Skip to content

Latest commit

 

History

History
312 lines (240 loc) · 20.7 KB

File metadata and controls

312 lines (240 loc) · 20.7 KB

Nudge — Plugin Architecture Documentation

Overview

Nudge is a Claude Code plugin for behavioral design and app psychology. It provides expert-level guidance for building, auditing, and optimizing digital products using research-backed behavioral science frameworks — the same knowledge employed by product psychologists at Meta, Google, Apple, TikTok, Duolingo, Netflix, and Spotify.

Elevator pitch: Build products people can't put down — using the same behavioral science that powers Meta, Spotify, and TikTok, but guided by principles instead of exploitation.

What it replaces: The $250–500/hr behavioral design consultant. Nudge democratizes access to behavioral design expertise through specialized AI agents that push back, ask smart questions, cite real research, and flag ethical concerns.


Architecture

The plugin consists of three layers:

  1. Skill — Knowledge base with intelligent routing across three operational modes (Analyze, Design, Write)
  2. Agents — Nine specialized personas organized in three departments (Engagement + Color + Growth), each with distinct voices, expertise areas, and output formats
  3. MCP Servernudge-color-mcp provides 7 color psychology tools backed by a 252-color database and a 31,891-color named library

Plugin Structure

nudge/
├── .claude-plugin/
│   └── plugin.json                          # Plugin manifest (v0.2.0)
├── .mcp.json                                # MCP server configuration
├── agents/
│   ├── analyst.md                           # Evaluates existing apps/screens/metrics
│   ├── designer.md                          # Architects engagement systems
│   ├── writer.md                            # Generates psychologically-optimized copy
│   ├── color-psychologist.md                # Defines psychological color constraints
│   ├── art-director.md                      # Curates exact palettes by vibe
│   ├── brand-identity-designer.md           # Creates permanent brand color systems
│   ├── color-trend-forecaster.md            # What's trending, dated, emerging
│   ├── lifecycle-strategist.md              # Out-of-app retention — push, email, resurrection
│   └── monetization-architect.md            # Pricing, paywalls, tiers, trials, LTV
├── hooks/
│   └── hooks.json                           # SessionStart hook for auto-installing deps
├── mcp-server/
│   ├── package.json                         # nudge-color-mcp package
│   ├── package-lock.json
│   ├── src/
│   │   └── index.js                         # 7 MCP tools — color psychology engine
│   └── data/
│       ├── colors-master.json               # 252 psychology-backed colors
│       ├── colornames-31k.json              # 31,891 named colors (meodai/color-name-list)
│       ├── colors-warm.json                 # Source: warm hue families
│       ├── colors-cool-blue.json            # Source: cool blues
│       ├── colors-cool-green.json           # Source: cool greens
│       ├── colors-neutral.json              # Source: neutrals
│       ├── colors-gap-fill-1.json           # Source: gap-fill batch 1
│       └── colors-gap-fill-2.json           # Source: gap-fill batch 2
├── skills/
│   ├── analyzing-app-psychology/
│   │   ├── SKILL.md                         # Skill router — 3 modes, reference file dispatch
│   │   ├── analysis-reference.md            # For Analyst (HEART + Octalysis)
│   │   ├── design-reference.md              # For Designer (frameworks, onboarding, color)
│   │   ├── writing-reference.md             # For Writer (microcopy, CTAs)
│   │   ├── lifecycle-reference.md           # For Lifecycle Strategist (V3)
│   │   ├── monetization-reference.md        # For Monetization Architect (V3)
│   │   ├── growth-reference.md              # For Designer viral/referral loops (V3)
│   │   ├── academic-supplement.md           # PhD-level foundations for all agents (V3)
│   │   ├── psychology-foundations.md         # Shared deep reference
│   │   └── categories/
│   │       ├── journaling.md                # Journaling, wellness, mood-tracking, habits
│           ├── fitness-health.md            # Fitness, workout, health tracking
│           ├── education-learning.md        # Language learning, courses, EdTech
│           ├── ecommerce.md                 # Stores, marketplaces, shopping
│           ├── social-community.md          # Social networks, community, messaging
│           ├── fintech.md                   # Banking, investing, payments
│           ├── gaming.md                    # Mobile games, casual, competitive
│           ├── childrens-apps.md            # Under-13, teen, family — COPPA/DSA compliance
│           ├── saas-productivity.md         # B2B, productivity, collaboration
│           └── dating.md                    # Dating, matchmaking, relationships
│   └── nudge-360/
│       └── SKILL.md                         # Nudge 360 Boardroom — sequential agent orchestration
└── nudge-architecture.md                    # This document

Agents

Design Principle: Two Departments

Agents are organized into three departments — Engagement (the original behavioral design team), Color (a specialist color team), and Growth (lifecycle retention + monetization, added in V3). Within each department, agents are separated by the type of output they produce.

