Skip to content
AletheIA Docs
Esc
navigateopen⌘Jpreview
On this page

Web App Security Review Checklist

Use this checklist when a Work Slice changes a web app, API, integration, model context path, logging behavior or security-sensitive client/server boundary.

1. Client/server authority

  • Does the client only request/display, or does it decide sensitive behavior?
  • Are price, permission, eligibility, payment, quota or critical transitions verified by an authoritative layer?
  • Is UI visibility being mistaken for authorization?

2. Secrets and config

  • Could secrets appear in frontend bundles, public config, logs, traces, examples or handoffs?
  • Are public and secret configuration paths separated?
  • Are missing checks marked unavailable instead of assumed safe?

3. Auth and data isolation

  • Is authorization explicit and source-backed?
  • Is tenant/user isolation enforced outside the UI?
  • Could a broader query, retrieval path or cache expose another tenant/user?

4. Model/agent context scoping

  • Is context scoped before inference?
  • Does a model or agent receive only context it is allowed to see?
  • Are external/retrieved contents classified as evidence/context rather than authority?

5. Inbound integrations

  • Are state-changing webhooks or inbound events verified?
  • Is idempotency/replay considered?
  • Are unverified events prevented from committing state?

6. Logs, traces and artifacts

  • Are sensitive payloads minimized/redacted?
  • Do handoffs/learnings/examples avoid copying secrets or restricted content?
  • Are operational artifacts reviewed as possible leak paths?

7. Untrusted rendering/content

  • Is external/user-generated content treated as untrusted?
  • Are content, evidence and commands separated?
  • Is rendering safety reviewed for user-visible and operator-visible surfaces?

8. Decision posture

Choose one:

  • proceed — required boundaries are source-backed and low-risk.
  • proceed_with_conditions — bounded follow-up is required.
  • require_human_review — evidence is missing or risk is high.
  • block — clear secret exposure, cross-tenant leak or unauthorized critical logic.

Was this page helpful?