build-verification-expert

Verifies build output integrity after code changes. Runs builds, validates artifact structure, checks for regressions in bundle size, asset references, and static export completeness. The last line of defense before a bad deploy ships.

仓库: curiositech/windags-skills
最近来源活动: 2026年3月31日 04:41
检测到的 SKILL.md 语言: 英语
星标: 10
分支: 2

安装方式

默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。

检查来源文件

决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。

返回搜索返回安装

name	build-verification-expert
license	Apache-2.0
description	Verifies build output integrity after code changes. Runs builds, validates artifact structure, checks for regressions in bundle size, asset references, and static export completeness. The last line of defense before a bad deploy ships.
category	Code Quality & Testing
tags	["build","verification","ci","artifacts","validation"]
allowed-tools	["Read","Bash(*)","Glob","Grep","Write","Edit"]
pairs-with	[{"skill":"ci-cache-optimizer","reason":"Build verification catches issues that stale CI caches silently introduce"},{"skill":"cloudflare-pages-cicd","reason":"Validates output structure before Cloudflare Pages deployment"},{"skill":"test-automation-expert","reason":"Build verification is the complement to test passes — tests pass but build broken is a real failure mode"},{"skill":"image-optimization-engineer","reason":"Asset size validation requires understanding image optimization tradeoffs"}]

Build Verification Expert

Systematic verification of build output after code changes. Catches the class of bugs that pass all tests but break in production: missing assets, broken static exports, inflated bundles, dead routes, and malformed sitemaps.

When to Use

Use for:

Verifying a build completes without errors or warnings after significant changes
Checking that static site exports contain all expected routes and pages
Validating artifact sizes haven't regressed beyond acceptable thresholds
Confirming asset references (images, fonts, scripts) resolve to real files
Auditing sitemap.xml and robots.txt for correctness after content changes
Comparing build output structure against a known-good baseline
Pre-deploy sanity checks in CI or before manual deploy
Diagnosing "works in dev but broken in production" issues

Do NOT use for:

Writing or running unit/integration tests (use test-automation-expert)
Performance profiling of the running application (use performance-profiling)
Setting up CI/CD pipelines (use github-actions-pipeline-builder or cloudflare-pages-cicd)
Optimizing bundle size proactively (use cost-optimizer or framework-specific skills)

Core Capabilities

1. Build Execution and Error Classification

Run the build and classify every diagnostic:

# Capture full build output with timing
time npm run build 2>&1 | tee /tmp/build-output.log

# Separate errors from warnings
grep -E "^(Error|error|ERROR)" /tmp/build-output.log > /tmp/build-errors.log
grep -iE "(warn|warning)" /tmp/build-output.log > /tmp/build-warnings.log

Error classification:

Severity	Pattern	Action
Fatal	`Error: Build failed`, non-zero exit code	Stop. Fix before proceeding.
Type error	`TS\d+:`, `Type '...' is not assignable`

# For Next.js static export
OUTPUT_DIR="out"

# Verify critical files exist
REQUIRED_FILES=(
  "index.html"
  "404.html"
  "_next/static"
  "sitemap.xml"
  "robots.txt"
  "favicon.ico"
)

for file in "${REQUIRED_FILES[@]}"; do
  if [ ! -e "$OUTPUT_DIR/$file" ]; then
    echo "MISSING: $file"
  fi
done

# Count HTML pages — compare against expected routes
ACTUAL_PAGES=$(find "$OUTPUT_DIR" -name "*.html" | wc -l)
echo "HTML pages found: $ACTUAL_PAGES"

Framework	Output Dir	Key Artifacts
Next.js (static)	`out/`	`index.html`, `_next/static/`, `_next/data/`
Next.js (server)	`.next/`	`.next/server/`, `.next/static/`
Vite/React	`dist/`	`index.html`, `assets/`
Astro	`dist/`	Per-route HTML, `_astro/`
Cloudflare Pages	`out/` or `dist/`	`_worker.js` (if functions), `_headers`, `_redirects`

# Extract all asset references from HTML files
grep -rhoP '(?:src|href)="([^"]*)"' "$OUTPUT_DIR"/*.html | \
  sed 's/.*="\(.*\)"/\1/' | \
  sort -u > /tmp/referenced-assets.txt

