Agent Skills › NeverSight/learn-skills.dev › Code Review

Code Review

GitHub

用于AEM EDS项目的代码审查，支持开发后自查和PR审查。检查代码质量、性能、可访问性及最佳实践，识别问题并提供修复建议与截图验证。

data/skills-md/adobe/helix-website/code-review/SKILL.md NeverSight/learn-skills.dev

Trigger Scenarios

完成编码准备提交或发起PR前进行自查收到新的Pull Request需要审查时通过GitHub Actions自动触发PR审查

Install

npx skills add NeverSight/learn-skills.dev --skill Code Review -g -y

More Options

Non-standard path

npx skills add https://github.com/NeverSight/learn-skills.dev/tree/main/data/skills-md/adobe/helix-website/code-review -g -y

Use without installing

npx skills use NeverSight/learn-skills.dev@Code Review

指定 Agent (Claude Code)

npx skills add NeverSight/learn-skills.dev --skill Code Review -a claude-code -g -y

安装 repo 全部 skill

npx skills add NeverSight/learn-skills.dev --all -g -y

预览 repo 内 skill

npx skills add NeverSight/learn-skills.dev --list

SKILL.md

Frontmatter

{
    "name": "Code Review",
    "description": "Review code for AEM Edge Delivery Services projects. Use at the end of development (before PR) for self-review, or to review pull requests. Validates code quality, performance, accessibility, and adherence to EDS best practices."
}

Code Review

Review code for AEM Edge Delivery Services (EDS) projects following established coding standards, performance requirements, and best practices.

When to Use This Skill

This skill supports two modes of operation:

Mode 1: Self-Review (End of Development)

Use this mode when you've finished writing code and want to review it before committing or opening a PR. This is the recommended workflow integration point.

When to invoke:

After completing implementation in the content-driven-development workflow (between Step 5 and Step 6)
Before running git add and git commit
When you want to catch issues early, before they reach PR review

How to invoke:

Automatically: CDD workflow invokes this skill after implementation
Manually: /code-review (reviews uncommitted changes in working directory)

What it does:

Reviews all modified/new files in working directory
Checks code quality, patterns, and best practices
Validates against EDS standards
Identifies issues to fix before committing
Captures visual screenshots for validation

Mode 2: PR Review

Use this mode to review an existing pull request (your own or someone else's).

When to invoke:

Reviewing a PR before merge
Automated review via GitHub Actions workflow
Manual review of a specific PR

How to invoke:

Manually: /code-review <PR-number> or /code-review <PR-URL>
Automated: Via GitHub workflow on pull_request event

What it does:

Fetches PR diff and changed files
Validates PR structure (preview URLs, description)
Reviews code quality
Posts review comment with findings and screenshots
Provides actionable fixes via GitHub suggestions or commits
Explains reasoning for each fix with references to review feedback

Review Workflow

Step 1: Identify Review Mode and Gather Context

For Self-Review (no PR number provided):

# See what files have been modified
git status

# See the actual changes
git diff

# For staged changes
git diff --staged

Understand the scope:

What files were modified?
What type of change is this? (new block, bug fix, feature, styling, refactor)
What is the test content URL? (from CDD workflow)

For PR Review (PR number provided):

# Get PR details
gh pr view <PR-number> --json title,body,author,baseRefName,headRefName,files,additions,deletions

# Get changed files
gh pr diff <PR-number>

# Get PR comments and reviews
gh api repos/{owner}/{repo}/pulls/<PR-number>/comments
gh api repos/{owner}/{repo}/pulls/<PR-number>/reviews

Understand the scope:

What type of change is this? (new block, bug fix, feature, styling, refactor)
What files are modified?
Is there a related GitHub issue?
Are there test/preview URLs provided?

Step 2: Validate Structure (PR Review Mode Only)

Skip this step for Self-Review mode.

Required elements for PRs (MUST HAVE):

Element	Requirement	Check
Preview URLs	Before/After URLs showing the change	Required
Description	Clear explanation of what changed and why	Required
Scope alignment	Changes match PR title and description	Required
Issue reference	Link to GitHub issue (if applicable)	Recommended

Preview URL format:

Before: https://main--{repo}--{owner}.aem.page/{path}
After: https://{branch}--{repo}--{owner}.aem.page/{path}

Flag if missing:

Missing preview URLs (blocks automated PSI checks)
Vague or missing description
Scope creep (changes unrelated to stated purpose)
Missing issue reference for bug fixes

Step 3: Code Quality Review

3.1 JavaScript Review

Linting & Style:

Code passes ESLint (airbnb-base configuration)
No eslint-disable comments without justification
No global eslint-disable directives
ES6+ features used appropriately
.js extensions included in imports

Architecture:

No frameworks in critical rendering path (LCP/TBT impact)
Third-party libraries loaded via loadScript() in blocks, not head.html
Consider IntersectionObserver for heavy libraries
aem.js is NOT modified (submit upstream PRs for improvements)
No build steps introduced without team consensus

Code Patterns:

Existing DOM elements re-used, not recreated
Block selectors scoped appropriately
No hardcoded values that should be configurable
Console statements cleaned up (no debug logs)
Proper error handling where needed

