Skip to content
cccskills
FreePublished on GitHub

fastly-compute-well-known-spa-fallback

Fix .well-known files (apple-app-site-association, assetlinks.json) being served as HTML by the @fastly/compute-js-static-publish SPA fallback instead of JSON. Use when: (1) iOS Universal Links or Android App Links are broken because the verification files return text/html instead of application/json, (2) PublisherServer with spaFile config intercepts /.well-known/ paths and returns index.html (200, text/html) instead of 404 for missing files, (3) apple-app-site-association (no file extension) gets wrong content type even when correctly stored in KV. Covers both apex domain and subdomain handler patterns.

The English summary is being prepared. Showing the original description.

See how to install

Files included (1)

  • SKILL.md9.9 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

Fastly Compute: .well-known Files Intercepted by SPA Fallback

Problem

When using @fastly/compute-js-static-publish with SPA fallback configured (spaFile: "/index.html"), the PublisherServer.serveRequest() returns index.html with status 200 and Content-Type text/html for ANY path not found in the KV store. This includes /.well-known/apple-app-site-association and /.well-known/assetlinks.json, which iOS and Android require to be served as application/json.

Symptoms:

  • iOS Universal Links don't work (Apple's verification fetches .well-known/apple-app-site-association and gets HTML)
  • Android App Links don't work (Google's verifier fetches .well-known/assetlinks.json and gets HTML)
  • curl -I https://yourdomain.com/.well-known/apple-app-site-association shows Content-Type: text/html
  • The files ARE published to KV (confirmed by npm run fastly:publish) but still return HTML

Non-Obvious Root Causes

  1. SPA fallback returns 200, not 404: The publisher's SPA mode returns index.html with HTTP 200 for missing paths. You cannot distinguish "file served" from "fallback served" by status code alone - you must inspect the Content-Type.

  2. apple-app-site-association has no file extension: The static publisher infers MIME type from file extension. With no extension, it cannot detect application/json, so even when the file IS published to KV, it may get application/octet-stream or be served incorrectly.

  3. includeWellKnown: true in publish-content.config.js is necessary but not sufficient: It ensures the files are uploaded to KV, but doesn't prevent the SPA fallback from intercepting the requests, and doesn't fix the content-type for extension-less files.

  4. Both apex domain and subdomain handlers need the fix: If your Compute handler has separate code paths for subdomains vs apex, both paths must intercept .well-known/ requests before reaching the SPA fallback.

Solution

Step 1: Ensure Files Are Published

In publish-content.config.js, confirm includeWellKnown: true is set:

// publish-content.config.js
module.exports = {
  // ...
  includeWellKnown: true,   // Must be true to include /.well-known/ files
  // ...
};

Then publish static content:

npm run fastly:publish

Step 2: Intercept .well-known Paths Before SPA Fallback

In your Compute entry point (compute-js/src/index.js), add a .well-known handler BEFORE any call to publisherServer.serveRequest(request) that has SPA fallback enabled.

Critical guard: Check that the publisher response is NOT text/html - if it is, the SPA fallback fired (file not in KV), so return 404 instead of the HTML.

// In your main handleRequest function, BEFORE the final publisherServer.serveRequest() call:

// Handle .well-known requests (must come before SPA fallback)
if (url.pathname.startsWith('/.well-known/')) {
  // Handle NIP-05 or other dynamic .well-known endpoints first
  if (url.pathname === '/.well-known/nostr.json') {
    return handleNip05(url);  // Your custom handler
  }

  // For all other .well-known files: fetch from static publisher
  const wkResponse = await publisherServer.serveRequest(request);

  // CRITICAL: Guard against SPA fallback. The publisher returns index.html (text/html)
  // for files not in KV. We must detect this and return 404 instead.
  if (
    wkResponse != null &&
    wkResponse.status === 200 &&
    !wkResponse.headers.get('Content-Type')?.includes('text/html')
  ) {
    const headers = new Headers(wkResponse.headers);

