How to Use The Forge

Forge UX Patterns is a structured, queryable repository of behavioral design patterns. It serves two audiences: human designers who browse and apply patterns to feature work, and AI agents that query the API to ground their suggestions in behavioral science.

On this page

For DesignersFor AI AgentsPattern LifecycleSchools of ThoughtBehavioral RuntimeDesign Principles

For Designers

The Pattern Catalog is your entry point. Patterns are grouped by lifecycle stage -- canonical patterns are the proven foundation, candidates are ready for team review but not yet fully validated.

Reading a pattern card

School badges -- which behavioral science tradition the pattern draws from

WendelKahnemanAgentic UX

Confidence % -- how battle-tested the pattern is (0-100)

95%

High confidence

70%

Medium

45%

Low

Cognitive load -- how much mental effort the pattern demands from users

Low|Medium|High

Autonomy levels -- which agent autonomy modes the pattern supports

SuggestConfirmAuto

Lifecycle stage -- how mature and validated the pattern is

DraftCandidateCanonicalDeprecated

Ethical guardrail -- the non-negotiable constraint (always visible, always respected)

Agent must cite the specific data source before executing any action on behalf of the user.

Copy prompt -- paste into any Claude conversation for pattern-grounded guidance

When starting a new feature, begin with the CREATE Action Funnel -- it is the diagnostic spine of the entire system. Map your user flow to its six stages (Cue, Reaction, Evaluation, Ability, Timing, Execute), identify where behavior breaks down, then pull in supporting patterns (Strategic Friction, Progressive Disclosure, etc.) to address the specific failure.

For AI Agents

Forge UX Patterns exposes a REST API returning clean JSON. Any agent should query the API before suggesting UI, workflow, or agent behavior changes.

Endpoints

GET /api

API documentation, all endpoints, query params, enum values, and the recommended agent system prompt.

GET /api/pattern/:slug

Single pattern with all populated fields. Canonical patterns return up to 35 fields including execution_model, behavioral_signals, and agent_decision_protocol.

GET /api/patterns

List and filter. Supports: school, tag, autonomy, lifecycle, evidence, impact, force, min_confidence, q, depth.

Recommended system prompt

You are a senior behavioral designer. Before suggesting any UI, workflow, or agent behavior, query Forge UX Patterns (/api/patterns or /api/pattern/[slug]) and reference the applicable pattern(s) by name. Prioritize patterns from the Agentic UX framework for any agent-driven interaction. Always respect the ethical guardrail defined in each pattern.

Key fields for agents

copy_prompt -- a tailored prompt for Claude referencing this specific pattern. Inject into system prompts.
agent_decision_protocol -- triggers, escalation strategy (L1-L5), and a worked example. Follow this when the pattern is relevant.
behavioral_signals -- telemetry conditions to monitor. When a signal fires, enter the escalation ladder at L1.
ethical_guardrail -- non-negotiable. Agents must respect this before acting.

Pattern Lifecycle

Every pattern moves through a lifecycle that reflects how much evidence supports it. Not every field is required at every stage -- the schema supports three depth tiers.

Draft

New pattern proposal. Core fields, description, guardrail, and an example. Lowest barrier to entry.

Candidate

Team-reviewed. Adds behavioral objective, when to use, example, agent rules, and copy prompt.

Canonical

Production validated. All fields populated: execution model, failure modes, KPIs, behavioral signals, decay monitoring, agent protocol.

Deprecated

Decay detected or superseded. Documented reason and link to replacement pattern.

Non-canonical patterns display a maturity checklist showing exactly which fields need to be populated for promotion. This checklist is visible on every candidate and draft pattern detail page.

Schools of Thought

Every pattern traces back to at least one established school of thought. These are the theoretical foundations -- reach for the right school based on the problem you are solving.

Behavioral Economics (Wendel)

CREATE Funnel, Behavioral Plans

High-stakes workflows: payments, deployments, approvals

Tiny Habits (Fogg)

B=MAT, Behavior Grid

Habit-forming features: daily standups, recurring check-ins

Dual-Process (Kahneman)

System 1 / System 2 thinking

Decisions requiring deliberation vs. speed

Human-Centered Design (Norman)

7 Principles, Affordances

Overall UX strategy, onboarding, discoverability

Agentic UX

Autonomy Dials, Explainability, Confidence Signaling

AI-driven analysis, automation, agent-first features

Ethical Design

Anti-pattern catalog, Guardrails

Protecting users from coercion, manipulation, AI hallucinations

The Behavioral Runtime

At canonical depth, patterns become runtime behavioral diagnostics. Agents read telemetry signals, diagnose failures, estimate intervention cost, and escalate through a formalized 5-level ladder. Agents always start at Level 1 and only escalate when the current level is insufficient.

Escalation Ladder

1

Diagnose

none

Map user state to pattern stage. Identify which behavioral gate is failing. No user-visible action.

2

Nudge

micro

Copy change, visual emphasis, timing adjustment. No structural change. Ships without design review.

3

Restructure

feature

Workflow modification, new component, flow reordering. Requires design review and human confirmation.

4

Constrain

feature

Dial back autonomy level. Force confirm_execution. Add Strategic Friction. Requires human approval.

5

Yield

n/a

Agent cannot resolve. Yield to human designer, product manager, or domain expert. Always triggered when confidence drops below threshold.

Rules

  • Agents must not skip levels. L3 requires evidence that L1-L2 were insufficient.
  • L5 (Yield) is mandatory when confidence drops below the pattern's confidence_weight threshold.
  • Levels 3+ require human approval before execution (enforced by Autonomy Dial at confirm_execution minimum).
  • Every level transition is logged to the Audit Trail.

Intervention Costs

micro

Copy, color, timing. No structural change. Agent can execute at suggest_only or higher.

feature

New component, flow change. Requires design review. Agent proposes, human confirms.

architectural

System redesign. Agent diagnoses and recommends, cannot execute. Always escalates to L5.

Design Principles

Six non-negotiable principles. Every pattern traces back to at least one.

Behavior Over Aesthetics

Wendel, Fogg

Design for action, not appearance.

B = MAT

BJ Fogg

Behavior requires Motivation + Ability + Trigger at the same moment.

CREATE Funnel

Wendel

Cue, Reaction, Evaluation, Ability, Timing, Execute.

Strategic Friction

Kahneman

Slow high-stakes decisions to force System 2 thinking.

Ethical Autonomy

2026 Standards

AI must be transparent, reversible, human-led.

Progressive Disclosure

Classic HCI

Reveal complexity only when the user is ready.