behavior-design

$npx mdskill add Mark393295827/third-brain-v5-skills/behavior-design

Design a behavior change system from goals into repeatable habits.

  • Decomposes a goal into minimum habits with concrete triggers.
  • Relies on goal decomposition and habit design principles.
  • Decides based on trigger, action, environment, and review metrics.
  • Outputs a behavior card with trigger, action, cue, reward, and fallback.

SKILL.md

.github/skills/behavior-designView on GitHub ↗
---
name: behavior-design
description: Design a behavior change system — decompose a goal into minimum habits, define triggers, build SOPs, and set up review cycles. Use when the user wants to build a habit, change behavior, or achieve a personal goal.
version: "1.1"
updated: "2026-05-22"
assumes: "The goal can be converted into repeated behavior with a concrete trigger and review metric."
conflicts_with: "Do not treat one-off strategy decisions as habits; use project-flow-ops or anthropic-os first."
---

# Behavior Design System

Transform goals into actionable behavior systems through decomposition, habit design, and review.

## Usage Template

**Prompt**
```text
Use behavior-design for this goal. Decompose it into minimum habits, triggers, SOPs, review cadence, and failure handling.
```

**Use Case**
- Converting a goal or intention into a behavior system the user can actually repeat.

**Expected Result**
- The agent creates a habit plan with trigger, minimum action, environment design, review metric, and fallback.

**Output Example**
- A behavior card with goal, trigger, 15-minute action, cue, reward, review metric, and fallback.

**Verification Case**
- The first action takes 15 minutes or less and has a concrete time, place, trigger, and success criterion.

**Verified Effect**
- A vague goal becomes a repeatable behavior loop with a trigger, minimum action, review metric, and fallback.

## Success Metrics

- Plan defines a concrete trigger, a minimum action of 15 minutes or less, a success metric, and a fallback.
- First repetition can be attempted today without buying tools or redesigning the whole environment.
- Review cadence and failure handling are written down.

## When to Use

- User says "I want to build a habit of X"
- User has a goal but hasn't broken it into actions
- User wants to change a behavior pattern
- User is reviewing why a habit didn't stick
- User wants to understand their relationship with AI tools

## Core Architecture

```
Goal → Habits  → Cues → SOPs → Review → Reward
           ↓
     Identity narrative: "I am the kind of person who..."
```

## Human Agency Scale (HAS) — Stanford Framework

When designing behavior systems involving AI tools, use the HAS framework to determine the right level of human involvement:

| Level | Description | When to Use | Example |
|:-----:|-------------|-------------|---------|
| **H1** | AI handles entirely, no human | Routine, low-stakes tasks | Auto-lint, auto-format |
| **H2** | AI needs minimal input | Tasks with clear success criteria | Code review, data entry |
| **H3** | Equal partnership | Creative/analytical work | Research synthesis, design |
| **H4** | Human drives, AI assists | High-stakes decisions | Investment analysis, strategy |
| **H5** | Human essential, AI supports | Relationship/empathy tasks | Coaching, conflict resolution |

**Key insight from Stanford research:**
- 45.2% of occupations prefer H3 (equal partnership) as the dominant level
- Workers generally prefer higher human agency than experts deem necessary
- Skills shift: from information processing → interpersonal competence

**Apply to behavior design:**
- For habits involving AI: Choose the appropriate HAS level
- For skill development: Focus on H4/H5 skills (interpersonal, strategic)
- For automation: Start with H1/H2 tasks, gradually expand

## Workflow

### B1: Define the Goal

Answer:
- Why is this important?
- What identity does it build? ("I want to be someone who...")
- What does success look like in 3 months?

