YepAPI/skills

RBAC Authorization

Rules

• Check permissions on EVERY route with middleware — never make authorization optional
• Deny by default: if no explicit permission is granted, access is denied
• Server-side enforcement: hiding UI elements is NOT security, check on the server
• Define role hierarchy: admin > editor > viewer — higher roles inherit lower permissions
• Resource-level checks: verify can(user, "edit", resource) — not just role-based access
• Audit all permission checks — log who accessed what and when, especially for admin actions
• Separate authentication (who are you?) from authorization (what can you do?)

// Permission check helper
type Action = "read" | "create" | "update" | "delete";
type Resource = { ownerId: string; [key: string]: any };

function can(user: { id: string; role: string }, action: Action, resource?: Resource): boolean {
  const permissions: Record<string, Action[]> = {
    admin: ["read", "create", "update", "delete"],
    editor: ["read", "create", "update"],
    viewer: ["read"],
  };
  const allowed = permissions[user.role];
  if (!allowed?.includes(action)) return false;
  // Resource-level check: editors can only modify their own resources
  if (resource && action !== "read" && user.role !== "admin") {
    return resource.ownerId === user.id;
  }
  return true;
}

// Express middleware guard
function requirePermission(action: Action) {
  return (req, res, next) => {
    if (!req.user) return res.status(401).json({ error: "Unauthenticated" });
    if (!can(req.user, action, req.resource)) {
      return res.status(403).json({ error: "Forbidden" });
    }
    next();
  };
}

app.put("/api/posts/:id", loadResource, requirePermission("update"), updatePost);

Avoid

• Checking permissions only in the UI — attackers call your API directly
• Allowing access unless explicitly denied — always deny by default
• Using a single isAdmin boolean — roles should be granular and extensible
• Forgetting resource-level checks — "editor" should not mean "can edit everyone's content"
• Skipping authorization on "internal" endpoints — they become public eventually