# /100x, Yours to Fork

Spawn a parallel research team. Four to seven Opus agents, each attacking your strategic question from a different angle. Synthesize the findings, stress-test with a red team, run a second pass if confidence is too low.

## Why fork it before you run it

The skill ships with generic research lenses (Art, Business, Technical, Growth, Competitive, Market). They work. But your industry has angles those defaults don't cover, and the skill's worth jumps when the lenses match your domain.

The skill also ships with the option to persist results to a Supabase database so future runs read prior runs. That's powerful, but it's optional and personal-stack specific. The customize prompt lets you decide where to save outputs (a Supabase table, a flat folder, an Obsidian vault, or nowhere).

## What's in this folder

- `100x.md` - the skill itself, with `{{PLACEHOLDER}}` markers where my preferences used to live
- `customize-prompt.md` - paste this into Claude Code along with `100x.md`. It asks 5 questions and writes a customized version to `~/.claude/skills/100x/SKILL.md`

## What was stripped

- References to my industry (web dev / Pharallax structural reads / Upwork pipeline)
- The Supabase `research_outputs` table query (replaced with a configurable persistence step)
- The Obsidian path I save outputs to (replaced with a placeholder)
- Permanent corrections from my memory ("Russ builds AND does outreach simultaneously, never gate one on the other") - replaced with a placeholder where you put your own permanent rules
- The `~/Documents/Obsidian Notes/100x/` save path
- The "173 runs" stat from the deck card (your run count starts at zero)

## What's still in

- The full Step 0 → Step 7 pipeline (load memory, premise gate, decompose, spawn, collect, synthesize, red team, save, persist)
- The red-team protocol (3-5 flaws, weakest assumption, confidence %, Round 2 gate at 60%)
- The convergence math (5/5 agents agreeing is signal, 1/5 is noise)
- The hard rule: always Opus for research agents, never sequential, minimum 4 max 7

## How to install

1. Open `customize-prompt.md`. Copy it.
2. Open Claude Code. Paste the customize prompt. Then paste `100x.md` underneath. Send.
3. Answer the 5 questions in plain language.
4. Claude Code writes the result to `~/.claude/skills/100x/SKILL.md`.
5. In any future session, type `/100x {your question}` and get back ranked plays with confidence levels.

## When to use it

`/100x` is overkill for simple questions. If the answer is one Google search away, ask Claude directly. The skill earns its keep when:

- The question is multi-dimensional (technical AND business AND competitive AND something else)
- The stakes are high enough to justify burning maybe $5-15 of API tokens on parallel Opus agents
- You want a structured output you can review later, not a chat transcript

The first thing `/100x` does is gate itself. If your question doesn't need the full team, the skill tells you and answers directly. Respect the gate.

## Run count

The deck shows my running total. Yours starts at 0. If you're like me, you'll cross 50 runs before you stop counting and 100 before you stop being surprised by what comes back.