Common Issues to Flag:

// BAD: CSS in JavaScript
element.style.backgroundColor = 'blue';

// GOOD: Use CSS classes
element.classList.add('highlighted');

// BAD: Hardcoded configuration
const temperature = 0.7;

// GOOD: Use config or constants
const { temperature } = CONFIG;

// BAD: Global eslint-disable
/* eslint-disable */

// GOOD: Specific, justified disables
/* eslint-disable-next-line no-console -- intentional debug output */

3.2 CSS Review

Linting & Style:

Code passes Stylelint (standard configuration)
No !important unless absolutely necessary (with justification)
Property order maintained (don't reorder in functional PRs)

Scoping & Selectors:

All selectors scoped to block: .{block-name} .selector or main .{block-name}
Private classes/variables prefixed with block name
Simple, readable selectors (add classes rather than complex selectors)
ARIA attributes used for styling when appropriate ([aria-expanded="true"])

Responsive Design:

Mobile-first approach (base styles for mobile, media queries for larger)
Standard breakpoints used: 600px, 900px, 1200px (all min-width)
No mixing of min-width and max-width queries
Layout works across all viewports

Frameworks & Preprocessors:

No CSS preprocessors (Sass, Less, PostCSS) without team consensus
No CSS frameworks (Tailwind, etc.) without team consensus
Native CSS features used (supported by evergreen browsers)

Common Issues to Flag:

/* BAD: Unscoped selector */
.title { color: red; }

/* GOOD: Scoped to block */
main .my-block .title { color: red; }

/* BAD: !important abuse */
.button { color: white !important; }

/* GOOD: Increase specificity instead */
main .my-block .button { color: white; }

/* BAD: Mixed breakpoint directions */
@media (max-width: 600px) { }
@media (min-width: 900px) { }

/* GOOD: Consistent mobile-first */
@media (min-width: 600px) { }
@media (min-width: 900px) { }

/* BAD: CSS in JS patterns */
element.innerHTML = '<style>.foo { color: red; }</style>';

/* GOOD: Use external CSS files */

3.3 HTML Review

Semantic HTML5 elements used appropriately
Proper heading hierarchy maintained
Accessibility attributes present (ARIA labels, alt text)
No inline styles or scripts in head.html
Marketing tech NOT in <head> (performance impact)

Step 4: Performance Review

Critical Requirements:

Lighthouse scores green (ideally 100) for mobile AND desktop
No third-party libraries in critical path (head.html)
No layout shifts introduced (CLS impact)
Images optimized and lazy-loaded appropriately

Performance Checklist:

Heavy operations use IntersectionObserver or delayed loading
No synchronous operations blocking render
Bundle size reasonable (no minification unless measurable Lighthouse gain)
Fonts loaded efficiently

Preview URL Verification: If preview URLs provided, check:

PageSpeed Insights scores
Core Web Vitals (LCP, CLS, INP)
Mobile and desktop performance

Step 5: Visual Validation with Screenshots

Purpose: Capture screenshots of the preview URL to validate visual appearance. For self-review, this confirms your changes look correct before committing. For PR review, this provides visual evidence in the review comment.

When to capture screenshots:

Always capture at least one screenshot of the primary changed page/component
For responsive changes, capture mobile (375px), tablet (768px), and desktop (1200px)
For visual changes (styling, layout), capture before AND after for comparison
For block changes, capture the specific block area

How to capture screenshots:

Option 1: Playwright (Recommended for automation)

// capture-screenshots.js
import { chromium } from 'playwright';

async function captureScreenshots(afterUrl, outputDir = './screenshots') {
  const browser = await chromium.launch();
  const page = await browser.newPage();

  // Desktop screenshot
  await page.setViewportSize({ width: 1200, height: 800 });
  await page.goto(afterUrl, { waitUntil: 'networkidle' });
  await page.waitForTimeout(1000); // Wait for animations
  await page.screenshot({
    path: `${outputDir}/desktop.png`,
    fullPage: true
  });

  // Tablet screenshot
  await page.setViewportSize({ width: 768, height: 1024 });
  await page.screenshot({
    path: `${outputDir}/tablet.png`,
    fullPage: true
  });

  // Mobile screenshot
  await page.setViewportSize({ width: 375, height: 667 });
  await page.screenshot({
    path: `${outputDir}/mobile.png`,
    fullPage: true
  });

  // Optional: Capture specific block/element
  const block = page.locator('.my-block');
  if (await block.count() > 0) {
    await block.screenshot({ path: `${outputDir}/block.png` });
  }

  await browser.close();

  return {
    desktop: `${outputDir}/desktop.png`,
    tablet: `${outputDir}/tablet.png`,
    mobile: `${outputDir}/mobile.png`
  };
}

// Usage
captureScreenshots('https://branch--repo--owner.aem.page/path');

Option 2: Using MCP Browser Tools

If you have MCP browser or Playwright tools available:

Navigate to the After preview URL
Take screenshots at different viewport sizes
Optionally take element-specific screenshots of changed blocks