    // Explicitly set correct content type.
    // apple-app-site-association has no extension, so the publisher may not detect JSON.
    const isJsonFile =
      url.pathname.endsWith('.json') ||
      url.pathname.endsWith('/apple-app-site-association') ||
      url.pathname === '/.well-known/apple-app-site-association';

    headers.set(
      'Content-Type',
      isJsonFile ? 'application/json' : (headers.get('Content-Type') || 'application/octet-stream')
    );
    headers.set('Cache-Control', 'public, max-age=3600');
    headers.append('Vary', 'X-Original-Host');  // If using multi-service routing

    return new Response(wkResponse.body, { status: 200, headers });
  }

  // File not in KV (or publisher returned SPA fallback) - return proper 404
  return new Response('Not Found', { status: 404 });
}

Step 3: Apply the Same Fix in Subdomain Handlers

If you have separate handling for subdomain requests, add the same guard there too. Subdomain paths hit a different code branch before reaching the apex domain handler:

if (subdomain) {
  if (url.pathname.startsWith('/.well-known/')) {
    if (url.pathname === '/.well-known/nostr.json') {
      return handleSubdomainNip05(subdomain);
    }

    // Same pattern: intercept, guard against SPA fallback, force JSON content type
    const wkResponse = await publisherServer.serveRequest(request);
    if (
      wkResponse != null &&
      wkResponse.status === 200 &&
      !wkResponse.headers.get('Content-Type')?.includes('text/html')
    ) {
      const headers = new Headers(wkResponse.headers);
      const contentType =
        url.pathname.endsWith('.json') || url.pathname.endsWith('/apple-app-site-association')
          ? 'application/json'
          : headers.get('Content-Type') || 'application/octet-stream';
      headers.set('Content-Type', contentType);
      headers.set('Cache-Control', 'public, max-age=3600');
      return new Response(wkResponse.body, { status: 200, headers });
    }
    return new Response('Not Found', { status: 404 });
  }

  // ... rest of subdomain handling
}

Verification

# Should return application/json, NOT text/html
curl -sI https://yourdomain.com/.well-known/apple-app-site-association | grep -i content-type

# Should return JSON body
curl -s https://yourdomain.com/.well-known/apple-app-site-association | head -c 100

# Android assetlinks.json
curl -sI https://yourdomain.com/.well-known/assetlinks.json | grep -i content-type

# Verify the SPA fallback guard works (path that does NOT exist in KV)
curl -sI https://yourdomain.com/.well-known/nonexistent-file
# Should return 404, not 200

Complete Working Example

From compute-js/src/index.js in divine-web:

// 4. Handle .well-known requests
if (url.pathname.startsWith('/.well-known/')) {
  // 4a. NIP-05 from KV store
  if (url.pathname === '/.well-known/nostr.json') {
    return await handleNip05(url);
  }

  // 4b. Serve other .well-known files (apple-app-site-association, assetlinks.json)
  // These must be served as JSON, not the SPA fallback.
  // apple-app-site-association has no file extension, so the static publisher
  // cannot detect its content type - we handle it explicitly here.
  const wkResponse = await publisherServer.serveRequest(request);
  // Guard: if publisher returns text/html, it's the SPA fallback, not the real file
  if (wkResponse != null && wkResponse.status === 200 && !wkResponse.headers.get('Content-Type')?.includes('text/html')) {
    const headers = new Headers(wkResponse.headers);
    // Ensure correct content type for app association files
    const contentType = url.pathname.endsWith('.json') || url.pathname.endsWith('/apple-app-site-association')
      ? 'application/json'
      : headers.get('Content-Type') || 'application/octet-stream';
    headers.set('Content-Type', contentType);
    headers.set('Cache-Control', 'public, max-age=3600');
    headers.append('Vary', 'X-Original-Host');
    return new Response(wkResponse.body, {
      status: 200,
      headers,
    });
  }
  // File not found in KV - return 404 instead of SPA fallback
  return new Response('Not Found', { status: 404 });
}

Deployment Checklist

After making code changes:

# 1. Publish static content first (uploads .well-known files to KV)
npm run fastly:publish

# 2. Deploy the edge worker code (with the .well-known interception logic)
npm run fastly:deploy

# NOTE: Order matters if files weren't in KV before. If you deploy code first,
# it will correctly return 404 for missing files. Then publish uploads the files.
# Either order works - the guard handles both cases.

