Security Skill for Claude

Elixir/Phoenix Security Reference

> Ash projects: AshAuthentication has its own strategy/token patterns — use the ash-framework skill. CSRF, XSS, and secret management patterns below still apply.

Quick reference for security patterns in Elixir/Phoenix.

Iron Laws — Never Violate These

• VALIDATE AT BOUNDARIES — Never trust client input. All data through changesets
• NEVER INTERPOLATE USER INPUT — Use Ecto's ^ operator, never string interpolation
• NO String.to_atom WITH USER INPUT — Atom exhaustion DoS. Use to_existing_atom/1
• AUTHORIZE EVERYWHERE — Check in contexts AND re-validate in LiveView events
• ESCAPE BY DEFAULT — Never use raw/1 with untrusted content
• SECRETS NEVER IN CODE — All secrets in runtime.exs from env vars
• LIVEVIEW EVENT PARAMS ARE UNTRUSTED — Users can alter forms, hooks, and every phx-value-* in DevTools. Validate and authorize against server-side state before acting

Quick Patterns

Timing-Safe Authentication

def authenticate(email, password) do
  user = Repo.get_by(User, email: email)

  cond do
    user && Argon2.verify_pass(password, user.hashed_password) ->
      {:ok, user}
    user ->
      {:error, :invalid_credentials}
    true ->
      Argon2.no_user_verify()  # Timing attack prevention
      {:error, :invalid_credentials}
  end
end

LiveView Authorization (CRITICAL)

# `id` is client input even when it came from phx-value-id.
# RE-AUTHORIZE IN EVERY EVENT HANDLER
def handle_event("delete", %{"id" => id}, socket) do
  post = Blog.get_post!(id)

  # Don't trust that mount authorized this action!
  with :ok <- Bodyguard.permit(Blog, :delete_post, socket.assigns.current_user, post) do
    Blog.delete_post(post)
    {:noreply, stream_delete(socket, :posts, post)}
  else
    _ -> {:noreply, put_flash(socket, :error, "Unauthorized")}
  end
end

Rendered LiveView events can expose IDs in HTML and websocket payloads. That is

not automatically a vulnerability: treat IDs as public identifiers, never as

proof of access. Use opaque references only when the identifier itself must not

be disclosed, and still perform server-side authorization.

SQL Injection Prevention

# ✅ SAFE: Parameterized queries
from(u in User, where: u.name == ^user_input)

# ❌ VULNERABLE: String interpolation
from(u in User, where: fragment("name = '#{user_input}'"))

Quick Decisions

What to validate?

• All user input → Ecto changesets
• File uploads → Extension + magic bytes + size
• Paths → Path.safe_relative/2 for traversal
• Atoms → String.to_existing_atom/1 only

What to escape?

• HTML output → Auto-escaped by default (<%= %>)
• User HTML → HtmlSanitizeEx with scrubber
• Never → raw/1 with untrusted content

Anti-patterns

| Wrong | Right |

|-------|-------|

| "SELECT * FROM users WHERE name = '#{name}'" | from(u in User, where: u.name == ^name) |

| String.to_atom(user_input) | String.to_existing_atom(user_input) |

| <%= raw @user_comment %> | <%= @user_comment %> |

| Hardcoded secrets in config | runtime.exs from env vars |

| Auth only in mount | Re-auth in every handle_event |

| Trusting phx-value-* or hidden IDs | Load server-side state and authorize it |

References

For detailed patterns, see:

• references/authentication.md - phx.gen.auth, MFA, sessions
• references/authorization.md - Bodyguard, scopes, LiveView auth
• references/input-validation.md - Changesets, file uploads, paths
• references/security-headers.md - CSP, CSRF, rate limiting, headers
• references/oauth-linking.md - OAuth account linking, token management
• references/rate-limiting.md - Composite key strategies, Hammer patterns
• references/advanced-patterns.md - SSRF prevention, secrets management, supply chain