Option 3: Manual capture with guidance

Instruct the reviewer or PR author to:

Open the After preview URL
Use browser DevTools to set viewport sizes
Take screenshots and attach to PR

Uploading screenshots to GitHub:

# Upload screenshot as PR comment with image
# First, upload to a hosting service or use GitHub's image upload

# Option A: Embed in PR comment (drag & drop in GitHub UI)
gh pr comment <PR-number> --body "## Visual Preview

### Desktop (1200px)
![Desktop Screenshot](screenshot-url-or-drag-drop)

### Mobile (375px)
![Mobile Screenshot](screenshot-url-or-drag-drop)
"

# Option B: Use GitHub's attachment API (for automation)
# Screenshots can be uploaded as part of the comment body

Screenshot checklist:

Primary page/component captured at desktop width
Mobile viewport captured (if responsive changes)
Specific block/component captured (if block changes)
Before/After comparison (if significant visual changes)
No sensitive data visible in screenshots
Screenshots uploaded and embedded in PR comment

Visual issues to look for:

Layout breaks or misalignment
Text overflow or truncation
Image sizing or aspect ratio issues
Color/contrast problems (especially in dark mode)
Missing or broken icons
Responsive layout issues at breakpoints
Unexpected visual differences from main branch

Step 6: Content & Authoring Review

Content Model (if applicable):

Content structure author-friendly
Backward compatibility maintained with existing content
No breaking changes requiring content migration
New content features only available after preview/publish

Static Resources:

No binaries/static assets committed (unless code-referenced)
User-facing strings sourced from content (placeholders, spreadsheets)
No hardcoded literals that should be translatable

Step 7: Security Review

No sensitive data committed (API keys, passwords, secrets)
No XSS vulnerabilities (unsafe innerHTML, unsanitized user input)
No SQL injection or command injection vectors
CSP headers appropriate for tool pages
External links have rel="noopener noreferrer"

Step 8: Generate Review Summary

Output depends on the review mode:

For Self-Review Mode (End of Development)

Report findings directly to continue the development workflow:

## Code Review Summary

### Files Reviewed
- `blocks/my-block/my-block.js` (new)
- `blocks/my-block/my-block.css` (new)

### Visual Validation
![Desktop Screenshot](path/to/screenshot.png)

✅ Layout renders correctly across viewports
✅ No console errors
✅ Responsive behavior verified

### Issues Found

#### Must Fix Before Committing
- [ ] `blocks/my-block/my-block.js:45` - Remove console.log debug statement
- [ ] `blocks/my-block/my-block.css:12` - Selector `.title` needs block scoping

#### Recommendations
- [ ] Consider using `loadScript()` for the external library

### Ready to Commit?
- [ ] All "Must Fix" issues resolved
- [ ] Linting passes: `npm run lint`
- [ ] Visual validation complete

After self-review: Fix any issues found, then proceed with committing and opening a PR.

For PR Review Mode

Structure the review comment for GitHub:

## PR Review Summary

### Overview
[Brief summary of the PR and its purpose]

### Preview URLs Validated
- [ ] Before: [URL]
- [ ] After: [URL]

### Visual Preview

#### Desktop (1200px)
![Desktop Screenshot](url-or-embedded-image)

#### Mobile (375px)
![Mobile Screenshot](url-or-embedded-image)

<details>
<summary>Additional Screenshots</summary>

#### Tablet (768px)
![Tablet Screenshot](url-or-embedded-image)

#### Block Detail
![Block Screenshot](url-or-embedded-image)

</details>

### Visual Assessment
- [ ] Layout renders correctly across viewports
- [ ] No visual regressions from main branch
- [ ] Colors and typography consistent
- [ ] Images and icons display properly

### Checklist Results

#### Must Fix (Blocking)
- [ ] [Critical issue with file:line reference]

#### Should Fix (High Priority)
- [ ] [Important issue with file:line reference]

#### Consider (Suggestions)
- [ ] [Nice-to-have improvement]

### Detailed Findings

#### [Category: e.g., JavaScript, CSS, Performance]
**File:** `path/to/file.js:123`
**Issue:** [Description of the issue]
**Suggestion:** [How to fix it]

Step 9: Provide Actionable Fixes (PR Review Mode Only)

Skip this step for Self-Review mode - in self-review, you fix issues directly in your working directory.

After identifying issues in a PR review, provide actionable fixes to make it easier for the PR author to address them. The goal is to provide one-click fixes whenever possible.

Quick Reference

PRIMARY METHOD: GitHub Suggestions (use for ~70-80% of fixable issues)

One-click acceptance by PR author
Proper git attribution
Works for changes < 20 lines
Native GitHub UI integration

SECONDARY: Guidance Comments (~20-30% of issues)

For subjective or architectural issues
When multiple approaches exist
"Consider" level suggestions

RARE: Fix Commits (avoid unless necessary)

Only when suggestions won't work
Multi-file complex refactors
Changes requiring extensive testing

Decision Tree: When to Use Which Approach

Approach	When to Use	Examples
GitHub Suggestions (PRIMARY)	Any change that can be expressed as a code replacement	Remove console.log, fix typos, add comments, refactor selectors, update functions, add error handling
Fix Commits (SECONDARY)	Changes that need testing, span many files, or are too large for suggestions	Complex multi-file refactors, security fixes requiring validation, changes >20 lines
Guidance Only (FALLBACK)	Architectural changes, subjective improvements, or when multiple approaches exist	"Consider using IntersectionObserver", design pattern suggestions, performance optimizations

IMPORTANT: Always prefer GitHub Suggestions when possible - they provide the best user experience with one-click acceptance and proper git attribution.

