Outcome-First Decisions — a business-decision skill for AI agents

The instrument

Every claim about demand sits on a rung.

The Buyer Evidence Ladder is the skill's measuring stick. Lower rungs size tests. Only the highlighted rungs justify building, hiring, automating, or scaling.

01Opinionnoise

02Complimentnoise

03Clicksignal, weak

04Reply or signupsignal

05Qualified conversationsignal

06Referralcommit

07Deposit, payment, signed pilot, signed contractcommit

08Repeat purchase, retained usage, renewal, marginproof

A buyer who pays today is more reliable than a hundred who say they would pay someday.

The output

Six parts. No hedging.

Every full answer is gated by a self-check: named buyer, one number, a 7-day proof test, a kill criterion, three actions for today. If any is missing, the answer is not ready.

sample verdict — saas idea, pre-buildrung 1 evidence

VERDICT Test first. Do not build. WHY The buyer is a category, not a segment. No evidence anyone pays to fix this. Building before a paid pilot is the most common failure path for solo SaaS. PROOF TEST Paid audit offer, $250 up front, 25 named prospects, 7 days. THRESHOLDS KEEP 5+ paid · CHANGE 1–4 paid · KILL 0 paid after 25 contacts — capture the learning and stop. TODAY 1. Write the 25-name list. 2. Write the 4-sentence ask, deadline Friday. 3. Send the first 5 messages.

What's inside

A complete decision system, in plain files.

6 core frameworks

Worth Filter, Cash Proof Sprint, One Number Map, Kill List Audit, Buyer-Problem-Path, Leverage Ladder.

7 supporting frameworks

Offer Sharpener, Sharp Ask Builder, Channel Bet Sheet, Price Integrity Check, Repeatability Test, Opportunity Cost Check, Weekly Decision Review.

Named patterns

23 named anti-patterns, 16 principles, 21 mental models — disambiguated, so the skill can call drift what it is.

12 industry overlays

Vertical-true proof tests and scoreboard defaults — plus an overlay builder that derives them for any vertical not on the list.

∞

Decision-operations layer

A typed decision ontology, a metrics bridge with provenance for every number, and a Portfolio Command Deck across all your bets.

Worked examples

Validation, prioritization, kill-audit, crisis-mode, and portfolio transcripts — plus 3 nested subskills, a buyer-conversation kit, and a calibrated decision journal.

Verticals

It speaks your industry's signals.

A clinic gets payer-path tests. A marketplace gets liquidity-wedge tests. A nonprofit gets funder-retention math. The buyer's money decides which overlay leads.

SaaSServices / AgencyCreator E-commerceB2BMarketplace Local businessEducation / CoachingHealthcare FintechHardware / PhysicalNonprofit + any other — derived in 10 minutes

Entry points

9 Claude Code slash commands.

/validateverdict + 7-day proof test + written kill criterion for one idea

/worth-filterscored comparison across 2+ options — do, test, defer, or drop

/kill-auditkeep/change/kill verdicts with reclaimed capacity allocated by name

/sharpenBuyer/Moment/Promise/Mechanism/Ask one-pager + pre-sale ask

/weekly-reviewfive-question retrospective + next-week plan, 20 minutes

/portfoliothe command deck — every bet, evidence rungs, kill dates, capacity

/log-decisionjournal entries with required confidence + calibration citation

/crisis-modecollapsed output: verdict, 3 hour-deadline actions, dollar kill threshold

/stuck-to-shippedthe 7-day protocol for "I'm busy and nothing is moving"

Install

One zip. Any compatible agent.

# download the latest release
gh release download --repo MeyerThorsten/outcome-first-decisions \
  --pattern 'outcome-first-decisions.zip'

# Claude Code
mkdir -p ~/.claude/skills && unzip outcome-first-decisions.zip -d ~/.claude/skills/

# Codex / OpenAI
mkdir -p ~/.codex/skills && unzip outcome-first-decisions.zip -d ~/.codex/skills/

# Cursor
mkdir -p ~/.cursor/skills && unzip outcome-first-decisions.zip -d ~/.cursor/skills/

# then, in a session:
Use the outcome-first-decisions skill to validate this idea.