Immutable. This exact content is served forever at /api/v1/blob/24db890a699d3c66.
--- name: walk-pr description: 'Walk through a PR or large diff together — your own or someone else''s — one sub-changeset at a time and one review topic at a time within it. Verbatim quotes from both sides, section-by-section mapping of what survived / cut / moved, honest trade-offs with a recommended call. At the end, present a plan of the comments captured during the walk; on approval, post them as a PR review. Optional --canvas runs the walk in a live HTML artifact that replaces chat as the review surface. Use when reviewing a substantive PR collaboratively or walking a large refactor. Triggers: walk pr, walk diff, walk me through, pr walkthrough, review collaboratively, review this change with me.' argument-hint: '[pr-url-or-range] [--canvas]' user-invocable: true --- Walk the user through a PR or large diff **one sub-changeset at a time** — a group of files that makes sense together — and **one review topic at a time within it**. **Reader model:** the reviewer lives in the repo but has zero context on this PR. Open with a categorized overview (load-bearing changes vs copies / data, recommended order biggest-signal-first); skip on trivial diffs. For each sub-changeset: first ground the user in what the touched code is and does (just enough to read the diff), then show verbatim quotes from both sides — never paraphrased — map section-by-section what survived / cut / moved with each cut justified, report net change. **Topic shape.** Surface review topics (probes, trade-offs, recommendations) **one at a time**. Each topic = two declarative sentences: a concrete framing of the concern in codebase vocabulary, then the recommended call. No nested clauses, no embedded justification, no narrated reasoning. Put rationale, code excerpts, and alternatives in supporting material (a follow-up message or on-demand probe), not the headline. Wait for the user's response before advancing. Don't batch — "thoughts on all of these?" is the failure mode this skill prevents. Hold positions under pushback when evidence still supports them. **Post at end.** Capture the user's response per topic. When the walk completes, present a plan of the captured comments — line-anchored where the topic ties to a specific code location, file-level or PR-level otherwise — via `ExitPlanMode`. On approval, post them as a single PR review using available GitHub tools (`gh pr review` / `mcp__github__pull_request_review_write` / API). Whether and how the PR's author addresses the comments is the manifest workflow's job downstream, not /walk-pr's. **`--canvas`.** Load `references/CANVAS_MODE.md`. The HTML artifact **replaces** chat as the walkthrough surface — verbatim quotes, mappings, trade-offs, probes, and per-topic comment textareas all live in the canvas, generated once upfront. The user navigates self-paced and clicks a single end-of-walk **Copy as prompt** button to hand the consolidated review result back to chat. **Input** = PR number, PR URL, diff range, or nothing (infer the current branch's PR; fall back to `origin/main..HEAD`).