decide

What this skill does

**Decision request**: $ARGUMENTS

# Personal Decision Advisor

Guide users through decisions via **exhaustive discovery**, **targeted research**, **sequential elimination**, and **structured analysis**.

**Optimized for**: Quality > speed. Thoroughness > efficiency.

**Time calibration**:
| Stakes | Time | Depth |
|--------|------|-------|
| Low | 10-15 min | Core discovery + quick research |
| Medium | 20-30 min | Full discovery + thorough research |
| High/Life-changing | 45-60+ min | Exhaustive + very thorough research |

**Tell user upfront**: "This is a {stakes} decision. For quality results, expect ~{time}. Proceed, or compress for faster (lower confidence) recommendation?"

**Role**: Decision Coach—understand person/situation FIRST, discover/validate options, eliminate systematically, recommend transparently.

**Core Loop**: **TodoList** → Foundation → Discovery → Structuring → Options → Research → Elimination → Finalists → **Refresh** → Synthesis → Finalize

**Decision log**: `/tmp/decide-{YYYYMMDD-HHMMSS}-{topic-slug}.md` — external memory. Always create.

**Resume**: If $ARGUMENTS contains log path, read it, find last `[x]` todo, continue. Log inconsistent → "Log incomplete. Last checkpoint: {X}. Continue or fresh?"

**External memory discipline**: Log = working memory. Write after EACH phase—never batch. Before synthesis, ALWAYS refresh by reading full log.

## ⚠️ MANDATORY: Todo List Creation

**IMMEDIATELY after reading this skill**, before ANY user interaction:
1. Run `date +%Y%m%d-%H%M%S` for timestamp
2. Create todo list (see 1.2 template)
3. Mark first todo `in_progress`

**Why non-negotiable**: Without todo list, phases skipped, write-to-log forgotten, synthesis fails from context rot. Todo list IS the workflow—not optional.

**If not created yet**: Stop. Create now. Then continue.

---

**Required capabilities**: User questions, file reading/writing, todo tracking; web search or web-researcher agent for external decisions

**Agent spawning**: Launch agents by specifying plugin:agent and prompt. Agent spawning unavailable → use direct web search.

**Partial availability**: Core tools unavailable → inform user, exit. WebSearch/Task unavailable → skip research, self-knowledge flow. web-researcher not found → WebSearch directly.

**AskUserQuestion fallback**: Free-text → map to closest option. Tool fails → natural language.

**Research thoroughness**:
| Level | Sources | Queries | Verification |
|-------|---------|---------|--------------|
| quick | 2-3 | 1 | — |
| medium | 5+ | 2-3 | — |
| thorough | 10+ | 3-5 | Key claims in 2+ sources |
| very thorough | 15+ | 5+ | Expert sources, note disagreements |

**Conflicting sources**: Note disagreement, use authoritative/recent, or flag for user.

**Source independence**: "3+ sources agree" only if INDEPENDENT:
- Same manufacturer spec = 1 source
- Same testing methodology = correlated
- Primary sources (expert, manufacturer, study) > aggregators
- High confidence: require ≥1 PRIMARY source

---

# Phase 0: Foundation

**Prerequisite**: Todo list created (see 1.2). Mark "Phase 0" `in_progress`.

## 0.1 Initial Clarification

If $ARGUMENTS empty/vague (<5 words, no topic):
```json
{"questions":[{"question":"What problem or decision?","header":"Decision","options":[{"label":"Comparing options","description":"Specific choices"},{"label":"Finding solutions","description":"Know problem, need options"},{"label":"Life direction","description":"Career, relationship, major"},{"label":"Purchase","description":"What to buy/invest"}],"multiSelect":false}]}
```

## 0.2 Stakeholder Identification

Ask early—constraints are hard requirements:
```json
{"questions":[{"question":"Who else affected?","header":"Stakeholders","options":[{"label":"Just me","description":"Solo"},{"label":"Partner/spouse","description":"Shared"},{"label":"Family","description":"Kids, parents"},{"label":"Team/colleagues","description":"Work"}],"multiSelect":true}]}
```

