# Explain It Back

> A pop quiz on a code change before you ship it - usually a diff an AI assistant wrote. Paste the diff; the page numbers it and highlights the lines worth a second look; the examiner walks you through the change and sets four, six or eight questions pinned to lines; you answer in your own words; your answers are marked and the paper comes back stamped: cleared to ship, re-read first, hold the merge, or answer more.

URL: https://explain-it-back.skillsafe.ai/
API guide: https://explain-it-back.skillsafe.ai/api.html

## What runs free in the browser
- The diff reader: git diffs, plain unified diffs and .patch files, or a bare code paste treated as one new file.
- The numbered listing: every row numbered by its line in the new file, removed rows shown unnumbered.
- The hot-spot scan: removed guards, removed or skipped tests, error handling, auth, data, async code, retries and timeouts, regexes, loop bounds, dependency manifests, TODOs.
- Hunk-aware clipping for long changes (hunks with hot spots are sent first), the score, the bubble answer sheet and every export (report .md, answers .csv).

## The metered lanes
quiz - input: task "quiz", diff (the numbered listing), ask, files, hot_spots, count ("4", "6" or "8"), optional focus and avoid. Output: verdict (ready | needs-diff), title, summary, walkthrough [{file, does, look}], questions [{id, kind, file, line, spot, ask, key}], tests, tips. Kinds: predict, why, break, edge, trace, test.
grade - input: task "grade", diff, ask, questions (the quiz's), answers [{id, answer, sure}]. Output: verdict (ship | reread | hold | needs-answers), title, summary, marks [{id, mark, note, reread}], next. Marks: solid, shaky, wrong, blank.
Rules, in order. Quiz: fewer than three changed rows of code -> needs-diff; otherwise ready. Grade: more than half blank -> needs-answers; a hot-spot question wrong or blank, or two wrong -> hold; anything not solid -> reread; all solid -> ship.
Answer keys must be provable from the listing; the examiner never invents a caller, a library behaviour or code the change does not show.

## Limits
A clean stamp means you can explain the change, not that the code is right. It is not a code review.

## Sources
Built on the SkillSafe skills @bmad-code-org/bmad-walkthrough (https://skillsafe.ai/skill/@bmad-code-org/bmad-walkthrough) - walk someone through a change: what it is for, what to look at closely, how to test it - and @lyndonkl/evaluation-rubrics (https://skillsafe.ai/skill/@lyndonkl/evaluation-rubrics) - explicit criteria and a fixed scale for consistent marking. Model: gpt-terra.