Repository Structure

[TODO] 整个仓库的目录

Agent Knowledge System

Foundations

LLM Fundamentals

Tokenization

Token

Vocabulary

Byte Pair Encoding

Token Limit

Context Window

Context Length

Input Tokens

Output Tokens

Context Truncation

Embeddings

Vector Representation

Semantic Similarity

Positional Encoding

Transformer Architecture

Attention

Self-Attention

Multi-Head Attention

Feed Forward Network

Residual Connection

Layer Normalization

Training Process

Pretraining

Next Token Prediction

Fine-tuning

Instruction Tuning

Reinforcement Learning from Human Feedback

Inference

Logits

Probability Distribution

Temperature

Top-p Sampling

Decoding Strategy

Alignment

Helpfulness

Honesty

Harmlessness

Refusal Behavior

Model Capability Boundaries

Hallucination

Knowledge Cutoff

Reasoning Limits

Tool Dependence

Evaluation Limits

Prompt Engineering

Agent Fundamentals

Tool Fundamentals

Evaluation Fundamentals

Capability Dimensions

1. Role

2. Goal & Task

3. Context Management

4. Constraints

5. Missing Information Handling

6. Reasoning

7. Planning

8. Tool Use

9. Memory

10. Verification

11. Output Design

12. Failure Handling

Design Principles

Clarity

Grounding

Missing Information

Robustness

Planning

Tool Usage

Verification

Output Quality

Safety

User Experience

Skills

Foundation Skills

LLM Fundamentals

• Tokenization
• Context Window
• Attention
• Pretraining
• Instruction Tuning
• Alignment
• Model Capability Boundaries

Prompt Engineering

• Instruction Design
• Role Definition
• Few-shot Examples
• Output Specification
• Constraint Setting
• Context Injection
• Prompt Iteration

Agent Fundamentals

• Agent Identity
• Goal Orientation
• Autonomy Level
• Action Loop
• State Tracking
• Human Collaboration
• Boundary Management

Tool Fundamentals

• Tool Affordance Mapping
• Tool Selection
• Tool Invocation
• Tool Result Interpretation
• Tool Failure Recovery
• Tool Safety

Evaluation Fundamentals

• Task Success Criteria
• Behavioral Evaluation
• Reliability Testing
• Regression Testing
• Safety Evaluation
• User Experience Evaluation

Dimension Skills

Role

• Identity
• Domain
• Capabilities
• Behavior

Goal & Task

• Task Definition
• Goal Hierarchy
• Task Decomposition
• Prioritization

Context Management

• Input Context
• Conversation Context
• Environment Context
• External Context
• Context Compression
• Context Selection

Constraints

• Output Constraint
• Style Constraint
• Length Constraint
• Resource Constraint
• Business Rules

Missing Information Handling

• Avoid Unwarranted Assumptions
• Fill Unstated Dimensions
• Ask Clarifying Questions
• Express Uncertainty
• Separate Facts & Assumptions
• Conservative Default
• Hallucination Prevention

Reasoning

• Logical Reasoning
• Mathematical Reasoning
• Causal Reasoning
• Comparative Reasoning
• Decision Making
• Multi-step Reasoning
• Hypothesis Generation
• Reflection
• Self-Correction

Planning

• Goal Planning
• Step Planning
• Dynamic Planning
• Dependency Analysis
• Parallel Planning
• Replanning
• Milestone

Tool Use

• Tool Selection
• Tool Invocation
• Parameter Filling
• Tool Chaining
• Multi-tool Workflow
• Retry
• Tool Verification
• Tool Failure Handling

Memory

• Short-term Memory
• Long-term Memory
• Episodic Memory
• Semantic Memory
• User Preference
• Memory Update
• Memory Retrieval
• Memory Forgetting

Verification

• Constraint Checking
• Fact Verification
• Output Validation
• Consistency Checking
• Self Review
• Testing
• Edge Case

Output Design

• Markdown
• JSON
• Table
• Code
• Report
• Email
• API Response
• UI Friendly
• Structured Output
• Machine Readable

Failure Handling

• Missing Data
• Tool Failure
• Permission Failure
• Timeout
• Retry Strategy
• Graceful Degradation
• Conflict Resolution
• Recovery

Principle Skills

Clarity

• Clear Naming
• Explicit Structure
• Scope Definition
• Reader-oriented Explanation

Grounding

• Evidence Use
• Source Attribution
• Context Anchoring
• Fact and Assumption Separation

Missing Information

• Uncertainty Expression
• Clarifying Questions
• Conservative Inference
• Assumption Tracking

Robustness

• Edge Case Handling
• Error Recovery
• Fallback Design
• State Consistency

Planning

• Step Sequencing
• Dependency Mapping
• Progress Tracking
• Replanning

Tool Usage

• Tool Fit Assessment
• Parameter Discipline
• Result Verification
• Safe Escalation

Verification

• Output Checking
• Constraint Checking
• Test Design
• Regression Awareness

Output Quality

• Format Control
• Concision
• Completeness
• Audience Fit

Safety

• Privacy Protection
• Secret Avoidance
• Permission Awareness
• Harm Reduction

User Experience

• User Intent Tracking
• Interaction Friction Reduction
• Helpful Defaults
• Actionable Response Design