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.
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.
A direct command skips the review prompt. Inspect the source before running it.
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.
[{"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
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 timingtime 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
Fix. These are bugs hiding as warnings.
Missing module
Module not found, Cannot resolve
Check imports, package.json, and tsconfig paths.
Deprecation
DeprecationWarning
Log for future cleanup, not blocking.
Size warning
exceeds the recommended size limit
Investigate if new, acceptable if known.
Critical rule: A build that exits 0 but emits TypeScript errors to stderr is NOT a passing build. Many frameworks continue past type errors in dev mode. Always check --strict or the equivalent.
2. Output Structure Validation
After a successful build, verify the output directory matches expectations:
Broken asset references are the most common post-build failure. An image path that works in dev (via dev server magic) but fails in production (because the path is wrong relative to the output).
# 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 resolveswhileread -r ref; do# Skip external URLs, data URIs, and anchorsif [[ "$ref" =~ ^(http|https|data:|#|mailto:) ]]; thencontinuefi# Resolve relative to output dir
target="$OUTPUT_DIR/$ref"if [ ! -e "$target" ]; thenecho"BROKEN REF: $ref"fidone < /tmp/referenced-assets.txt
Common broken reference patterns:
/images/logo.png when the actual file is at /img/logo.png
Hashed filenames (main.abc123.js) not matching because the hash changed
Absolute paths that assumed a subdirectory deploy but built for root
Case sensitivity differences between macOS (case-insensitive) and Linux (case-sensitive)
4. Bundle Size Tracking
Size regressions are silent killers. A single unguarded import can add 200KB.
# Validate sitemap.xml is well-formed XML
xmllint --noout "$OUTPUT_DIR/sitemap.xml" 2>&1
# Extract URLs from sitemap and verify they correspond to real pages
grep -oP '<loc>\K[^<]+'"$OUTPUT_DIR/sitemap.xml" | whileread -r url; do# Convert URL to local path
path=$(echo"$url" | sed 's|https\?://[^/]*||')
local_path="$OUTPUT_DIR${path}index.html"if [ ! -f "$local_path" ] && [ ! -f "$OUTPUT_DIR${path%.html}.html" ]; thenecho"SITEMAP ORPHAN: $url has no corresponding page"fidone# Check robots.txtif [ -f "$OUTPUT_DIR/robots.txt" ]; thenecho"robots.txt exists"# Verify sitemap reference
grep -q "Sitemap:""$OUTPUT_DIR/robots.txt" || echo"WARNING: robots.txt missing Sitemap directive"elseecho"WARNING: No robots.txt found"fi
6. Static Export Completeness
For statically exported sites, every route must produce an HTML file:
# Extract all internal links from the built site
grep -rhoP 'href="(/[^"]*)"'"$OUTPUT_DIR" | \
sed 's/href="//;s/"//' | \
sort -u > /tmp/internal-links.txt
# Check each link resolves to a filewhileread -r link; do# Try both /path/index.html and /path.htmlif [ ! -f "$OUTPUT_DIR${link}index.html" ] && \
[ ! -f "$OUTPUT_DIR${link}.html" ] && \
[ ! -f "$OUTPUT_DIR${link}" ]; thenecho"DEAD INTERNAL LINK: $link"fidone < /tmp/internal-links.txt
Decision Points
When to Fail vs Warn
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.
Build Comparison Strategy
When comparing against a baseline:
Always compare against the last known-good build, not against "what I expect."
Store baselines as JSON manifests: { "files": [...], "sizes": {...}, "routes": [...] }
Diff the manifests to surface additions, removals, and size changes.
New files are usually fine. Missing files are usually bad.
Anti-Patterns
Anti-Pattern: "Build Passed, Ship It"
What it looks like: CI shows green checkmark because npm run build exited 0.
Why wrong: A zero exit code means the build tool finished. It does not mean the output is correct, complete, or deployable. Frameworks routinely exit 0 while emitting warnings about missing pages, oversized bundles, or unresolved imports that fall back to runtime errors.
Instead: Run explicit post-build verification. The build command is step 1. Verification is step 2. They are not the same step.
Anti-Pattern: Manual Spot-Checking
What it looks like: Developer opens the built site in a browser, clicks around for 30 seconds, says "looks good."
Why wrong: Humans cannot visually verify 50 routes, 200 asset references, and bundle sizes in 30 seconds. They check the homepage and maybe one other page.
Instead: Automated scripts that exhaustively check every route and asset. Human review is for visual fidelity, not structural integrity.
Anti-Pattern: Ignoring Warnings
What it looks like: Build output shows 47 warnings. Developer says "those have always been there."
Why wrong: Warning count creep is how technical debt compounds. Each warning was once new. Nobody investigated it. Now there are 47 and nobody reads them. The 48th warning — the one that actually matters — is invisible.
Instead: Treat warning count as a ratchet. Today's count is the ceiling. New warnings must be resolved or explicitly acknowledged before merge. Track warning count in CI and fail if it increases.
Anti-Pattern: Size Limits Without Context
What it looks like: "Our bundle limit is 500KB" applied uniformly to every page.
Why wrong: A landing page and an admin dashboard have different size budgets. The landing page needs to be fast for first-time visitors on mobile. The admin dashboard is used by logged-in users on desktop who will tolerate a larger initial load.
Instead: Per-route or per-entry-point size budgets. Use tools like bundlesize, size-limit, or next/bundle-analyzer to set granular limits.
Quality Checklist
Before declaring a build verified:
Build completed with exit code 0
Zero TypeScript/compilation errors in output log
All expected route HTML files exist in output directory
All asset references (src, href) resolve to real files in output
No case-sensitivity mismatches between references and filenames
Bundle sizes within established budgets (JS, CSS, images)
sitemap.xml is well-formed and all URLs correspond to real pages
robots.txt exists and references the sitemap
favicon.ico and meta assets present
No new warnings introduced (warning count stable or decreased)
Output directory contains no unexpected large files (>1MB)
If static export: internal links all resolve to HTML files
If Cloudflare/Vercel/Netlify: platform-specific files present (_redirects, _headers, _worker.js)
Output Contract
This skill produces:
Build verification report listing pass/fail/warn for each check category
Size diff comparing current build against baseline (if baseline exists)
Broken reference list with file:line locations of each broken asset reference
Route coverage report mapping expected routes to their HTML file presence
Actionable fix list prioritized by severity (FAIL items first, then WARN)
References
references/build-manifest-schema.md -- JSON schema for build baseline manifests used in comparisons
references/framework-output-maps.md -- Expected output structures for Next.js, Vite, Astro, and Cloudflare Pages