### B2: Decompose into Minimum Habits

Break the goal into 3 minimum habits — actions so small they can't fail:

| Habit | Minimum Viable Action | Trigger | Frequency |
|-------|----------------------|---------|-----------|
| 1 | ≤2 minutes | After existing habit X | Daily |
| 2 | ≤5 minutes | When situation Y occurs | 3x/week |
| 3 | ≤15 minutes | At time Z | Weekly |

### B3: Define SOPs

For each habit, write the execution SOP:

```
WHEN [trigger]
THEN:
1. [step 1 — what to do]
2. [step 2 — what to do]
3. [step 3 — what to do]
AFTER: [immediate reward]

Barrier removal: [what prevents doing this?]
```

### B4: Set Up Review

- Daily: check off habit completion (≤30 sec)
- Weekly: review completion rate + resistance patterns
- Monthly: adjust habits + upgrade minimum bar

### B5: Reframe Identity

Instead of "I want to read more" → "I am a reader."
Instead of "I want to exercise" → "I am someone who moves daily."

## Behavior Design Principles

1. **Minimum start** (< 5 min) — if it takes willpower to start, the habit won't stick
2. **Trigger binding** — attach new habit to an existing routine
3. **Friction removal** — prepare the environment to make it easy
4. **Immediate reward** — the brain needs dopamine within seconds, not months
5. **Identity narrative** — lasting change comes from identity shift, not goal completion

## Quality Gates

- [ ] Goal framed as identity, not outcome
- [ ] 3 minimum habits defined (≤2/5/15 min)
- [ ] Each habit has a clear trigger
- [ ] Friction removal identified for each
- [ ] Review schedule set

More from Mark393295827/third-brain-v5-skills

SkillDescription
agent-teams-commandCommand V6 multi-agent work with bounded roles, ownership, worktree isolation, IPC, integration gates, cleanup, attention budgets, and verification loops. Use when the user needs Claude Code Agent Teams, parallel agents, delegation strategy, Ender-style commander training, von Neumann-style command architecture, Palantir-style ontology command boards, FDE field discovery, EDD integration gates, dynamic workflow tradeoffs, or multi-agent orchestration grounded in Obsidian wiki operating rules.
ai-six-sigma-property-osDesign an AI Six Sigma Black Belt operating model for property service, maintenance dispatch, environmental testing, quote generation, CRM follow-up, and workflow quality dashboards. Use when the user needs a Property Agent OS, AI + Ontology + DMAIC management system, CTQ metrics, agent-team roles, work-order states, or MVP roadmap for operations quality.
anthropic-osImprove a personal or team operating system with self-evolving loops, CASH allocation, 3B creativity, predictive coding, and diagnostics. Use when the user wants to redesign a work method, learning loop, or cognitive operating system.
cognitive-compileDeep learning compile framework — transforms raw information into actionable judgment. Use when the user wants to deeply understand a topic, not just capture it.
context-managerManage the LLM's context window — token budgeting, prompt assembly, truncation strategies. Use when approaching context limits or optimizing prompt costs.
creativity-engineGenerate, validate, and output new ideas based on existing knowledge. Combines combinatorial creativity, cross-domain analogy, and minimum experiments. Use when the user wants fresh ideas, new product concepts, or creative solutions.
daily-okrExecute a V6 daily knowledge compound closed loop — 7 Key Results from input to feedback with scoring, evidence, wiki write-back, and optional scheduled Obsidian daily-loop note. Use when the user wants to do a daily review, plan their day, run a knowledge workflow, or complete the generated daily knowledge-management loop.
loop-engineeringTurn a repeatable task into a bounded, evidence-driven agent loop. Use when Codex needs to decide whether a task merits a Goal, Loop, Automation, or AutoResearch pattern; define a loop contract; check trigger/state/tools/codebase readiness; choose single-agent versus maker-checker versus manager-workers topology; prevent runaway iteration; or repair a loop that stalls, self-grades, exceeds review budget, or writes without verified evidence.
session-learnExtract reusable knowledge from a work session and save concepts, entities, corrections, patterns, ideas, decisions, and gaps to the wiki. Use when ending a session or when the user says to extract knowledge.
startup-evaluationEvaluate startup health using entrepreneurship, VC, and execution frameworks. Use when assessing a startup idea, company, pitch, due diligence target, fundraising readiness, or business model health.