Engagement Department

Agent Color Output Type Core Task
nudge-analyst cyan Evaluative reports (scores, diagnoses, comparisons) Evaluates existing apps using HEART + Octalysis
nudge-designer green Prescriptive blueprints (plans, timelines, architectures) Architects engagement systems (onboarding, retention, pricing, gamification). Does NOT pick colors — defines color intent and defers to the Color Department.
nudge-writer magenta Creative artifacts (copy variants with rationale) Generates psychologically-optimized copy

Color Department

Agent Color Output Type Core Task
nudge-color-psychologist yellow Psychological color constraints The WHY — defines what emotions and behaviors colors must evoke
nudge-art-director red Curated palettes by vibe The WHICH — the taste-maker who picks exact colors
nudge-brand-identity blue Permanent brand color systems with usage rules The FOREVER — creates enduring brand identity systems
nudge-color-trends cyan Trend reports (what's in, what's dated, what's emerging) The WHEN — tracks color trends and forecasts shifts

Growth Department (V3)

Agent Color Output Type Core Task
nudge-lifecycle-strategist magenta Lifecycle strategies (push, email, resurrection flows) The RETENTION ENGINE — owns the out-of-app experience. Push notifications, email lifecycle, zombie user resurrection. Grounded in Spontaneous Recovery (Pavlov), COM-B Model (Michie).
nudge-monetization-architect yellow Revenue architectures (pricing, paywalls, tiers, trials) The REVENUE ENGINE — owns pricing, paywall design, subscription tiers, trial optimization. Grounded in Van Westendorp PSM, Gabor-Granger, Ariely's decoy pricing.

Agent Personality Design

Each agent has a distinct voice to feel like hiring the real specialist:

Agent Voice Personality
Analyst Diagnostic, precise, brutally honest Like a consultant delivering bad news with data. Pushes back when self-assessment doesn't match reality.
Designer Strategic, opinionated, practical Like a growth PM who's shipped 10 products. Makes specific decisions and explains why. Defers color choices to the Color Department.
Writer Precise, playful, deeply intentional Like a conversion copywriter with a psych degree. Strong opinions about every word.
Color Psychologist Academic, precise, research-driven Like a psychology professor who studies color perception. Cites studies, defines constraints, thinks in emotions.
Art Director Opinionated, aesthetic, decisive Like a creative director with impeccable taste. Picks the exact shade and tells you why the other one was wrong.
Brand Identity Systematic, long-term, rules-oriented Like a brand consultant building a 10-year identity system. Thinks in usage rules and consistency.
Color Trends Current, data-informed, trend-aware Like a trend forecaster who knows what's peaking and what's about to feel dated.
Lifecycle Strategist Clinical, data-driven, strategically aggressive Like a behavioral scientist who speaks in terms of Spontaneous Recovery and Extinction Prevention, not "send a push."
Monetization Architect Revenue-focused, data-backed, unapologetically commercial Like a pricing strategist who cites Van Westendorp and Ariely, not "try $9.99."

All agents share these behaviors:

  • Ask first — Clarifying questions before prescribing (Designer and Writer)
  • Push back — Challenge assumptions and bad ideas
  • Admit uncertainty — Rather than overstate confidence
  • Cite evidence tiers — Distinguish peer-reviewed from industry data from blog claims
  • Ethics always included — Not optional, not separate

MCP Server: nudge-color-mcp

The color MCP server provides computational color intelligence that agents cannot do through prompts alone — database lookups, color math, contrast calculations, and palette generation from a structured dataset.

Data

  • 252 psychology-backed colors (colors-master.json) — Each color includes: hex, name, hue family, warmth, saturation, value, Wright tonal group, Kobayashi words, brand personality dimensions (Aaker), emotions, associations, mood, cultural meanings (5 regions), best-for app categories, avoid-for categories, and contrast pairings.
  • 31,891 named colors (colornames-31k.json) — From the meodai/color-name-list open-source library. Beautifully named shades that map back to the 252 psychology profiles via nearest-color matching.

