# `MingaAdversarial`
[🔗](https://github.com/jsmestad/minga-adversarial/blob/main/lib/minga_adversarial.ex#L1)

An adversarial pair programmer that challenges the assumptions in your code.

On demand (or, depending on the skepticism dial, on save) it sends the
current file to the model with an adversarial prompt ("what does this code
assume that could break?") and surfaces the answers as advisory gutter
findings (amber `?`) you can read on hover and step through with the normal
diagnostic navigation. Findings clear themselves as you edit.

The **skepticism dial** (`:skepticism` option) controls how eager it is:

  * `:off`: disabled; the analyze command is a no-op.
  * `:manual`: only the `adversarial-analyze` command runs it (default).
  * `:on_save`: also analyzes on save.
  * `:paranoid`: analyzes on save with a maximally skeptical prompt.

Default is `:manual`, so there is zero background model spend until you ask.

Keybindings (under the `SPC a` "+ai" leader):

  * `SPC a A`: challenge assumptions in the current file now

Also `M-x adversarial-analyze`, `M-x adversarial-clear`, and
`M-x adversarial-cycle-skepticism`.

---

*Consult [api-reference.md](api-reference.md) for complete listing*
