Menu
PR code review workflow: pull diff, two-pass findings, draft and prioritize comments, then optionally post via gh api.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | gh-pr-code-review |
| description | PR code review workflow: pull diff, two-pass findings, draft and prioritize comments, then optionally post via gh api. |
This skill guides a repeatable pull request review workflow using the GitHub CLI.
It emphasizes two-pass review (to find more issues), drafting comments before posting, and posting only after explicit user confirmation.
Use this skill when the user asks you to review a pull request and leave actionable feedback.
Inputs you should accept:
gh pr view / gh pr diff-R owner/repo if not in a git checkoutgh authenticated (gh auth status)jqA code review in two passes, then a drafting and posting phase:
A) Pull down the PR diff and identify issues
B) Repeat the review, finding only new issues
C) Draft review comments but do not post yet
D) Prioritize comments by criticality:
critical: correctness/security/data loss/broken API/test failuresoptional: maintainability/performance/edge cases/docs/test gapstrivial: nits/style/micro-optimizations (do not show by default)E) Show only critical and optional items to the user. Then, if (and only if) the user explicitly confirms, post:
gh pr view <PR> --json number,title,url,baseRefName,headRefName,headRefOid,files,additions,deletions
gh pr diff <PR> --color=never
Optional: check CI signal while reviewing:
gh pr checks <PR>
Review the diff and record issues as structured items. For each issue capture:
severity: critical | optional | trivialpath: file path when applicableside: RIGHT (new code / additions / context) or LEFT (deletions)line: line number in the diff blob when you can anchor reliablybody: the draft comment text (short, actionable)Focus this pass primarily on:
Re-review the same diff from scratch with a different lens:
Critical rule: do not repeat issues from stage A. Only append genuinely new issues.
Convert each issue into a comment draft that is ready to paste/post:
Assign each comment a criticality:
critical: must-fixoptional: should-fixtrivial: nice-to-fix"Post these to the PR now? (yes/no)"
path + line + side.Use the Issues API to post one summary comment:
gh api -X POST "repos/{owner}/{repo}/issues/<PR_NUMBER>/comments" \
-f body="<summary markdown>"
Endpoint reference: POST /repos/{owner}/{repo}/issues/{issue_number}/comments.
Use the Pull Request Review Comments API:
SHA="$(gh pr view <PR> --json headRefOid -q .headRefOid)"
gh api -X POST "repos/{owner}/{repo}/pulls/<PR_NUMBER>/comments" \
-f body="<comment markdown>" \
-f commit_id="$SHA" \
-f path="path/to/file.py" \
-F line=42 \
-f side=RIGHT
Endpoint reference: POST /repos/{owner}/{repo}/pulls/{pull_number}/comments.
Inline comments require an exact file path, a line number, and a side.
Use this deterministic approach for unified diffs:
Each hunk header looks like @@ -old_start,old_count +new_start,new_count @@.
Initialize counters: old_line = old_start, new_line = new_start.
Walk each subsequent hunk line:
' ' (context): increment both old_line and new_line.'+' (addition): increment new_line only.'-' (deletion): increment old_line only.To comment on the new code, use side=RIGHT and set line=new_line.
To comment on a deletion, use side=LEFT and set line=old_line.
If you cannot reliably determine line/side (for example, complex hunks or
uncertainty), do not post an inline comment. Include the item only in the
summary timeline comment.
Guide for authoring Agent Skills with strong YAML `description` triggers, progressive disclosure, and bundled resources. Use when creating or updating a skill, running init_skill.py or package_skill.py, or improving a bland skill description so agents load the skill on the right user tasks.
Build an awesome marimo notebook that brings a research paper or dataset to life — for competitions, tutorials, demos, or blog companions. Use when asked to "make an awesome marimo notebook," "build a notebook for this paper," "create a competition submission," or pair-program an interactive research explainer on molab. Covers: what makes a notebook engaging (intuition game, real experiment, novel extension, design system), paper selection, narrative arc and proportions, custom anywidget patterns, wigglystuff CellTour, effective GPU usage, memory-efficient training (fused STE, gradient checkpointing), parallel review subagents, and a pre-publication polish checklist. CRITICAL: mo is auto-injected (import in ONE cell); every top-level name must be unique across ALL cells; custom anywidget needs model.save_changes(); use mo.output.replace() for live charts; guard all torch.cuda calls for CPU-only machines.
Create VS Code-style typewriter animation videos with mechanical keyboard sounds. Supports 4 typing speeds, ghost text autocomplete, strikethrough corrections, emoji picker, file tab switching, syntax highlighting (6 languages), Chinese IME (pinyin + direct mode), inline images, image stacks, animated checkboxes, Excalidraw hand-drawn diagrams, 8 visual themes, 4 sound packs, and A-roll narration sync. Built on Remotion (React-based video rendering). Use when the user wants to create code typing videos, explainer B-roll, or typewriter-style animations.
Create animated videos using Remotion from topics, product URLs, Google reviews, talking-head videos, CSV data, or channel/brand URLs. Supports 6 video types: educational explainers, product launch demos, testimonial/social proof, avatar video overlays, data visualization dashboards, and lower third overlays. Each follows a 2-step workflow: research/scrape/analyze then design and animate with spring animations, SVG diagrams, and count-up effects. Requires the Remotion best practices skill (install with `npx skills add remotion-dev/skills`). Use when the user asks to create a Remotion video, explainer video, educational video, product demo video, testimonial video, video with animated overlays, data visualization video, animated dashboard, lower third, transparent overlay, short-form vertical video for mobile, or multi-product video from a website.
Improve coherence in drafts through a four-pass sub-agent workflow: resolve argument conflicts, smooth section transitions, smooth paragraph transitions, and run a final coherence audit. Use when a user says things like 'make this coherent', 'tighten the flow', 'this feels choppy', 'can you smooth transitions', or asks to revise prose in Markdown, notes, essays, blog posts, emails, or docs so ideas connect cleanly without adding new points.
Break long-lived pull request branches into a mergeable stack of small PRs with clear dependency order and story flow. Use when a branch has grown too large, when the user asks to split a PR into stacked PRs, or when each PR must be reviewable in about 5 minutes while preserving logical narrative across the stack.