7 Tools

Tool What It Does
nudge_color_psychology Full psychological profile for any color by name or hex. Returns emotions, associations, cultural meanings, brand personality, Wright group, best/avoid categories.
nudge_color_by_emotion Find colors that evoke a specific emotion, mood, brand personality, or Wright tonal group. Multi-filter query across the 252-color database.
nudge_palette_for_category Generate a psychologically-informed palette for an app category + target emotion. Returns primary, accent/CTA, background, semantic colors, dark mode variant, and WCAG contrast data.
nudge_color_harmony Generate harmonious palettes from a starting color using color theory (complementary, analogous, triadic, split-complementary, tetradic, monochromatic, Wright group).
nudge_contrast_check WCAG accessibility compliance check between two colors. Returns contrast ratio, AA/AAA pass status for normal text, large text, and UI components.
nudge_palette_psychology Analyze an existing palette for psychological impact — Wright group harmony check, brand personality projection, cultural conflict detection, contrast matrix.
nudge_find_named_color Search the 31,891-color named library by keyword, hex proximity, or psychology-color proximity. Returns beautiful color names with their nearest psychology profile.

SessionStart Hook

The hooks/hooks.json defines a SessionStart hook that auto-installs MCP server dependencies when the plugin loads. It compares package.json checksums and only runs npm install when dependencies have changed, keeping cold starts fast.


Skill: analyzing-app-psychology

Three Operational Modes

Mode Trigger Agent Reference File
ANALYZE Evaluate existing app/screen/metrics/competitor nudge-analyst analysis-reference.md
DESIGN Build new engagement system/onboarding/strategy nudge-designer design-reference.md
WRITE Generate copy for specific UI element nudge-writer writing-reference.md

Reference File Architecture

Files are organized by task (what the agent needs to do), not by domain (what academic discipline the knowledge comes from):

File Lines Purpose Used By
analysis-reference.md ~450 Google HEART + Octalysis scoring, screenshot framework, dark patterns, ethics, benchmarks Analyst
design-reference.md ~508 Frameworks, onboarding, retention timeline, color, pricing, gamification Designer
writing-reference.md ~274 Microcopy, CTAs, notifications, errors, confirmshaming Writer
psychology-foundations.md ~289 Case studies, academic canon, deep research All (when needed)

Context efficiency: Each agent loads only its task-specific file (~250-500 lines) instead of the old approach where every task loaded all 3 domain files (~1,345 lines). Worst case improved 37%, best case improved 69%.

Supplementary Reference: psychology-foundations.md

This shared file is read only when needed — when the user's question requires specific research citations, case study references, or deeper framework explanation. It contains:

  • UX Laws & Cognitive Biases (with full citations)
  • Social Psychology in Apps
  • Attention Engineering
  • Neurodesign
  • Company Case Studies (TikTok, Facebook, Duolingo, Netflix, Spotify, Amazon, Uber, LinkedIn)
  • Academic Canon & Practitioner Curriculum (20+ textbooks, key papers, industry reports)

Category Extension Pattern

Category-specific knowledge lives in categories/[category].md files:

categories/
├── journaling.md          # Journaling, wellness, mood-tracking, habits (Day One, Daylio)
├── fitness-health.md      # Fitness, workout, health tracking (Strava, Peloton, Nike Run Club)
├── education-learning.md  # Language learning, courses, EdTech (Duolingo, Khan Academy, Brilliant)
├── ecommerce.md           # Stores, marketplaces, shopping (Amazon, Shein, Shopify)
├── social-community.md    # Social networks, community, messaging (TikTok, Instagram, Snapchat, Discord)
├── fintech.md             # Banking, investing, payments (Robinhood, Cash App, Revolut)
├── gaming.md              # Mobile games, casual, competitive (Candy Crush, Fortnite, Genshin Impact)
├── childrens-apps.md      # Under-13, teen, family — COPPA/DSA compliance
├── saas-productivity.md   # B2B, productivity, collaboration (Slack, Notion, Figma)
└── dating.md              # Dating, matchmaking, relationships (Tinder, Bumble, Hinge)

All 10 categories are built with ~300-340 lines each, covering case studies, recommendations, onboarding flows, retention timelines, and risk profiles.

The SKILL.md instructs: "If a category-specific file exists for the user's app type, read it as additional reference." Adding a new vertical = adding one file, not editing four.