Approach 1: GitHub Inline Suggestions (PRIMARY APPROACH - Recommended)

Use GitHub's native suggestion feature for most fixes. This provides the best user experience with one-click acceptance and proper git attribution.

When to use:

✅ ANY change that can be expressed as a line replacement
✅ Single-line to multi-line changes (up to ~20 lines works well)
✅ Clear, actionable fixes with known solutions
✅ Independent changes that can be applied separately
✅ Changes that don't require extensive testing before commit

When NOT to use:

❌ Changes requiring testing/validation before commit
❌ Very large changes (>20 lines become unwieldy in GitHub UI)
❌ Changes spanning many files (better as fix commits)
❌ Subjective/architectural suggestions with multiple valid approaches

Benefits:

🚀 One-click acceptance by PR author
✅ Proper co-author attribution in git history
🎯 Batch multiple suggestions into single commit
📱 Works in GitHub mobile app
⚡ Zero copy/paste errors
🔍 Clear before/after diff in GitHub UI

How to create suggestions:

GitHub suggestions are created using the Pull Request Reviews API with a special markdown syntax in the comment body:

```suggestion
// The corrected code here
```

Complete workflow with examples:

# Step 1: Get PR information
PR_NUMBER=196
OWNER="adobe"
REPO="helix-tools-website"

# Get the current HEAD commit SHA (required for review API)
COMMIT_SHA=$(gh api repos/$OWNER/$REPO/pulls/$PR_NUMBER --jq '.head.sha')

# Step 2: Analyze the diff to find line positions
# IMPORTANT: Use 'position' in the diff, NOT 'line' in the original file
# Position is the line number in the unified diff output starting from the first diff hunk

# Get the diff to understand positions
gh pr diff $PR_NUMBER --repo $OWNER/$REPO > /tmp/pr-$PR_NUMBER.diff

# Step 3: Create review JSON with suggestions
# Each comment needs:
# - path: file path relative to repo root
# - position: line number IN THE DIFF (not in the file!)
# - body: description + ```suggestion block

cat > /tmp/review-suggestions.json <<JSON
{
  "commit_id": "$COMMIT_SHA",
  "event": "COMMENT",
  "comments": [
    {
      "path": "tools/page-status/diff.js",
      "position": 58,
      "body": "**Fix: Add XSS Safety Documentation** (BLOCKING)\\n\\nAdd a comment to document that this HTML injection is safe:\\n\\n\`\`\`suggestion\\n      const previewBodyHtml = previewDom.querySelector('body').innerHTML;\\n\\n      // XSS Safe: previewBodyHtml is sanitized by mdToDocDom from trusted admin API\\n      const newPageHtml = \\\`\\n\`\`\`\\n\\nThis addresses the security concern by making it clear that XSS has been considered."
    },
    {
      "path": "tools/page-status/diff.js",
      "position": 6,
      "body": "**Fix: Improve Error Handling Pattern**\\n\\nAdd an \\\`ok\\\` flag for more consistent error handling:\\n\\n\`\`\`suggestion\\n  * @returns {Promise<{content: string|null, status: number, ok: boolean}>} Content, status, and success flag\\n\`\`\`"
    },
    {
      "path": "tools/page-status/diff.js",
      "position": 12,
      "body": "**Fix: Return consistent result object**\\n\\n\`\`\`suggestion\\n    return { content: null, status: res.status, ok: false };\\n\`\`\`"
    },
    {
      "path": "tools/page-status/diff.js",
      "position": 16,
      "body": "**Fix: Include ok flag in success case**\\n\\n\`\`\`suggestion\\n  return { content, status: res.status, ok: true };\\n\`\`\`"
    },
    {
      "path": "tools/page-status/diff.css",
      "position": 41,
      "body": "**Fix: Remove stylelint-disable by refactoring selector**\\n\\nUse \\\`.diff-new-page\\\` as intermediate selector to avoid specificity conflict:\\n\\n\`\`\`suggestion\\n.page-diff .diff-new-page .doc-diff-side-header {\\n  padding: var(--spacing-s) var(--spacing-m);\\n\`\`\`"
    }
  ]
}
JSON

# Step 4: Submit the review with all suggestions at once
gh api \
  --method POST \
  -H "Accept: application/vnd.github+json" \
  repos/$OWNER/$REPO/pulls/$PR_NUMBER/reviews \
  --input /tmp/review-suggestions.json

# Step 5: Add a friendly summary comment
gh pr comment $PR_NUMBER --repo $OWNER/$REPO --body "$(cat <<'EOF'
## ✨ One-Click Fix Suggestions Added!

