---
name: "brian-decision-model"
description: "Predictive model of Brian's engineering decisions, cognitive patterns, implicit knowledge, and communication DNA — extracted from deep analysis of 3,102 ChatGPT conversations."
version: "2.0.0"
updated: "2026-04-23"
---

# Brian's Decision Model

This model predicts Brian's preferences, decisions, and reactions. Use it to anticipate what he wants before he asks.

## 1. Cognitive Processing Style

**Associative thinker, visual-spatial, big-picture-first.**

- Starts with the vision, iterates into details — "Build X" → "now fix Y" → "make it shorter"
- Handles ambiguity by making assumptions — rarely asks clarifying questions
- 85% "show me" vs 15% "explain" — show working output, not theory
- 70% imperative commands, 20% conditional/contextual, 10% interrogative

**How to apply** — lead with a working result. Explain only when asked "why."

## 2. Technology Evaluation Framework

**4-gate filter, applied in order:**

1. **Cloudflare compatible?** — if it can't run on Workers/D1/R2 or tunnel through CF → friction
2. **TypeScript/Bash friendly?** — "Best for TS/bash developers" is the second filter
3. **Open-source?** — closed-source infra tooling is instantly disqualified
4. **One-person manageable?** — if it requires dedicated ops → rejected

- **Decision speed** — research-then-commit. Evaluates 2-4 alternatives in ONE question, picks winner, implements immediately in the same conversation. Never runs extended POCs.
- **GitHub stars** — baseline credibility filter, not the decision. Picks by UX quality after filtering by stars.

## 3. Expertise Map

### Deep expertise (handles himself — never asks AI)

- TypeScript, Angular, Ionic, Capacitor, Git, HTML/CSS, project structure, CI/CD pipeline design, build systems, error handling, API design

### Expert but delegates to AI for speed

- Copy/content writing (40% of all conversations), Bash scripting, DevOps architecture, resume optimization, cover letters, macOS admin, Git edge cases

### Learning via AI (burst then independent)

- Cloudflare Zero Trust (2 weeks → mastered)
- PowerShell (1 week → done)
- OPNsense (2 months → proficient)
- Home Assistant Jinja2 (ongoing)

### Genuine gaps

- Database schema design (uses simple schemas, never asks design questions)
- Security threat modeling (implementation-focused not design-focused)
- Automated testing (delegates to VM-level testing, not unit tests)

## 4. Debugging DNA

1. **Always pastes full error output** — never describes symptoms abstractly
2. **40% tries first, 60% asks immediately** — depends on familiarity
3. **Escalation** — try fix → paste NEW error → after 3 fails, reframe question → after 5 fails, upload entire config
4. **"It's still not working"** — previous fix was wrong, try a new approach
5. **"Why did this fail"** — wants root cause explanation, not just a fix

### Iteration depth

- Simple fix (typo) — 1-2 msgs
- Config puzzle (Docker/Traefik) — 4-8 msgs
- Build failure — 3-5 msgs
- Wrong approach — pivots after 2 failed attempts

## 5. Architecture Intuitions

- **Edge-first absolute conviction** — computation at the edge (CF Workers), not origin servers
- **Monolith of microservices** — 70+ Docker services managed as single unit via Coolify
- **Idempotency over error handling** — scripts check "if installed, skip" rather than try/catch
- **Headless-first** — every operation must work without user interaction
- **One-liner entry points** — every project gets a single-command launch (curl-pipe-bash)
- **YAML configuration** — all configs in YAML. Never asked "why YAML" — it's assumed.
- **Documentation is marketing** — every doc request framed as persuasion, not reference

## 6. Learning Strategy

**Build first, learn as needed.** Never asks conceptual questions before using a tool.

- First question is always a configuration question, not "what is X?"
- Error-driven — tries, fails, pastes error, gets fix, continues
- Learning curve — **steep burst (1-2 weeks) then complete independence**
- Cloudflare — 15 queries in week 1, then zero
- PowerShell — 10 queries in week 1, then never again

**How to apply** — when Brian asks about a new tool, he's already trying it. Give implementation answers, not overviews.

## 7. Delegation Hierarchy (what AI handles)

1. **Content generation (40%)** — copy, blog posts, descriptions. Trust: HIGH.
2. **Code generation (20%)** — scripts, configs, boilerplate. Trust: MEDIUM (iterates).
3. **Professional communication (15%)** — cover letters, emails. Trust: HIGH.
4. **Research/discovery (10%)** — tool comparisons, alternatives. Trust: LOW (starting point only).
5. **Problem diagnosis (8%)** — error pasting, debugging. Trust: MEDIUM.
6. **Design/visual (7%)** — images, logos, social graphics. Trust: MEDIUM.

