including global settings, project-specific rules, and environment-specific adaptations.
Multi-Perspective Analysis
Multi-Agent Systems & Team FrameworksAnalyzing problems or solutions from multiple distinct viewpoints or roles to ensure comprehensive coverage and identify blind spots.
Structured Commit Workflow
Multi-Agent Systems & Team FrameworksSystematic approach to creating well-formatted commits with conventional commit messages, semantic typing, and automated validation steps.
Five Whys Root Cause Analysis
Multi-Agent Systems & Team FrameworksSystematic questioning technique that asks 'Why?' iteratively to drill down from symptoms to root causes of problems.
Visual Documentation Generation
Multi-Agent Systems & Team FrameworksAutomated creation of diagrams, flowcharts, and visual documentation from code structure, data models, or process descriptions.
Context Priming
Multi-Agent Systems & Team FrameworksSystematic technique for loading comprehensive project understanding by analyzing key files, structure, and conventions before performing tasks.
Meta-Prompt Improvement
Multi-Agent Systems & Team FrameworksSystematic approach for continuously improving AI assistant prompts and rules based on emerging patterns, feedback, and performance metrics.
Browser Automation Prompting
Multi-Agent Systems & Team FrameworksStructured patterns for automating web browser interactions, including element selection, timing management, and error handling strategies.
Comprehensive Code Analysis
Multi-Agent Systems & Team FrameworksMulti-faceted code inspection methodology covering knowledge graphs, quality metrics, performance, security, architecture, and test coverage.
Automated Screenshot Documentation
Multi-Agent Systems & Team FrameworksSystematic capture of application states and UI elements for documentation, testing, and visual verification purposes.
Action-Selector Pattern
Secure Agent ArchitecturesA security pattern where an agent can trigger pre-defined actions but is sandboxed from their outputs. This prevents feedback loops where tainted data from a tool's output could influence subsequent actions.
Plan-Then-Execute Pattern
Secure Agent ArchitecturesAn agent generates a complete, static plan of action (e.g., a sequence of tool calls) *before* any exposure to untrusted input. This plan is executed without modification, preventing runtime deviations based on tainted data.
LLM Map-Reduce Pattern
Secure Agent ArchitecturesA pattern where a primary coordinating agent delegates the processing of multiple pieces of untrusted data to isolated, single-purpose sub-agents (the 'map' step). The results are then aggregated in a sanitized, structured format (the 'reduce' step).
Dual LLM Pattern
Secure Agent ArchitecturesA security architecture using two LLMs: a 'privileged' LLM that can access tools and sensitive data, and a 'quarantined' LLM that handles all untrusted user input. The privileged LLM is never exposed to untrusted content.
Code-Then-Execute Pattern (CaMeL)
Secure Agent ArchitecturesAn advanced pattern, often seen as an evolution of the Dual LLM pattern, where a privileged LLM generates code in a secure, sandboxed Domain-Specific Language (DSL). This DSL defines the workflow and data flow, allowing for rigorous analysis and 'taint tracking' of untrusted data.
Context Minimization Pattern
Secure Agent ArchitecturesA security tactic where potentially malicious user input is deliberately removed from the LLM's context window at a strategic point in the workflow. This severs the causal link between a potential injection attempt and subsequent actions.
Universal Adversarial Triggers
Secure Agent ArchitecturesAttack technique using carefully crafted token sequences that cause models to produce harmful outputs regardless of input context.
Role-Playing Jailbreaks
Secure Agent ArchitecturesAttack method where users instruct the model to take on fictional personas or characters that are not bound by the model's safety guidelines.
Multi-Turn Jailbreaks
Secure Agent ArchitecturesGradual manipulation technique where attackers build trust and slowly escalate requests across multiple conversation turns to bypass safety measures.
Instruction Hierarchy Attacks
Secure Agent ArchitecturesExploitation of conflicts between system instructions and user inputs, where attackers try to override system-level safety instructions with user-level commands.
Constitutional AI Defense
Secure Agent ArchitecturesDefense technique that trains models using a set of principles (constitution) to self-correct harmful outputs and maintain alignment with human values.
Adversarial Training Defense
Secure Agent ArchitecturesTraining technique that exposes models to adversarial examples during training to improve robustness against attacks and jailbreaks.
Response Prefilling
Prompt Structure & EngineeringStarting the model's response with a specific string to guide its output format and content.
XML Tagging / Structured Prompting
Prompt Structure & EngineeringUsing XML tags to clearly separate different parts of the prompt (instructions, data, examples) to prevent confusion.
System Prompts
Prompt Structure & EngineeringUsing the system role to set the overall behavior, persona, and constraints of the model before user interaction.
Context Engineering
Prompt Structure & EngineeringStrategically organizing and optimizing the information provided in the context window to maximize relevance and model comprehension.
No techniques found matching your search.