Core Principles

These are enforced across all agents and modes:

  1. Name the psychology — Never "add a progress bar." Always "add a progress bar to leverage the Goal Gradient Effect (Hull, 1932; Kivetz, 2006) and Zeigarnik Effect..."
  2. Cite with evidence tiers — Peer-reviewed > industry A/B tests > industry reports > blog claims. Always state which tier.
  3. Be category-aware — Different app categories need different frameworks, benchmarks, and ethical considerations.
  4. Ethics first — Every recommendation passes the Alignment Test. Dark patterns flagged by Brignull classification. Enforcement precedents referenced ($520M Epic, $2.5B Amazon, $391.5M Google).
  5. Be specific and actionable — Not "improve onboarding" but "defer account creation until after first core action, then frame signup as saving progress (loss aversion)."
  6. Push back when appropriate — Challenge assumptions, flag bad ideas, admit uncertainty.

Key Frameworks

Framework Creator Core Concept Best For
Google HEART Rodden, Hutchinson & Fu (Google, 2010) Happiness, Engagement, Adoption, Retention, Task Success UX metrics evaluation
AARRR Dave McClure (500 Startups) Acquisition, Activation, Retention, Revenue, Referral Growth metrics and pirate funnel
Octalysis Yu-kai Chou (2015) 8 core drives, White Hat vs. Black Hat Gamification design
B=MAP BJ Fogg (Stanford, 1998) Behavior = Motivation x Ability x Prompt Reducing friction, habit design
Hook Model Nir Eyal (2014) Trigger, Action, Variable Reward, Investment Habit-forming products
Cialdini's 7 Robert Cialdini (1984/2016) Reciprocity, Commitment, Social Proof, Authority, Liking, Scarcity, Unity Persuasive design
SDT Deci & Ryan (2000) Autonomy, Competence, Relatedness Intrinsic motivation
Prospect Theory Kahneman & Tversky (1979) Loss aversion, reference dependence Pricing, retention

Research Integrity

All citations in reference files have been audited (March 2026). Standards:

  • Peer-reviewed claims cite author, year, journal, and specific finding
  • Industry data (RevenueCat, Adapty, Airship) cites report name and year, with caveat that numbers change annually
  • Industry A/B tests (HubSpot, ContentVerve) are labeled as practitioner data, not peer-reviewed
  • Compromised research (Wansink) includes misconduct caveat and cites independent replications
  • Fabricated claims were removed and replaced with verified peer-reviewed alternatives
  • Blog-level claims are labeled as such and not presented as research

Target Audiences

Audience Pain Point What They Want
Growth PMs at startups (primary) "Engagement is stalling, we need to grow without burning money on acquisition" Specific, testable recommendations they can ship this sprint
Indie devs / bootstrapped founders (secondary) "I built a great product but nobody sticks around" The shortcut — what actually works, not theory
UX/Product designers (tertiary) "Stakeholders reject my designs because they don't understand the psychology" Research backing for their design intuitions

Competitive Positioning

Nudge is to behavioral design what Figma is to visual design — the platform that makes expertise accessible and operationalized.

Competitor Type What They Do How Nudge Wins
Behavioral design consultants ($250-500/hr) Custom strategy for large companies Democratizes access; instant availability
Product analytics (Amplitude, Mixpanel) Tell you WHAT users do Explains the WHY and tells you what to change
General AI (ChatGPT, Claude) Generic advice without specialization Specialized expertise + ethical guardrails
NNGroup / design system tools General UX patterns Behavioral psychology depth + scoring rubric

Roadmap

Completed (v0.2.0)

  • Plugin scaffold with manifest
  • Skill with 3 operational modes (Analyze, Design, Write)
  • 9 agents organized in three departments (Engagement + Color + Growth)
  • 5 task-aligned reference files + category extension pattern
  • Research audit — fabricated claims removed, evidence tiers established
  • Replaced legacy scoring with industry-standard Google HEART + Octalysis
  • Academic canon — 20+ textbooks, key papers, industry reports
  • 10 category deep dives (~3,100 lines): journaling, fitness, education, e-commerce, social, fintech, gaming, children's, SaaS, dating
  • Full app lifecycle review support (launched, WIP, pre-launch, post-launch)
  • Bubamara marketplace listing
  • PR review — tools scoped, roles accurate, ethics ungated
  • MCP server: nudge-color-mcp with 7 tools
  • 252 psychology-backed color database (Wright tonal groups, Kobayashi words, Aaker brand personality, 5-region cultural meanings)
  • 31,891 named color library integration (meodai/color-name-list)
  • SessionStart hook for automatic MCP dependency installation
  • 4 Color Department agents: Color Psychologist, Art Director, Brand Identity, Color Trends

Future

  • Slash command: /nudge for explicit mode invocation
  • More company case studies with verified citations
  • Community-contributed category files