I've added **GitHub Suggestions** that you can apply with a single click! 

### How to Apply

1. Go to the **Files changed** tab
2. Find the inline comments with suggestions
3. Click **"Commit suggestion"** to apply individually
4. Or click **"Add suggestion to batch"** on multiple, then **"Commit suggestions"** to batch

### What's Included

- ✅ [BLOCKING] XSS safety documentation
- ✅ Error handling improvements  
- ✅ CSS selector refactoring (removes linter disables)

After applying, run \`npm run lint\` to verify all checks pass!
EOF
)"

echo "✅ Review with suggestions posted successfully!"
echo "View at: https://github.com/$OWNER/$REPO/pull/$PR_NUMBER"

Key points:

Use position (diff position) NOT line (file line number) - This is critical!
The position is the line number in the unified diff output, counting from the first @@ hunk marker
Multiple suggestions in one review enables batch application
Each suggestion creates a properly attributed co-authored commit when accepted
Escape special characters in JSON (quotes, backticks, newlines)
Always include context in the body before the suggestion block

How to determine the correct position value:

The position is the line number in the unified diff, NOT the line number in the file. Here's how to find it:

Get the diff:
```
gh pr diff <PR-number> > pr.diff
```
Open the diff and count lines from the top, including:
- File headers (--- a/file and +++ b/file)
- Hunk headers (@@ -old,lines +new,lines @@)
- Context lines (no prefix or space prefix)
- Removed lines (- prefix)
- Added lines (+ prefix)
The position is the line number of the ADDED line you want to comment on

Example from actual diff:

--- a/tools/page-status/diff.js
+++ b/tools/page-status/diff.js
  async function fetchContent(url) {
    const res = await fetch(url);
-   if (!res.ok) throw new Error(`Failed to fetch ${url}: ${res.status}`);
-   return res.text();
+   if (!res.ok) {
+     return { content: null, status: res.status };  ← Position 12 (counting from top)
+   }

Best practices for suggestions:

Keep suggestions focused and atomic (one fix per suggestion)
Always include context - explain WHY the change is needed before the suggestion block
Test the suggestion mentally or locally before posting
Only suggest changes you're confident are correct
Include surrounding lines for context (GitHub shows ~3 lines of context)
Use position in the diff, not line in the file (critical!)
Batch related suggestions in one review for easier application
Add a friendly summary comment explaining how to apply
Format suggestion body with markdown (bold headers, inline code)
Escape special characters properly in JSON (quotes, backticks, backslashes)

Approach 2: Fix Commits (For Complex or Multi-File Fixes)

For more complex fixes, create commits directly on the PR branch. This is especially useful when:

Fixes span multiple files
Changes need to be tested together
Refactoring is required
You want to demonstrate the fix working

Prerequisites:

Ensure you have write access to the repository or the PR is from a fork you can access
Verify the PR author has enabled maintainer edits (usually default for same-org PRs)

Workflow:

# 1. Get PR branch information
PR_INFO=$(gh pr view <PR-number> --json headRefName,headRepository,headRepositoryOwner)
BRANCH=$(echo $PR_INFO | jq -r '.headRefName')
REPO_OWNER=$(echo $PR_INFO | jq -r '.headRepositoryOwner.login')

# 2. Fetch the PR branch
git fetch origin pull/<PR-number>/head:pr-<PR-number>

# 3. Check out the PR branch
git checkout pr-<PR-number>

# 4. Make fixes based on review findings
# Example: Fix CSS linter issues by refactoring selectors

# 5. Test your fixes
npm run lint
# Run any relevant tests

# 6. Commit with detailed reasoning
git commit -m "$(cat <<'EOF'
fix: refactor CSS selectors to eliminate linter disables

Refactored CSS selectors in diff.css to resolve specificity conflicts
without using stylelint-disable comments. Changes:

- Increased specificity using .diff-new-page parent selector
- Reordered rules to prevent descending specificity issues
- Maintains same visual appearance and functionality

Addresses code review feedback: https://github.com/{owner}/{repo}/pull/<PR-number>#issuecomment-XXXXX
EOF
)"

# 7. Push to the PR branch
git push origin pr-<PR-number>:$BRANCH

# 8. Add comment to PR explaining what you fixed
gh pr comment <PR-number> --body "$(cat <<'EOF'
## Fixes Applied

I've pushed commits to address some of the review findings:

### Commit 1: Refactor CSS selectors
- ✅ Eliminated all `stylelint-disable` comments
- ✅ Resolved specificity conflicts properly
- ✅ Maintained visual appearance

### Commit 2: Standardize error handling  
- ✅ Updated fetchContent to return consistent result objects
- ✅ Updated all callers to use new pattern
- ✅ Added JSDoc for clarity

### Still Need Action From You:
- [ ] **[BLOCKING]** Add XSS safety comment (see suggestion in review)
- [ ] Consider extracting renderDiffPanel helper (optional)

All linting passes now. Please review the commits and address the remaining item.
EOF
)"