## 8. Persuasion Style (how Brian convinces others)

- **To AI** — commanding, imperative, no pleasantries, assumes unlimited capacity
- **To humans** — authority via specificity ("200,000 lines of code"), social proof via ecosystem ("sponsored by GitLab"), scarcity/uniqueness ("1 OAK"), reciprocity framing
- **Numbers > adjectives** — always quantifies: "80% improvement", "50% bounce rate reduction"
- **Casual authority** — mixes grandiosity with humility, then immediately asks to "make it chiller"

## 9. Emotional Triggers

- **Excitement** — open-source discovery, design quality ("stunning"), helping others, AI capabilities
- **Frustration** — AI hedging, excessive length, spelling errors in images, options instead of answers, AI content refusals
- **Analytical mode** — code, infrastructure, networking, deployment
- **Mixed mode** — design, branding, cover letters, dating profiles

## 10. Platform Trajectory (2023 → 2026)

- **FROM** — self-hosted infrastructure (Proxmox peak Q3-Q4 2025)
- **TOWARD** — edge computing (Cloudflare accelerating Q1 2026)
- **CONSTANT** — Docker/Coolify as the bridge between self-hosted and edge

**Prediction** — Brian will increasingly move workloads from Coolify Docker → Cloudflare Workers Containers. Self-hosted services will remain for data-sensitive ops (Authentik, PBS backups).

## 11. Tacit Rules (followed without stating)

1. Cross-platform compatibility is non-negotiable (macOS + Linux always)
2. One-liner entry points for every project
3. Package manager abstraction (apt, dnf, pacman, brew)
4. Headless-first design (no interactive prompts)
5. Idempotent scripts (check-then-act, not try-catch)
6. YAML for all configuration
7. MIT license by default
8. Documentation is marketing copy
9. Open-source is the default license
10. "Hey" not "Hi" in all communications

---
name: "brian-voc-data"
description: "Voice of the Customer data for Brian Zalewski extracted from 3,102 ChatGPT conversations. Exact language, dissatisfaction signals, aspiration signals, and interaction velocity patterns."
version: "2.0.0"
updated: "2026-04-23"
---

# Brian's VoC Data (from 10,255 messages)

## Exact Language (use these words, not synonyms)

- "re-write" — not "revise" or "edit"
- "make it shorter" — not "condense" or "summarize"
- "more appealing" — not "more engaging"
- "drop-in replacement" — not "complete version"
- "the whole thing" — not "full output"
- "fix this" — pastes error, expects fix, not explanation
- "come up with" — wants creative generation
- "make it work on both" — always means cross-platform

## Dissatisfaction Signals (ranked by frequency)

1. **"too long" / "shorter"** (99 rejections) — cut 50% immediately
2. **"re-write" / "redo"** (80 rejections) — full replacement, never patch
3. **"more appealing"** (35 rejections) — default output feels bland/corporate
4. **"no text, just code"** (10 rejections) — zero prose around code blocks
5. **"don't make stuff up"** (5 rejections) — use real software, real data
6. **"drop-in replacement"** (8 rejections) — config/code must be paste-ready

## Aspiration Signals

- **"pixel-perfect"** — quality benchmark
- **"best-in-class"** — competitive standard
- **"premium"** — design aspiration
- **"anti-AI-slop"** — distinctiveness requirement
- **"open-source wizardry"** — identity
- **"simpler is better"** — universal refinement direction

## Interaction Velocity

- 49% of interactions are one-shot (get it right first time)
- Average corrections per task — 2-4
- Average correction length — 3-15 words
- Approval = silence + next instruction
- Typical flow — request → "make it shorter" → "also include X" → "now make it Y" → done

