# Next Move

> Describe one stuck situation in plain language. Next Move proposes an ordered plan naming
> which proven frameworks apply to it and why; you confirm, drop, reorder or swap steps in
> one click, and a second pass works the confirmed plan through against your actual details.

https://next-move.skillsafe.ai/

## What makes this different from a form

You never pick a mode, a lane or a technique. There is one text box. The model reads the
situation and decides the route; your job is only to confirm or edit the route it proposes.
An app that asks you to choose a technique up front has already made you do the hard part.

## The three beats

1. **Ask.** One textarea. One situation, in your own words.
2. **Plan.** A run returns an ordered list of steps. Each step names the technique it applies,
   why it applies (in words drawn from what you wrote), and what it will produce. Some steps
   carry a one-click alternative. Steps the model considered and rejected are listed with
   reasons, so the routing is legible rather than magic.
3. **Execute.** On confirm, a second run works the confirmed steps through, in your order.

## The free part

Before any run, and without an account, the page reads your text in the browser and extracts
what you said must happen, must not happen, is already done, plus deadlines, figures and
counterparties. Those become chips under the box. They are also sent into the run, and the
returned plan is reconciled against them — so a constraint the model quietly dropped is
shown as "NOT COVERED" rather than passing silently. The reconciliation count is a total over
the live plan, so dropping a step moves it.

## The technique catalogue

24 named techniques drawn from ten skills. The catalogue itself is NOT in this page, not in
the JavaScript, and not in the system prompt. It ships as a private reference corpus that
nothing serves over HTTP; a run retrieves it server-side and the browser only ever receives
the subset the model chose to cite.

## Contract

- Input: `task` (`plan` | `execute`), `query`, optional `context`, `plan[]` on execute,
  `prescan`, `clipped`. Fields go at the top level — an `{"input": {...}}` wrapper is
  accepted by the platform and charged while hiding every field from the model.
- Plan output: `reading`, `steps[]` (`id`, `skill`, `name`, `why`, `produces`, `optional`,
  `alt`), `declined[]`, `unknown`.
- Execute output: `summary`, `results[]` (`id`, `skill`, `name`, `body`, `findings[]`),
  `open[]`, `caveats[]`.
- `results[]` covers exactly the confirmed step ids, in order. A missing one renders as
  "not returned" rather than silently shortening the list.
- Full reference, with worked examples in eight languages: https://next-move.skillsafe.ai/api.html

## Honest limits

- Runs are metered and need a signed-in account. An estimate is always shown first, and the
  estimate itself is free.
- The model reasons over what you typed. It never invents a figure: if arithmetic needs a
  number you did not give, it asks for it instead of estimating.
- This is judgement support, not advice. It does not know your business.

## Sources

Built on ten skills by @guia-matthieu (https://github.com/guia-matthieu/clawfu-skills):
never-split-difference, six-thinking-hats, eisenhower-matrix, first-principles,
persuasion-principles, spin-selling, high-output-management, cognitive-biases,
objection-mapping, forecast-scenarios.

The underlying frameworks are the published work of their authors — Chris Voss, Edward de
Bono, Dwight Eisenhower, Robert Cialdini, Neil Rackham, Andrew Grove and others. This app
routes between them; it does not reproduce them.