Commit message format for fixes:

fix: <short description>

<Detailed explanation of what was fixed and why>

Changes:
- <specific change 1>
- <specific change 2>

Addresses review feedback: <link to review comment>

Approach 3: Hybrid Approach (RECOMMENDED FOR MOST PRs)

For the majority of PRs, use this hybrid strategy:

GitHub suggestions for all fixable issues (primary method)
Guidance/comments for subjective or architectural issues
Fix commits only for very complex multi-file refactors

Typical workflow:

# 1. Post comprehensive review comment with summary (from Step 8)
gh pr comment <PR-number> --repo {owner}/{repo} --body "<detailed review summary>"

# 2. Submit a review with GitHub suggestions for ALL fixable issues
# (Create JSON as shown in Approach 1 with all suggestions)
gh api POST repos/{owner}/{repo}/pulls/<PR-number>/reviews \
  --input /tmp/review-suggestions.json

# 3. Add a friendly summary about the suggestions
gh pr comment <PR-number> --repo {owner}/{repo} --body "$(cat <<'EOF'
## ✨ One-Click Suggestions Added!

I've added GitHub Suggestions for the fixable issues:
- ✅ [BLOCKING] Security documentation
- ✅ Error handling improvements
- ✅ CSS selector refactoring

Go to **Files changed** tab and click **"Commit suggestion"** to apply!
EOF
)"

# 4. For subjective/architectural issues, add guidance comments separately
# (Only if needed - most issues should have suggestions)

Real-world example distribution:

For a typical PR with 10 issues:

GitHub suggestions: 7-8 issues (concrete fixes)
Guidance comments: 2-3 issues (architectural, subjective, or "consider" level)
Fix commits: 0-1 (only if critically needed)

This approach provides:

✅ Maximum ease of use (one-click fixes)
✅ Clear distinction between concrete fixes and suggestions
✅ Reduced review cycles
✅ Better PR author experience

Real-World Example

PR #196: Support unpublished pages in diff tool

Issues identified in review:

XSS safety documentation needed (BLOCKING)
Four CSS stylelint-disable comments (should fix)
Inconsistent error handling (should fix)
CSS variable fallback inconsistency (optional)

Action taken:

# Created two reviews with 9 total GitHub Suggestions
# - 4 suggestions for diff.js (XSS comment, error handling)
# - 5 suggestions for diff.css (selector refactoring)

# Posted friendly summary comment explaining one-click application

# Result: PR author can fix all issues in ~30 seconds vs 5-10 minutes of manual work

View the actual implementation:

JavaScript suggestions: https://github.com/adobe/helix-tools-website/pull/196#pullrequestreview-3747855930
CSS suggestions: https://github.com/adobe/helix-tools-website/pull/196#pullrequestreview-3747857266
Summary comment: https://github.com/adobe/helix-tools-website/pull/196#issuecomment-3843945119

Time saved:

Traditional review: Author spends 5-10 minutes copying code, editing files, testing
With suggestions: Author spends 30 seconds clicking "Commit suggestions"
Time savings: 90%+ reduction in fix application time

Guidelines for Choosing Approach

Use GitHub Suggestions when: (DEFAULT - use this ~70-80% of the time)

✅ You know the exact fix needed
✅ Change is < 20 lines (works well in GitHub UI)
✅ Fix is objective and unambiguous
✅ Can be applied without extensive testing
✅ Independent of other changes
✅ Examples: Add comments, remove debug code, fix typos, refactor selectors, update error handling, add JSDoc, fix linting issues

Use Fix Commits when: (RARE - use only when suggestions won't work)

✅ Changes span many files (>5 files)
✅ Complex refactoring requiring testing
✅ Changes are interdependent and must be tested together
✅ Security fixes that need immediate attention and validation
✅ You're already working on the PR branch for other reasons
✅ Examples: Large refactors, multi-file renames, complex security patches

Use Guidance Only when: (~20-30% of issues)

✅ Architectural decisions needed
✅ Multiple valid approaches exist
✅ Requires domain knowledge or context from PR author
✅ Subjective improvements ("Consider", "Think about")
✅ Performance optimizations with trade-offs
✅ Examples: "Consider using IntersectionObserver", "You might want to extract this to a util", "Have you thought about caching?"

Decision flowchart:

Issue identified
    ↓
Do I know the exact fix? ──NO──→ Use Guidance
    ↓ YES
Is it < 20 lines? ──NO──→ Use Fix Commit or Guidance
    ↓ YES
Use GitHub Suggestion ✅ (BEST OPTION)

Quality Standards for Fixes

Before posting suggestions or commits:

Verify correctness: Test locally or mentally verify the fix is correct
Check scope: Ensure your fix doesn't introduce new issues
Maintain style: Match existing code style and patterns
Run linters: Ensure fixes don't break linting
Be respectful: Frame fixes as helpful suggestions, not criticism
Link context: Reference the review comment explaining WHY

Don't:

Don't fix issues outside the PR scope
Don't change code style unrelated to the issue
Don't add features or enhancements
Don't push commits without explaining what you fixed
Don't fix subjective issues without discussion

