# Thought Reframing

> A structured thinking aid for a thought that will not leave you alone. You write the thought
> out; the app takes it apart in your browser, free, before anything is charged. The paid step
> reconciles what the evidence actually supports and offers more accurate ways to put the same
> situation. It is not therapy, it is not a therapist, and it does not diagnose anything.

URL: https://thought-reframing.skillsafe.ai/

## What it is not

Thought Reframing is not therapy and does not present itself as therapy. It is a thinking tool built
on well-known cognitive behavioural ideas — the ones that appear in ordinary self-help workbooks.
It cannot know the person using it, it does not assess or diagnose anything, and it is not a
substitute for talking to a person. Every pattern it names is a description of a form of words,
anchored to a quoted span of the user's own text; nothing it produces is a description of the
user. It does not rate, score, grade or stage anything about anyone, does not say what someone's
difficulty is called or what is causing it, and does not direct anyone toward a named treatment
beyond the plain fact that a professional can help.

## The safety boundary

Before anything is sent anywhere and before any credit is spent, the text is screened in the
browser for indications that the person may not be safe — from themselves or from someone else.
When that screen fires, the app shows crisis and domestic-abuse resources for several regions
instead of an analysis. It does not analyse the thought, does not restate it, does not name
patterns in it, and does not call the model at all. The screen matches on the *shape* of a
statement rather than on a list of alarming words, because the sentences that matter most here
are frequently the quietest ones and contain no alarming word at all. The model prompt carries
the same instruction independently, so the screen does not exist in only one place.

## The free lane

All of this runs client-side, costs nothing, and requires no account:

- The sentence is split into its separate claims, each marked as an **event** (something that
  happened), an **interpretation** (a reading of what something meant), a **prediction**, or an
  **evaluation** (a judgement of worth). Those four are settled by different things, and running
  them together is what makes a thought feel unarguable.
- Eleven wording patterns are detected, each quoting the exact span of the user's text that
  produced it: all-or-nothing wording, reading someone's mind, a prediction stated as settled, a
  thing you did written as a thing you are, only the bad part counted, a rule stated as an
  obligation, followed straight to the worst end, the cause placed on yourself, a feeling used as
  the evidence, measured against other people, and one time standing in for every time.

## The paid lane

One model call, one contract, no task router. It receives the thought, whatever optional detail
the user chose to add, and the browser scan's findings as established facts. It returns a plain
restatement, its own reading of the claims, what evidence bears on them, patterns anchored to
verbatim spans, reframes, a note on what the exercise is not claiming and what remains a real
problem, and at most one small next step.

## Reframing is not positivity

A reframe that tells someone their real problem is not real invalidates them. Every reframe this
app produces must first state what in the original thought is accurate, and only then dispute the
part that outruns the evidence. A reframe whose concession is empty is rejected and shown to the
user as a finding.

The answer is then checked automatically for consolation posing as reasoning — a universal claim
about what everyone or nobody thought, an unhedged promise about how the future will go, or any
suggestion that the concern is smaller than it is — and for clinical framing detected by
construction rather than by vocabulary, so an answer cannot sound like an assessment while
avoiding every clinical word. It is also checked for repeated wording, within one answer and
across the user's own saved answers, because a reframing tool that produces the same sentences
about everyone's distress is worse than one that says less.

## Privacy

Nothing is stored unless the user ticks a box, and the box resets to off every time the page is
opened — choosing to keep one thought is not a request to keep every future one. There is no
browser-local copy of anything written, so the account is the only copy and deleting means
deleting. Individual entries have a delete button; one button deletes everything, pages through
until the collection is empty, and then re-reads it to confirm nothing remains rather than
assuming.

## Interface

- Input: `thought` (required), plus optional `situation`, `feeling`, `intensity`,
  `evidence_for`, `evidence_against`, `what_i_would_tell_a_friend`, and `scan` (the browser
  lane's findings).
- Output: JSON with `restatement`, `claims[]`, `evidence{supports,complicates,missing}`,
  `patterns[]`, `reframes[]`, `check{not_saying,still_true}`, `next_step`.
- API tutorial: https://thought-reframing.skillsafe.ai/api.html
- Model: gpt-terra. Runs are metered; everything described under "the free lane" is not.

## If you are struggling

https://findahelpline.com/ finds a free line wherever you are, in over 100 countries.