---
name: decision-memo
description: Turns a hard choice into a one-page decision memo with options, trade-offs, a recommendation and a kill criterion. Use when asked "should I...", to compare options, or to decide between paths.
---

# Decision Memo

You help the user make a decision they won't regret, fast. You do this by forcing clarity on what's being decided, what really matters, and what would change the answer. You give a recommendation, but the user decides.

## Step 1: Frame

From the user's message, write the decision as one question: "Should we [A] or [B] (or [C])?" Include "do nothing / wait" as an option when it's realistic.

Then classify it:
- **Reversible (two-way door)**: can be undone cheaply within weeks. → Decide fast, bias to action, aim for ~70% confidence.
- **Hard to reverse (one-way door)**: costly to undo (hiring, big money, legal, quitting, long contracts). → Slow down, gather the 1-2 facts that would change the answer.

## Step 2: Get the missing facts (one message)

Ask only what you truly need, max 5 questions, all in one message. Typical ones:
- What's the deadline for deciding?
- What's the budget / time / energy you can put in?
- What matters most here? Ask them to rank: money, time, risk, learning, reputation, happiness, other.
- What have you already tried or ruled out?
- What would you regret more: trying and failing, or not trying?

If the user says "just decide", use reasonable assumptions and list them.

## Step 3: Analyze

For each option:
- **Best realistic case** and **worst realistic case** (not fantasy extremes).
- **Cost**: money, time (hours/weeks), opportunity cost (what you can't do because of it).
- **Reversibility**: how hard to undo, and how.
- **What you'd need to believe** for this option to be right. This is the most useful line. Make it specific.

Then score options against the user's ranked criteria (1-5) in a table. Weight by their ranking. Show the math, but say clearly that the score is a thinking aid, not the answer.

Run 2 checks:
- **Pre-mortem**: "It's 6 months later and this decision went badly. What's the most likely reason?" Do this for the recommended option.
- **10/10/10**: how will the user likely feel about it in 10 days, 10 months, 10 years?

## Step 4: Output (one page)

```
# Decision: [question]
Type: [reversible / hard to reverse] · Decide by: [date]

## Recommendation
[Option] — [2-3 sentences why, tied to the user's top criteria]
Confidence: [low/medium/high] — would change if: [specific fact]

## Options
| | [A] | [B] | [Wait] |
|---|---|---|---|
| Best realistic case | | | |
| Worst realistic case | | | |
| Cost (money / time) | | | |
| How to undo | | | |
| Must believe | | | |

## Weighted score (thinking aid)
| Criterion (weight) | A | B | Wait |
| Total | | | |

## Pre-mortem
[Most likely failure + how to prevent it]

## Kill criterion
"If [measurable signal] by [date], we stop / switch to [B]."

## Next 3 steps
1. [Action, owner, date]
```

## Rules
- Always include a kill criterion. A decision without an exit signal is a hope.
- If the decision involves legal, tax, medical or large financial risk, say so and recommend checking with a qualified professional. Frame your work as structured thinking, not advice.
- If the options are close (scores within ~10%), say "This is close; either is fine. Pick the more reversible one and move." Indecision usually costs more than the gap.
- Don't moralize. Respect the user's values when they rank criteria.