# Check each reference resolves
while read -r ref; do
  # Skip external URLs, data URIs, and anchors
  if [[ "$ref" =~ ^(http|https|data:|#|mailto:) ]]; then
    continue
  fi

  # Resolve relative to output dir
  target="$OUTPUT_DIR/$ref"
  if [ ! -e "$target" ]; then
    echo "BROKEN REF: $ref"
  fi
done < /tmp/referenced-assets.txt

# Capture current sizes
find "$OUTPUT_DIR" -type f -name "*.js" -exec du -k {} + | sort -rn > /tmp/js-sizes.txt
find "$OUTPUT_DIR" -type f -name "*.css" -exec du -k {} + | sort -rn > /tmp/css-sizes.txt

# Total JS payload
JS_TOTAL=$(find "$OUTPUT_DIR" -name "*.js" -exec cat {} + | wc -c)
echo "Total JS: $(echo "scale=1; $JS_TOTAL / 1024" | bc)KB"

# Compare against threshold
JS_LIMIT_KB=500
JS_ACTUAL_KB=$((JS_TOTAL / 1024))
if [ "$JS_ACTUAL_KB" -gt "$JS_LIMIT_KB" ]; then
  echo "WARNING: JS bundle ($JS_ACTUAL_KB KB) exceeds limit ($JS_LIMIT_KB KB)"
fi

Asset Type	Warning Threshold	Hard Limit	Why
Total JS	300KB gzipped	500KB gzipped	LCP impact on mobile
Single chunk	150KB gzipped	250KB gzipped	Parse time
Total CSS	50KB gzipped	100KB gzipped	Render blocking
Individual image	200KB	500KB	Layout shift, bandwidth
Total HTML per page	100KB	200KB	Time to first byte

# Validate sitemap.xml is well-formed XML
xmllint --noout "$OUTPUT_DIR/sitemap.xml" 2>&1

# Extract URLs from sitemap and verify they correspond to real pages
grep -oP '<loc>\K[^<]+' "$OUTPUT_DIR/sitemap.xml" | while read -r url; do
  # Convert URL to local path
  path=$(echo "$url" | sed 's|https\?://[^/]*||')
  local_path="$OUTPUT_DIR${path}index.html"

  if [ ! -f "$local_path" ] && [ ! -f "$OUTPUT_DIR${path%.html}.html" ]; then
    echo "SITEMAP ORPHAN: $url has no corresponding page"
  fi
done

# Check robots.txt
if [ -f "$OUTPUT_DIR/robots.txt" ]; then
  echo "robots.txt exists"
  # Verify sitemap reference
  grep -q "Sitemap:" "$OUTPUT_DIR/robots.txt" || echo "WARNING: robots.txt missing Sitemap directive"
else
  echo "WARNING: No robots.txt found"
fi

# Extract all internal links from the built site
grep -rhoP 'href="(/[^"]*)"' "$OUTPUT_DIR" | \
  sed 's/href="//;s/"//' | \
  sort -u > /tmp/internal-links.txt

# Check each link resolves to a file
while read -r link; do
  # Try both /path/index.html and /path.html
  if [ ! -f "$OUTPUT_DIR${link}index.html" ] && \
     [ ! -f "$OUTPUT_DIR${link}.html" ] && \
     [ ! -f "$OUTPUT_DIR${link}" ]; then
    echo "DEAD INTERNAL LINK: $link"
  fi
done < /tmp/internal-links.txt

Condition	Verdict	Reasoning
Build exits non-zero	FAIL	Obvious. Nothing to verify.
TypeScript errors in output	FAIL	Type errors are bugs. Period.
Missing route HTML file	FAIL	404 in production.
Broken asset reference	FAIL	Visible breakage to users.
Bundle size exceeds hard limit	FAIL	Performance degradation ships.
Bundle size exceeds warning threshold	WARN	Investigate, but may be intentional.
Deprecation warnings	WARN	Track, schedule cleanup.
New unrecognized file type in output	WARN	Could be intentional, could be accidental.
Sitemap URL with no page	WARN	SEO issue, not user-facing breakage.