PR Diff Summary

Turns a raw pull-request diff into a concise, reviewer-facing changelog with risk flags — so a busy reviewer can grasp a PR in under two minutes instead of scrolling hundreds of changed lines.

What it does

The InTouch AI assistant loads this skill as a prompt and follows it to:

1. Parse the change set (files touched, status, line counts).
2. Classify the PR (feature, bugfix, refactor, docs, test, build/ci, dependency, config, revert).
3. Write a plain-English changelog, collapsing mechanical churn into a single bullet.
4. Scan for risk — security (secrets, auth, SQL injection, disabled TLS), data (schema/migrations, destructive deletes), reliability (removed tests, swallowed exceptions, changed retry/timeout logic), and compatibility (changed public APIs, renamed exports, major dependency bumps).
5. Assign an overall risk level (low / medium / high) and name the files a reviewer should read first.

What to feed it

Either form works:

• Unified diff text — the output of git diff, git show, or a GitHub .diff URL.
• PR API JSON — { title, description, files: [{ path, status, additions, deletions, patch }] }.

Optional extras (PR title, description, branch names, linked issue ids) are used when present and never fabricated when absent. The diff itself is always treated as the source of truth — nothing is assumed beyond what the changed lines show.

What it returns

A compact markdown report:

• PR Summary — one/two-sentence description, type labels, risk level, file/line counts.
• Changelog — one bullet per meaningful change.
• Risk Flags — each flagged file with a one-line reason (omitted with an explicit "No risk flags identified." when clean).
• Review First — the 1–3 files or hunks to read first.

Output is kept tight (around 250 words for a small PR) with no preamble or filler.