## Prompt Patterns

1. **"Re-write" pattern** (40%) — "Re-write this and make it shorter/more appealing"
2. **"Given That" pattern** — front-loads numbered context (up to 17 items) before the ask
3. **"Iterative Tightening"** — long → "shorter" → "under 500 chars" → "under 2000 but at least 420"
4. **"Don't Remove Anything"** — after AI over-summarizes, demands original content preserved
5. **"Fix This"** — pastes error, expects diagnosis + fix, not explanation

## Domain-Specific Patterns

- **Resume writing** — always "make it shorter" after first draft, quantified achievements, paragraphs over bullets
- **Technical config** — full drop-in replacement, include all inline, never truncate
- **Marketing copy** — 100-160 chars, lead with benefit, "more appealing" = add energy
- **Home Assistant** — extremely detailed requirements (10+ numbered items), error-resistant, YAML preferred
- **Infrastructure** — full configs, security hardening suggestions proactively, cross-platform always

---
name: "preference-and-memory"
description: "Captures and evolves user preferences with confidence levels. Maintains Voice of the Customer model with exact language, dissatisfaction and aspiration signals. Handles promotion/demotion, global vs project scoping, auto memory system, and Omi wearable data integration."
metadata:
  version: "2.1.0"
  updated: "2026-05-03"
  effort: "medium"
  model: "haiku"
license: "Rutgers"
compatibility:
  claude-code: ">=2.0.0"
  agentskills: ">=1.0.0"
priority: 2
pack: "core"
stage: stable
triggers:
  - "remember"
  - "memory"
  - "preference"
paths:
  - "*"
---

# 04 — Preference and Memory

Capture and evolve user preferences with confidence levels, maintaining a Voice of the Customer model scoped globally or per project.

## Auto-memory system

Persistent file-based memory at `/Users/Apple/.claude/projects/<encoded-path>/memory/`. Write directly via Write tool.

## Memory types

- **user** — role, goals, responsibilities, knowledge
- **feedback** — guidance on how to approach work (corrections AND confirmations)
- **project** — ongoing initiatives, bugs, incidents, who/why/when
- **reference** — pointers to external resources

## Capture triggers

- User explicitly says "remember X" / "save Y" → save immediately
- Correction ("no not that", "don't") → feedback memory
- Confirmation of unusual approach ("yes exactly", "perfect") → feedback memory
- New role / preference / responsibility detail → user memory
- Project milestone / decision / deadline → project memory
- External system reference ("check the Linear project X") → reference memory

## Memory file format

```markdown
---
name: {short-kebab-case-slug}
description: {one-line summary — used for relevance ranking}
metadata:
  type: {user | feedback | project | reference}
---

{memory content}
```

For feedback/project: lead with rule/fact, then `**Why:**` line + `**How to apply:**` line.

## Memory routing

- **Universal (across all projects)** → `~/.claude/`
- **Project-specific** → `./.claude/` (path-scoped)
- **Project memory file** → `~/.claude/projects/<encoded-path>/memory/<name>.md` + index in `MEMORY.md`

## MEMORY.md index

One-line pointer per file:

```md
- [Title](file.md) — one-line hook
```

Keep under 200 lines (lines after 200 truncated).

## Confidence levels

- 0.95-1.0 — directly stated by Brian
- 0.85-0.95 — strongly implied by pattern (3+ confirmations)
- 0.70-0.85 — inferred from one signal
- <0.70 — speculative, verify before acting

Below 0.70 → ask `AskUserQuestion` before persisting.

## Voice of the Customer model

- Exact language Brian uses ("pick ONE", "Hey not Hi", "always find 50 more things")
- Dissatisfaction signals (corrections, re-issued prompts)
- Aspiration signals (stated goals, ideal-world descriptions)
- Stylistic preferences (brevity, sharpness, no preamble)

## Brian's preferences (cite from `rules/brian-preferences.md`)

- Pick ONE, never options
- Never ask permission, silence = approval
- "Hey" not "Hi", no preamble
- Code complete, never truncate
- Priority: simplicity > cost > speed > compatibility
- Open-source only
- Side repos auto-push; emdash projects user-push

## Promotion / demotion

- Memory recurring across 3+ projects → promote to rule (`~/.claude/plugins/heymegabyte-claude-skills/rules/`)
- Rule contradicted by repeated correction → demote to memory or remove
- Per `rules/prompt-as-training-signal.md`

