Patterns

Patterns are composable designs that connect primitives into reliable workflows. Each pattern solves a recurring problem in agentic system design -- from enforcing execution contracts to managing delegation hierarchies.

Execution Patterns

Pattern	Purpose
Closed Loop Execution	Build-review-fix cycles that converge on quality
Closed Loop Systems Win	Why feedback loops outperform open-loop approaches
Review Fix Closed Loop	Specific implementation of the review-fix cycle
Phase Gate	Quality checkpoint between workflow phases
Phase Gate with Artifact Extraction	Phase gates that capture outputs as artifacts
Plan as Prompt	Using the plan itself as the execution prompt

Delegation Patterns

Pattern	Purpose
Agent Delegation Hierarchy	Structured chain of command for multi-agent systems
Reduce and Delegate	Break work down, then hand pieces to specialists
Scout Swarm	Parallel scouts that explore before the main agent acts
Stakeholder Trifecta	Three-role pattern for balanced decision making
Work Claim	Agents claim work items to prevent conflicts

Enforcement Patterns

Pattern	Purpose
Classification Gate	Route work by type before processing
Hook Enforced Contract	Use hooks to guarantee contract compliance
Hook Event Selection	Choose the right lifecycle event for each hook
Layered Enforcement	Multiple enforcement layers for defense in depth
Template Declared Contracts	Templates that declare their own requirements

Prompt Patterns

Pattern	Purpose
Adaptive Template Activation	Templates that activate based on context
Higher Order Prompt	Prompts that generate other prompts
Metaprompt Variable Injection	Injecting variables into metaprompts at runtime
Prompt Template Authoring	Best practices for writing prompt templates
Slash Command Template	Standard structure for slash command definitions
PITER Framework	Purpose-Input-Transform-Evaluate-Report

Composition Patterns

Pattern	Purpose
Skill Composition	Combining skills into compound capabilities
Builder Validator Pair	Paired agents that build and validate
Act Learn Reuse	Execute, extract lessons, apply to future work
Progressive Disclosure	Reveal complexity gradually as needed
Progressive Autonomy Staging	Increase agent autonomy in controlled stages
Domain Expert	Compound of 3-5 command surfaces sharing a knowledge domain

Maintenance Patterns

Pattern	Purpose
Correction Task	Structured approach to fixing agent mistakes
Orphan Reaper	Clean up abandoned or orphaned artifacts
Task Create Script	Automated task creation from templates
Shared Briefing Document	Single source of truth for multi-agent context