Troubleshooting GitHub Suggestions

Common issues and solutions:

Problem	Cause	Solution
"Validation Failed: line could not be resolved"	Used `line` instead of `position`	Use `position` (diff line number) not `line` (file line number)
Suggestion doesn't appear inline	Wrong position value	Count lines in the diff carefully, including headers
Can't apply suggestion	Merge conflict or branch updated	Author needs to update branch first
Suggestion formatting broken	Unescaped JSON characters	Escape quotes, backticks, newlines in JSON
"Invalid commit_id"	Used old commit SHA	Get current HEAD SHA before creating review

How to verify your review before posting:

# 1. Validate JSON syntax
cat /tmp/review-suggestions.json | jq . > /dev/null && echo "✅ Valid JSON"

# 2. Check position values against diff
gh pr diff <PR-number> > pr.diff
# Manually verify each position value in your JSON matches an added line

# 3. Test with one suggestion first
# Before posting 10 suggestions, test with 1 to ensure positions are correct

Success Criteria

A good fix provision should:

Use GitHub Suggestions as the primary method (for ~70-80% of fixable issues)
Make it easy for the author to address issues (one-click accept)
Provide working, tested code (not untested suggestions)
Explain the reasoning behind each fix in the comment body
Reference the original review feedback from Step 8
Be respectful and collaborative in tone
Focus only on issues identified in the review
Not introduce new issues or regressions
Include a friendly summary comment explaining how to apply suggestions
Use proper position values (diff lines, not file lines)
Batch related suggestions in one review for efficient application

Quick Start Template

Copy this template to quickly create a review with GitHub Suggestions:

#!/bin/bash
# Quick script to create GitHub Suggestions for PR review

PR_NUMBER=YOUR_PR_NUMBER
OWNER="adobe"
REPO="helix-tools-website"

# Get commit SHA
COMMIT_SHA=$(gh api repos/$OWNER/$REPO/pulls/$PR_NUMBER --jq '.head.sha')
echo "✅ PR Head SHA: $COMMIT_SHA"

# Create review JSON
cat > /tmp/review-$PR_NUMBER.json <<JSON
{
  "commit_id": "$COMMIT_SHA",
  "event": "COMMENT",
  "comments": [
    {
      "path": "path/to/file.js",
      "position": DIFF_LINE_NUMBER,
      "body": "**Fix: Issue Title**\\n\\nExplanation of the issue.\\n\\n\`\`\`suggestion\\nYour fixed code here\\n\`\`\`\\n\\nReasoning for the fix."
    }
  ]
}
JSON

# Submit review
gh api POST repos/$OWNER/$REPO/pulls/$PR_NUMBER/reviews \
  --input /tmp/review-$PR_NUMBER.json

# Add summary comment
gh pr comment $PR_NUMBER --repo $OWNER/$REPO --body "✨ GitHub Suggestions added! Go to **Files changed** tab and click **Commit suggestion** to apply."

echo "✅ Review posted! View at: https://github.com/$OWNER/$REPO/pull/$PR_NUMBER"

To use:

Set PR_NUMBER, OWNER, REPO
Replace the comments array with your actual suggestions
Run the script

Review Priority Levels

Must Fix (Blocking)

Issues that MUST be resolved before merge:

Missing preview URLs
Linting failures
Security vulnerabilities
Breaking existing functionality
Performance regressions (Lighthouse score drop)
Accessibility violations
Modifications to aem.js

Should Fix (High Priority)

Issues that SHOULD be resolved:

!important usage without justification
Unscoped CSS selectors
Hardcoded values that should be configurable
Missing error handling
Console statements left in code
CSS in JavaScript

Consider (Suggestions)

Improvements to consider:

Code organization
Naming conventions
Documentation
Additional test coverage
Modern API usage opportunities

Common Review Patterns

Based on actual PR reviews, watch for these patterns:

CSS Issues

"No CSS in JS, please" - Inline styles should use CSS classes
"Use proper CSS" - Avoid style manipulation in JavaScript
"Do we need the !important?" - Strong preference against !important
"Solvable with more CSS specificity" - Increase specificity instead of !important

JavaScript Issues

"Why is it hardcoded here?" - Configuration should be externalized
"No global eslint-disable directives" - Specific, justified disables only
"Clean up console messages" - Remove debug logging
"Use proper feature (e.g., decorateIcons, loadScript)" - Leverage existing utilities

Architecture Issues

"Use existing patterns" - Check if similar functionality exists
"Consider IntersectionObserver" - For lazy loading
"Extract and align design tokens" - Use CSS custom properties

Content Issues

"Check the type in content config" - Validate against expected schema
"Is this feature needed?" - Question value vs complexity

Review Response Templates

Approval

## Approved

Preview URLs verified and changes look good.

### Visual Preview
![Desktop Screenshot](url-or-embedded-image)

<details>
<summary>Mobile View</summary>

![Mobile Screenshot](url-or-embedded-image)

</details>

**Verified:**
- [x] Code quality and linting
- [x] Performance (Lighthouse scores)
- [x] Visual appearance (screenshots captured)
- [x] Responsive behavior
- [x] Accessibility basics

[Any additional notes]

Request Changes

## Changes Requested

### Blocking Issues
[List with file:line references]

### Suggestions
[List of recommendations]

Please address the blocking issues before merge.