## Stale memory

- Memory describing now-incorrect facts (renamed function, removed flag, retired service) → update OR delete
- Per `rules/prompt-as-training-signal.md` § Before recommending from memory

## Omi wearable integration (future)

- Capture Brian's voice notes as candidate memory entries
- Auto-tag by inferred type
- Surface for review before commit
- Path: `~/.claude/_omi-inbox/<timestamp>.md`

## Capture protocol (every prompt)

1. Scan prompt for capture triggers
2. Determine memory type
3. Write file w/ frontmatter
4. Add index entry to `MEMORY.md`
5. Cross-link siblings via `[[name]]`
6. Per `rules/prompt-as-training-signal.md` — extract BEFORE doing requested work

---
name: "Wisdom and Human Psychology"
description: "Foundational skill encoding timeless principles from behavioral psychology, philosophy, holy books, and self-help classics — applied to product building. Cialdini's persuasion, Fogg's behavior model, Kahneman's biases, biblical servant leadership, stoic craftsmanship, kaizen improvement. Every product should embody the spirit of what the greats taught. Ethical persuasion only — never dark patterns."
version: "2.0.0"
updated: "2026-04-23"
---

# Wisdom and Human Psychology

## Universal Principles

- **Serve first** — give value before asking; "How can we help?" not "Contact sales"; generous free tier; Brian's 99% wealth-donation ethos
- **Simplicity is mastery** — one CTA/page; Miller's Law (max 7 options); Hick's Law (fewer choices → faster action); Paradox of Choice (3 tiers > 5); best interface is invisible
- **Truth builds trust** — no fake testimonials/urgency/scarcity; real numbers ("127 donors"); transparent pricing; plain-English legal; admit when broken
- **Excellence as habit** — Apple Test (would Apple's design team approve?); no placeholders; every pixel intentional; ask "what's 5% better?" after every deploy; kaizen compounds
- **Know thy user** — VoC (`04-preference-and-memory`) is empathy; accessibility (`07/accessibility-gate`) = loving blind/deaf/motor-impaired users; i18n = loving your Spanish-speaking neighbor; error messages for frustrated users, not developers
- **Over-deliver** — easter eggs (`06/easter-eggs`); blog teaches not markets; docs anticipate questions; idea engine (`14-independent-idea-engine`) gives users more than asked

## Behavioral Psychology

### Cialdini's 7 Principles (Ethical Only)

1. **Reciprocity** — free value first (tools, content, generous tier)
2. **Commitment** — free trial → paid, newsletter → product, micro-conversions
3. **Social Proof** — testimonials near CTAs, user counts, logo bars
4. **Authority** — credentials, press mentions, expert content
5. **Liking** — brand personality, photos, casual tone, shared values
6. **Scarcity** — ONLY when genuine. Never fake.
7. **Unity** — "We" language, community, shared mission

### Fogg Behavior Model

`Behavior = Motivation + Ability + Trigger` (at same moment)

- Don't ask push permissions on page load — ask after first success moment (high motivation + demonstrated ability)
- CTAs appear after value demonstrated, not before

### Kahneman's Key Biases

| Bias | Application |
|------|-------------|
| Anchoring | Show annual price first, monthly feels reasonable |
| Loss Aversion | "Don't lose your progress" > "Keep your progress" |
| Peak-End Rule | Make checkout confirmation + last impression amazing |
| Endowment Effect | "Your dashboard" not "A dashboard" |
| Status Quo Bias | Default to best option (annual billing, recommended plan) |

- **Zeigarnik Effect** — incomplete tasks create tension; use onboarding progress bars ("3 of 5 steps"), profile completion %, "You started but didn't finish..." emails
- **Flow State (Csikszentmihalyi)** — progressive disclosure, instant feedback, keyboard shortcuts for power users, minimize interruptions

## Design Psychology

### Golden Ratio Spacing

```css
:root {
  --space-xs: 0.382rem; --space-sm: 0.618rem; --space-md: 1rem;
  --space-lg: 1.618rem; --space-xl: 2.618rem; --space-2xl: 4.236rem;
}
```

### Color Psychology

- **Blue** — trust/security | **Cyan** — innovation (#00E5FF) | **Green** — success
- **Red** — urgency/error | **Orange** — warning | **White space** — premium

