Diff-to-PR Narrative Architect
Diff-to-PR Narrative Architect
AI coding agents now produce diffs faster than reviewers can read them. This skill is the human-narrative layer they don't produce well: given a diff, it writes what a PR description generator writes, plus the one thing that actually saves review time - a risk callout list that tells the reviewer where to look closely and why.
What it does
Paste a git diff (or a file list + description of changes), and you get:
- PR Description - what changed, why, and how to test it
- Changelog Entry - Keep-a-Changelog style, written for the right audience (end users or internal devs)
- Reviewer Risk Callouts - specific, evidence-backed flags: auth/payment code touched, a function's blast radius bigger than the diff shows, weakened error handling, breaking API signature changes, missing test coverage, and more
Why it's different
Never pads with generic warnings. If nothing risky is in the diff, it says so plainly instead of manufacturing a callout - false alarms are what make reviewers stop trusting risk flags in the first place.
Who it's for
Developers and teams shipping fast with AI coding agents (Claude Code, Cursor, Copilot, Codex) who need PRs reviewers can actually trust and skim quickly.
Pairs well with
AGENTS.md & CLAUDE.md Architect - one sets up repo context for your coding agent, this one explains what a specific change actually did.
FAQ
Do I need the full diff?
A full diff gives the best results, but a file list plus a description of what changed also works - the skill will ask if it needs more detail to flag something specific.
Will it flag risks that aren't really there?
No - every flag is tied to something actually present in the diff. No generic checklist padding.
