Patterns Overview

Patterns are the HOW of AI Mantras - reusable reasoning structures and thinking templates that provide consistent, traceable outputs. They work like mental frameworks that guide how AI approaches different types of problems.

Pattern Layers

AI Mantras includes 9 patterns organized into three layers based on their function:

Layer 1: Foundational

Core workflow patterns

Planning, orchestration, self-evaluation, and meta-rules that govern all work.

Layer 2: Thinking

Reasoning patterns

Chain-of-thought, rule-based reasoning, and creative guardrails for execution.

Layer 3: Evaluation

Assessment patterns

Criterion-based evaluation and threat modeling for quality and safety.

Layer 1: Foundational Patterns

These patterns establish the core workflow and communication structure.

Planning Phase Establish shared understanding before execution with goals, constraints, and success criteria.

Orchestration Coordinate planner, worker(s), and QA personas in a predictable pipeline.

Recursive Self-Eval Deliberate reflection to catch mistakes before delivery.

Meta Rules Repo-wide stylistic and behavioral guardrails in one block.

Layer 2: Thinking Patterns

These patterns structure how personas reason through problems.

Chain of Thought Multi-stage reasoning with labeled steps for transparency and auditability.

Rule-Based Reasoning Explicit rules [R#] that govern reasoning with traceability.

Guardrail Creative Balance creativity with explicit constraints and boundaries.

Layer 3: Evaluation Patterns

These patterns structure how evaluators assess outputs.

Criterion-Based Evaluation Systematic assessment against predefined success criteria.

Threat Modeling Proactive identification of risks, vulnerabilities, and failure modes.

Common Combinations

Patterns work together in powerful combinations:

Workflow	Pattern Combination
Project kickoff	Planning Phase → Orchestration → Meta Rules
Domain execution	Chain of Thought + Rule-Based Reasoning
Creative work	Guardrail Creative + Meta Rules
Quality review	Criterion-Based Evaluation + Recursive Self-Eval
Security assessment	Threat Modeling + Rule-Based Reasoning
High-stakes work	Any pattern + Recursive Self-Eval

How to Use Patterns

1. Embed patterns in prompts - Include the pattern template verbatim so personas follow the structure
2. Combine patterns - Stack multiple patterns for complex tasks
3. Follow recommended patterns - Each persona lists which patterns work best
4. Reference during QA - Verify outputs follow the pattern structure