Comment

## Review Notes

[Non-blocking observations and questions]

Integration with GitHub Workflow

When triggered via GitHub Actions, the skill should:

Receive: PR number, repository info, event context
Execute: Full review workflow above
Output:
- Review comment on the PR
- Appropriate review status (approve/request-changes/comment)
- Summary posted as PR comment

GitHub Actions integration points:

pull_request event triggers
gh pr review for posting reviews
gh pr comment for detailed feedback

Resources

EDS-Specific Resources

EDS Development Guidelines: https://www.aem.live/docs/dev-collab-and-good-practices
Performance Best Practices: https://www.aem.live/developer/keeping-it-100
Block Development: https://www.aem.live/developer/block-collection
David's Model: https://www.aem.live/docs/davidsmodel

GitHub Code Review Resources

GitHub Suggestions Documentation: https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/reviewing-changes-in-pull-requests/incorporating-feedback-in-your-pull-request
Pull Request Reviews API: https://docs.github.com/en/rest/pulls/reviews
Review Comments API: https://docs.github.com/en/rest/pulls/comments
Creating Review Comments: https://docs.github.com/en/rest/pulls/comments#create-a-review-comment-for-a-pull-request

Success Criteria

For Self-Review Mode

A complete self-review should:

Review all modified/new files
Check code against all quality criteria
Run linting and fix issues
Capture visual screenshots of test content
Verify responsive behavior across viewports
Identify issues to fix before committing
Confirm code is ready for commit and PR

For PR Review Mode

A complete PR review should:

Validate all PR structure requirements (preview URLs, description)
Check code against all quality criteria
Verify performance requirements
Capture and include visual screenshots
Assess visual appearance for regressions
Assess content/authoring impact
Identify security concerns
Provide actionable feedback with specific references
Include screenshots in PR review comment
Provide GitHub Suggestions for all fixable issues (primary method - ~70-80% of issues)
Submit suggestions as a single review for batch application
Include a friendly summary comment explaining how to apply suggestions
Explain reasoning for each fix in the suggestion comment body
Use guidance comments only for subjective/architectural issues
Use fix commits only when suggestions won't work (rare)
Use appropriate review status (approve/request-changes/comment)

Integration with Content-Driven Development

This skill integrates with the content-driven-development workflow:

CDD Workflow:
Step 1: Start dev server
Step 2: Analyze & plan
Step 3: Design content model
Step 4: Identify/create test content
Step 5: Implement (building-blocks skill)
    └── testing-blocks skill (browser testing)
        └── **code-review skill (self-review)** ← Invoke here
Step 6: Lint & test
Step 7: Final validation
Step 8: Ship it (commit & PR)

Recommended invocation point: After implementation and testing-blocks skill complete, before final linting and committing.

What this catches before PR:

Code quality issues
EDS pattern violations
Security concerns
Performance problems
Visual regressions

Benefits of self-review:

Catch issues early (cheaper to fix)
Cleaner PRs with fewer review cycles
Learn from immediate feedback
Consistent code quality

Version History

e0220ca Current 2026-07-05 23:26

Metadata

Files: 0
Version: e0220ca
Hash: ee6ae4b5
Indexed: 2026-07-05 23:26

Code Review

Code Review

Trigger Scenarios

Install

SKILL.md

Code Review

When to Use This Skill

Mode 1: Self-Review (End of Development)

Mode 2: PR Review

Review Workflow

Step 1: Identify Review Mode and Gather Context

Step 2: Validate Structure (PR Review Mode Only)

Step 3: Code Quality Review

3.1 JavaScript Review

3.2 CSS Review

3.3 HTML Review

Step 4: Performance Review

Step 5: Visual Validation with Screenshots

Step 6: Content & Authoring Review

Step 7: Security Review

Step 8: Generate Review Summary

For Self-Review Mode (End of Development)

For PR Review Mode

Step 9: Provide Actionable Fixes (PR Review Mode Only)

Quick Reference

Decision Tree: When to Use Which Approach

Approach 1: GitHub Inline Suggestions (PRIMARY APPROACH - Recommended)

Approach 2: Fix Commits (For Complex or Multi-File Fixes)

Approach 3: Hybrid Approach (RECOMMENDED FOR MOST PRs)

Real-World Example

Guidelines for Choosing Approach

Quality Standards for Fixes

Troubleshooting GitHub Suggestions

Success Criteria

Quick Start Template

Review Priority Levels

Must Fix (Blocking)

Should Fix (High Priority)

Consider (Suggestions)

Common Review Patterns

CSS Issues

JavaScript Issues

Architecture Issues

Content Issues

Review Response Templates

Approval

Request Changes

Comment

Integration with GitHub Workflow

Resources

EDS-Specific Resources

GitHub Code Review Resources

Success Criteria

For Self-Review Mode

For PR Review Mode

Integration with Content-Driven Development

Version History

Same Skill Collection