diff-walkthrough
Present a changeset for comprehension: reorganize the diff by reviewer value (core logic first, wiring condensed, boilerplate summarized), distill dense hunks into pseudocode, trace surprising behavior changes on a concrete input, and tag the few genuinely tricky hunks. Use when asked to "walk me through the diff/PR", "explain what changed", or to catch someone up on a branch. Not a review: it explains, it does not judge — for findings use /review (or /basic-review), for quality use /clean-code.
- License 许可
- MIT
- 最近更新
- 2026-11-01
- Stars
- ★ 12
- Forks
- 0
01 前言 Front Matter
- name
- diff-walkthrough
- version
- 1.0.0
- triggers
- [ "walk me through the diff", "walk me through this pr", "explain what changed", "catch me up on this branch" ]
- description
- Present a changeset for comprehension: reorganize the diff by reviewer value (core logic first, wiring condensed, boilerplate summarized), distill dense hunks into pseudocode, trace surprising behavior changes on a concrete input, and tag the few genuinely tricky hunks. Use when asked to "walk me through the diff/PR", "explain what changed", or to catch someone up on a branch. Not a review: it explains, it does not judge — for findings use /review (or /basic-review), for quality use /clean-code.
- allowed-tools
- [ "Bash", "Read", "Grep", "Glob" ]
- metadata.upstream
- https://github.com/cursor/plugins (pr-review-canvas/skills/pr-review-canvas, MIT)
03 SKILL.md 正文
/diff-walkthrough
Present a changeset reorganized for reviewer comprehension — not in file-tree
order, not a raw diff dump. This skill explains what happened; it does not
review. Findings belong to /review or /basic-review.
Gather the diff
Accept a branch, a range, staged/uncommitted changes, or a PR reference (with
gh pr diff when the remote is reachable). If the user did not say which
changeset they mean, stop and ask. Never guess from recent history or fall
back to a silent default — a walkthrough of the wrong diff is worse than no
walkthrough.
Group changes for comprehension
Do not present files in alphabetical or tree order. Reorganize into sections ordered by reviewer value:
- Core logic — new behavior, algorithm changes, state transitions, API surface changes. Show the real hunks with surrounding context.
- Wiring & integration — route registration, dependency injection, config plumbing that connects the core logic. Condensed: enough to confirm correctness.
- Boilerplate & mechanical — renames, import reordering, generated code, formatting, type re-exports. Summarize as a list of files and stats; no inline hunks unless something is genuinely surprising there.
Lead with core logic: attention is freshest at the top.
Distill dense logic into pseudocode
When a core hunk is dense — nested conditions, state machines, retry/backoff, multi-step transforms — add a few lines of pseudocode next to it that strip syntax and error handling to expose the essential control flow. Only for hunks that are genuinely hard to scan; straightforward changes need no mirror.
Trace surprising behavior on a concrete example
When a hunk changes behavior in a way that is hard to predict from reading it — reordered operations, new short-circuits, altered edge cases — pick one small, realistic input and walk it through the old and new code paths side by side, highlighting the step where they diverge and the observable outcome. Reserve this for genuinely surprising changes.
Call attention to tricky things — sparingly
When a hunk hides something risky or easy to miss, mark it with a short tag
(Subtle, Breaking, Race condition, Perf) and one sentence. Overuse
destroys the signal; a few tags per walkthrough is the ceiling.
Tone
Write reviewer-facing commentary, not a changelog: why it changed, how files interact ("the new validator in core.py is invoked by the route added in routes.py"), anything the diff alone does not make obvious. One or two sentences per note. No findings, no verdicts, no "LGTM".
Adapted from cursor/plugins
pr-review-canvas/skills/pr-review-canvas/SKILL.md (MIT, © Cursor) — the
canvas-SDK presentation layer replaced with markdown conventions.
04 同仓库更多 Skills
youtube
Download YouTube video transcripts when user provides a YouTube URL or asks to download/get/fetch a transcript from YouTube. Also use when user wants to transcribe or get captions/subtitles from a YouTube video.
write-plan
Use when you have a spec or requirements for a multi-step task, before touching code. Produces an engineering plan complete with TDD tasks, DRY/SOLID/YAGNI principles, a required Test Plan & Verification section, and a coverage target — then auto-runs /plan-eng-review as an independent double-check before execution.
worktrees
Use when starting feature work that needs isolation from current workspace or before executing implementation plans - creates isolated git worktrees with smart directory selection and safety verification
web-perf
Analyzes web performance using Chrome DevTools MCP. Measures Core Web Vitals (LCP, INP, CLS) and supplementary metrics (FCP, TBT, Speed Index), identifies render-blocking resources, network dependency chains, layout shifts, caching issues, and accessibility gaps. Use when asked to audit, profile, debug, or optimize page load performance, Lighthouse scores, or site speed. Biases towards retrieval from current documentation over pre-trained knowledge.
verify
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always
validate-idea
Validate a business idea using the minimalist entrepreneur framework. Use when someone has a business idea and wants to test if it's worth pursuing before building anything.