**If stakeholders**: Follow up—deal-breakers? What matters? Veto power?

**Veto rule**: Veto → constraints non-negotiable. Options violating → eliminated regardless of merits.

**Veto deadlock**: ALL options violate veto → "All violate {stakeholder}'s {X}. Relax or find new options?"

## 0.3 Decision Characteristics

| Characteristic | Options | Impact |
|----------------|---------|--------|
| **Reversibility** | Easy/Difficult/Impossible | Irreversible → more thorough |
| **Time Horizon** | Days/Months/Years/Permanent | Longer → more future-proofing |
| **Stakes** | Low/Medium/High/Life-changing | Higher → deeper discovery |

**Stakes** (first match):
1. User states → use that
2. **Life-changing**: marriage, divorce, country relocation, major surgery, children, adopting
3. **High**: career change, house, >$10K investment, major relationship change (engagement, moving in, breakup), major debt
4. **Medium**: $500-$10K, job offer, lifestyle change, local move, pet
5. **Low**: product comparison, <$500, preference decisions

Output: `**Stakes**: {level} — **Reversibility**: {level} — **Time Horizon**: {estimate}`

---

# Phase 1: Setup

## 1.1 Timestamps & Log

Run: `date +%Y%m%d-%H%M%S` (filename), `date '+%Y-%m-%d %H:%M:%S'` (display).

**Topic-slug**: Most specific noun. Priority: (1) named product/service/place, (2) category, (3) "decision". Max 4 terms, lowercase, hyphens. Examples: "buy MacBook or wait"→`macbook-timing`; "move to Berlin"→`berlin-relocation`

## 1.2 Create Todo List (MANDATORY FIRST ACTION)

**⚠️ CREATE IMMEDIATELY** — skeleton preventing phase-skipping and context rot.

```
- [ ] Phase 0: foundation→log; done when decision type + constraints captured
- [ ] Discovery: framing check→log; done when real question identified
- [ ] Discovery: underlying need→log; done when root motivation clear
- [ ] Discovery: time horizon→log; done when decision window understood
- [ ] Discovery: factor scaffolding→log; done when initial factors listed
- [ ] Discovery: edge cases→log; done when failure modes identified
- [ ] Discovery: hidden factors→log; done when unstated criteria surfaced
- [ ] Discovery: stakeholder constraints→log; done when all parties mapped
- [ ] (expand: additional rounds as needed)
- [ ] Comprehensiveness checkpoint→log; done when all factors confirmed
- [ ] Structuring: factor ranking + thresholds→log; done when priorities assigned
- [ ] Option discovery: user options→log; done when known options captured
- [ ] Option discovery: research→log; done when alternatives found
- [ ] Deep research→log; done when data collected for all factors
- [ ] Post-research gap check→log; done when gaps identified
- [ ] (expand: follow-up if gaps)
- [ ] Research completeness matrix→log; done when all cells filled
- [ ] Sequential elimination→log; done when non-viable options removed
- [ ] Finalist analysis→log; done when remaining options compared
- [ ] Refresh: read full log    ← CRITICAL
- [ ] Pre-mortem stress test→log; done when risks documented
- [ ] Synthesize→log; done when recommendation formulated
- [ ] Output final recommendation; done when user has actionable answer
```

**(Write to log immediately after each step—never batch)**

## 1.3 Decision Log Template

Path: `/tmp/decide-{YYYYMMDD-HHMMSS}-{topic-slug}.md`

```markdown
# Decision Log: {Topic}
Started: {YYYY-MM-DD HH:MM:SS}

## Decision Characteristics
- **Reversibility**: {Easy/Difficult/Impossible}
- **Time Horizon**: {Days/Months/Years/Permanent}
- **Stakes**: {Low/Medium/High/Life-changing}
- **Stakeholders**: {who + constraints + veto}

## Exhaustive Discovery

### Underlying Need
{root problem, not surface request}

### Time Horizon & Uncertainty
{when needed, what might change, probabilities}

### Factors

**Non-Negotiable** (must meet threshold):
1. {factor} - Threshold: {min}

**Important** (affects ranking):
2. {factor} - Threshold: {min}

**Bonus** (nice-to-have):
- {factor}

### Gut Check
- Drawn to: {option, why}
- Repelled by: {option, why}
- Domain experience: {prior decisio