Notes

  • This pattern applies to any Fastly Compute service using @fastly/compute-js-static-publish with spaFile configured.
  • The SPA fallback is intentional for client-side routing, but it breaks any path that needs a real 404 (like .well-known verification files).
  • The content-type detection by file extension is a fundamental limitation of static publishing - extension-less files always need explicit handling.
  • If you serve multiple domains (apex + subdomains), each code path that can call publisherServer.serveRequest() needs the .well-known interception guard.
  • After fastly:publish, allow up to 2-3 minutes for KV propagation before testing.

References

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

Fix ArgoCD ExternalSecret deployment failing with "namespace X is not permitted in project Y". Use when: (1) ExternalSecret shows OutOfSync in ArgoCD but won't sync, (2) ArgoCD application status shows "namespace X is not permitted in project 'infrastructure'", (3) ExternalSecret targets a namespace managed by a different ArgoCD project, (4) Using apps-of-apps pattern with separate infrastructure and application projects.

The English summary is being prepared. Showing the original description.

divinevideo/divine-mobile265Updated Oct 23, 2026

Art direction for any content — reads text, PDF, Word, HTML, PPT, then proposes 2-3 creative directions with photography style, mood, and visual language. After selection, generates AI image prompts and visual briefs section-by-section. Use when the user shares content and needs visual direction, image sourcing, or creative direction for any material.

The English summary is being prepared. Showing the original description.

divinevideo/divine-mobile265Updated Oct 23, 2026

Fix "Null check operator used on a null value" errors when an object is set to null during an async await. Use when: (1) Object reference is nullified while awaiting, (2) Code accesses object with ! after await returns, (3) Cancel/dispose operations run concurrently with async operations on same object. Solution: capture local reference before await.

The English summary is being prepared. Showing the original description.

divinevideo/divine-mobile265Updated Oct 23, 2026

Add custom metadata headers (x-amz-meta-*) to AWS v4 signed requests for GCS S3-compatible API. Use when: (1) Adding custom metadata to GCS uploads via S3 API, (2) Getting signature mismatch errors after adding new headers, (3) x-amz-meta-* headers being ignored or causing 403 errors. Custom headers MUST be included in canonical headers and signed headers list.

The English summary is being prepared. Showing the original description.

divinevideo/divine-mobile265Updated Oct 23, 2026

Fix password/secret authentication failures caused by trailing newlines when creating Google Cloud secrets (or similar) with bash here-strings. Use when: (1) Password authentication fails with correct password, (2) Secret created with `<<< "value"` syntax, (3) Error like "password authentication failed" or "invalid token" despite correct value. Bash here-strings (`<<<`) add a trailing newline that corrupts secrets.

The English summary is being prepared. Showing the original description.

divinevideo/divine-mobile265Updated Oct 23, 2026

Fix silent video/media processing failures caused by URL extraction code that filters on file extensions (.mp4, .webm, .webp). Use when: (1) Media moderation, transcoding, or analysis silently skips files from Blossom or content-addressed storage servers, (2) URL extraction from Nostr event tags (imeta, r tags) drops URLs without recognized extensions, (3) CDN fallback URLs append .mp4 but the actual server uses extensionless content-addressed paths like /{sha256}. Common in Nostr video events (kind 34236) where different clients use different URL formats.

The English summary is being prepared. Showing the original description.

divinevideo/divine-mobile265Updated Oct 23, 2026

See all skills by divinevideo

Report a problem with this skill