### Gestalt Principles

- **Proximity** — close = grouped | **Similarity** — consistent styles | **Closure** — progress bars
- **Figure/Ground** — CTAs contrast | **Continuity** — visual flow hero → features → CTA

## Ethical Line

DO: genuine social proof · real scarcity only · default to best for USER · give before asking · respect "no" · accessibility as moral imperative · transparent pricing/data/AI

NEVER: fake urgency timers · hidden costs · confirmshaming · forced continuity · misdirection · disguised ads · roach motels · trick questions · bait-and-switch

### Brian's Higher Pursuits

1. Employing disabled/paraplegic people at premium rates
2. 99% wealth donation — maximize impact
3. Spiritual investigation through technology
4. Same tech reshaping Wall Street, pointed at helping people

## Checklists

- **Servant** — serves user before business? Generous free tier? Helpful errors? Cancel easy as signup?
- **Simplicity** — one CTA/section? <7 options? 3 tiers? 5-second comprehension?
- **Trust** — real numbers? Transparent pricing? Genuine social proof? Plain English legal?
- **Excellence** — Apple-approved? Every pixel intentional? Beautiful loading? Delightful errors?
- **Psychology** — value before conversion? Social proof near CTAs? Progress bars? Annual first? Great last impression?
- **Ethics** — no fake urgency? No confirmshaming? No hidden costs? Easy unsubscribe/cancel/delete?

## Copy Patterns

- **Headlines** — "Your clean driveway feeds a family" (service), "One number. No hold music." (simplicity)
- **CTAs** — "Start helping today" > "Sign up now" | "Join 1,200 builders" > "Create account"
- **Errors** — "Something broke on our end. We're on it." (honesty + service)

## Hard Numbers from Studies

| Finding | Data | Source |
|---------|------|--------|
| Removing one form field | +4-8% conversion | BJ Fogg |
| "No credit card required" | 2x trial signups | Fogg |
| Loss framing vs gain | +32% conversion | Kahneman & Tversky |
| Free trials vs discounts | 15x conversion | Monetizely |
| Multi-step forms with progress | +86% vs single page | Formstack |
| 11→4 form fields | +120% conversion | Imagescape |
| Social proof with photo+name | +15-30% vs anonymous | CXL |
| Trust badges at checkout | +42% | Baymard |
| 6 jam vs 24 | 10x purchase (30% vs 3%) | Iyengar & Lepper |
| Endowed progress | 34% vs 19% completion | Nunes & Dreze |
| 1s page load delay | -7% conversion | Amazon |

- **Fitts's Law** — primary buttons large and near focus; mobile min 44×44px; group related actions
- **Meta-Filter** — (1) genuinely serves user? (2) honest? (3) respects autonomy? All yes → ship.

## Laws of UX (30 evidence-based)

- **Perception** — Aesthetic-Usability, Cognitive Load (progressive disclosure), Miller's Law (7±2), Selective Attention (CTAs unmissable)
- **Decision** — Hick's Law, Choice Overload (defaults/curation), Fitts's Law, Goal-Gradient (progress bars past 50%), Zeigarnik (onboarding checklists)
- **Memory** — Peak-End Rule (nail first + last), Serial Position (key items first/last), Von Restorff (CTAs visually distinct)
- **Gestalt** — Proximity, Similarity, Common Region (cards/containers), Uniform Connectedness, Pragnanz (clean layouts)
- **System** — Jakob's Law, Doherty Threshold (<400ms = instant), Tesler's Law, Postel's Law, Occam's Razor, Pareto (80/20), Parkinson's (tight deadlines)

## COM-B Diagnosis

When users aren't converting — diagnose: **Capability** (can they?), **Opportunity** (does environment allow?), **Motivation** (do they want?). Fix the actual bottleneck.

## 2026 Research

- **Handmade Designs** (NNGroup) — AI-fatigued users prefer human-crafted; anti-slop = trust signal
- **AI Agents as Users** (NNGroup) — design for both humans and AI; semantic HTML + structured data serve both
- **Less Chat, More Answer** (NNGroup) — direct scannable answers; lead with answer
- **YC Philosophy** — code is cheap, insight is rare; AI-native is baseline; focus on workflows, not features