v0.2.2

Security and integrity

Creative frontend work touches secrets, third-party code, user input and public claims. Integrity (honest content and honest reporting) is part of quality.

Contents

  1. Secrets and configuration
  2. Client-side code is public
  3. Untrusted content and HTML sinks
  4. Forms, input and data handling
  5. Third-party scripts and embeds
  6. Links, navigation and redirects
  7. Dependencies
  8. Content integrity (no fabrication)
  9. Interaction integrity (no dark patterns)
  10. Reporting integrity

1. Secrets and configuration

  • Never print, copy, commit or paste the contents of .env*, key files or credentials. inspect_repo.mjs lists such files by name only.
  • Add new required configuration to .env.example with placeholder values and a comment.
  • If a task needs a secret value, ask the user to set it. Don't request it in chat unless they choose to share it.

2. Client-side code is public

  • Anything bundled for the browser is readable by anyone. Never put secrets in NEXT_PUBLIC_*, VITE_*, PUBLIC_* or inline scripts.
  • Call privileged APIs from the server (route handlers, server actions, edge functions), with the key kept server-side.
  • slop_scan.mjs flags likely client-side secrets (it never prints the matched value).

3. Untrusted content and HTML sinks

  • dangerouslySetInnerHTML, v-html, innerHTML, insertAdjacentHTML and document.write need trusted or sanitised input (for example DOMPurify), configured for the expected tags.
  • Markdown or CMS rich text is rendered with a sanitising renderer. Don't trust CMS editors with script injection.
  • Never build URLs, CSS or HTML strings from user input without encoding. Validate href protocols (allow http, https, mailto, tel).
  • JSON-LD and inline data: serialise with escaping of < to prevent breaking out of <script>.

4. Forms, input and data handling

  • Validate on the server; client validation is for usability only.
  • Collect the minimum personal data needed. Make consent explicit and non-pre-checked.
  • Rate-limit or protect public forms (honeypot, server-side checks). CAPTCHAs must have accessible alternatives (WCAG 3.3.8).
  • Don't log personal data to the console or analytics.

5. Third-party scripts and embeds

  • Each script is a supply-chain and privacy risk and a performance cost. Add only with a purpose, load async/defer, and prefer facades for embeds.
  • Use Subresource Integrity for static CDN scripts where versions are pinned. Use a Content-Security-Policy where the hosting allows; avoid unsafe-inline by using nonces or hashes.
  • Sandbox iframes (sandbox, allow lists) where the embed permits.
  • target="_blank" implies noopener in all modern engines. Adding rel="noopener noreferrer" is harmless. Use noreferrer when referrer privacy matters.
  • Validate redirect targets against an allow-list (no open redirects via ?next= parameters).
  • External links can be identified (icon or text) when leaving the site is significant.

7. Dependencies

  • Install from the official registry with the project's package manager; keep the lockfile.
  • Prefer well-maintained packages; check their weekly use, maintenance activity and licence.
  • Don't add packages that duplicate existing capability. Run the project's audit tooling if present (npm audit and similar), and report findings without auto-upgrading majors.

8. Content integrity (no fabrication)

Category C failures, never acceptable in deliverables:

  • Invented testimonials, reviews, quotes or ratings.
  • Fabricated client or partner logos, press mentions or awards.
  • Unsupported statistics ("10,000+ teams", "3× faster") without a source the client provided.
  • Generated or stock faces presented as real customers or team members.
  • Mock product screenshots presented as the real product.
  • Fake scarcity or urgency ("Only 2 left!", countdowns that reset).

Do instead: designed, labelled slots for missing proof; illustrative sample data labelled as sample; a request to the user for real assets in the report.

9. Interaction integrity (no dark patterns)

  • No pre-checked consent or upsells; no confirm-shaming copy; no disguised ads; no hidden costs revealed at the last step; no roach-motel cancellation flows.
  • The consent UI offers "reject" as easily as "accept" (and must meet local law; flag this for the client).
  • Visual hierarchy may emphasise the recommended action but must not hide or disguise alternatives.

10. Reporting integrity

  • Report only checks that ran, with their real output.
  • Separate what was verified, what was inferred and what was not tested.
  • Disclose demo or mocked behaviour and any placeholder content left in place.
  • Don't describe untested features as complete.