Rails Security

Ruby on Rails Security Audit

Audit Rails applications (Rails 6, 7, 8).

When this skill applies

• Reviewing controllers, models, views
• Auditing strong parameters / mass assignment
• Reviewing ActiveRecord queries for injection
• Checking Devise / Pundit / CanCanCan setup
• Auditing secrets and credential management

Workflow

Follow ../_shared/audit-workflow.md.

Phase 1: Stack detection

grep -E "rails" Gemfile | head
bundle exec rails --version 2>/dev/null

Phase 2: Inventory

# Controllers
find app/controllers -name '*.rb' | head

# Models
find app/models -name '*.rb' | head

# Routes
cat config/routes.rb 2>/dev/null | head -100

# Initializers (security-relevant)
ls config/initializers/

# Brakeman recommended
which brakeman 2>/dev/null || echo "Install: gem install brakeman"

Phase 3: Detection — the checks

Strong parameters

• RLS-SP-1 Every controller action accepting params for create/update uses a permit allowlist:

``ruby def user_params params.require(:user).permit(:email, :name) # role, admin flags explicitly NOT in permit end ``

• RLS-SP-2 No params.permit! (allows everything — equivalent to no protection).
• RLS-SP-3 Nested attributes use permit(:foo, addresses_attributes: [:street, :city]) not permit!.

SQL injection (ActiveRecord)

• RLS-SQL-1 where("name = '#{params[:name]}'") is injection. Use placeholders:

``ruby User.where("name = ?", params[:name]) User.where(name: params[:name]) ``

• RLS-SQL-2 Dynamic ORDER BY: User.order(params[:sort]) — Rails 6+ raises on unknown columns, but the safe path is User.order(sort_column => sort_direction) with allowlisted values.
• RLS-SQL-3 find_by_sql("SELECT * WHERE id = #{params[:id]}") is injection. Use placeholders.
• RLS-SQL-4 Raw SQL via connection.execute reviewed.

Template XSS (ERB)

• RLS-XSS-1 ERB auto-escapes `. (double equals) and raw(...), htmlsafe, .htmlsafe` skip escaping.
• RLS-XSS-2 sanitize(html, ...) used for partial HTML; safer than raw.
• RLS-XSS-3 link_to(text, params[:url]) — Rails normalizes javascript: URLs in link_to? Confirm — depends on version; validate URLs server-side.
• RLS-XSS-4 content_tag(:a, name, href: params[:url]) does not validate URL.

CSRF

• RLS-CSRF-1 protect_from_forgery with: :exception in ApplicationController (Rails 5+ default; verify not removed).
• RLS-CSRF-2 API-only controllers (ActionController::API) skip CSRF by default — verify auth is token/JWT-based (not cookie).
• RLS-CSRF-3 No skip_before_action :verify_authenticity_token on state-changing endpoints unless replaced with equivalent (webhook signatures).
• RLS-CSRF-4 Rails 7+ Origin check by default for non-GET; verify config.

Authentication (Devise)

• RLS-AUTH-1 Devise modules appropriate: :database_authenticatable, :registerable, :recoverable, :trackable, :lockable, :timeoutable, :validatable.
• RLS-AUTH-2 :lockable enabled to prevent brute force.
• RLS-AUTH-3 Password reset tokens single-use and time-limited (Devise defaults reasonable).
• RLS-AUTH-4 Email enumeration prevented: same response for "email sent" whether or not the email exists.
• RLS-AUTH-5 before_action :authenticate_user! on protected controllers.

Authorization (Pundit / CanCanCan)

• RLS-AZ-1 Every controller action uses authorize @resource (Pundit) or load_and_authorize_resource (CanCanCan).
• RLS-AZ-2 Pundit's verify_authorized and verify_policy_scoped in after_action to catch missed authz.
• RLS-AZ-3 Policy scopes restrict index queries to current_user-visible.
• RLS-AZ-4 Default-deny: missing policies → 403.

class ApplicationController < ActionController::Base
  include Pundit::Authorization
  
  after_action :verify_authorized, except: :index
  after_action :verify_policy_scoped, only: :index
  
  rescue_from Pundit::NotAuthorizedError, with: :user_not_authorized
  
  private
  def user_not_authorized
    render plain: 'Forbidden', status: :forbidden
  end
end

Mass assignment via attribute hash

• RLS-MA-1 User.new(params[:user]) without strong params = mass assignment.
• RLS-MA-2 User.update(params[:user]) similarly.
• RLS-MA-3 attr_accessible / attr_protected (deprecated; pre-Rails 4) not relied on.

Open redirects

• RLS-OR-1 redirect_to params[:return_to] — validate the URL against allowlist or use redirect_to params[:return_to], allow_other_host: false (Rails 7+ default for safety).
• RLS-OR-2 Devise after_sign_in_path_for doesn't blindly use stored_location_for.

Insecure deserialization

• RLS-DES-1 Marshal.load(user_input) is RCE. Never deserialize untrusted data with Marshal.
• RLS-DES-2 YAML.load(user_input) (not safe) → RCE. Use YAML.safe_load.
• RLS-DES-3 JSON.load invokes Object.from_json and is unsafe; use JSON.parse.

Secrets and credentials

• RLS-SEC-1 config/credentials.yml.enc encrypted; master.key not committed (in .gitignore).
• RLS-SEC-2 Rails.application.credentials.secret_key_base set in production.
• RLS-SEC-3 Environment-specific credentials (config/credentials/production.yml.enc) used.
• RLS-SEC-4 No secrets in config/secrets.yml (deprecated) committed.

Cookies and sessions

• RLS-CK-1 Session store config secure: Rails.application.config.session_store :cookie_store, key: ..., secure: true, httponly: true, same_site: :lax.
• RLS-CK-2 Cookie store has size limit (4KB); if storing large data, use Redis/DB store.
• RLS-CK-3 Signed/encrypted cookies used for sensitive data (cookies.signed[:foo], cookies.encrypted[:foo]).

File uploads (Active Storage, CarrierWave, Shrine)

• RLS-UP-1 Active Storage's allowlist of content types if exposing direct uploads.
• RLS-UP-2 Active Storage variants and previews not run on untrusted content (ImageMagick CVEs).
• RLS-UP-3 S3 / blob URLs not directly user-controllable.

Rails Admin / ActiveAdmin

• RLS-ADM-1 Admin gem authentication enforced separately from app auth.
• RLS-ADM-2 Admin actions audited (paper_trail, audited gem).
• RLS-ADM-3 Admin URL changed from default /admin for security through obscurity.

Headers

• RLS-HDR-1 config.force_ssl = true in production.
• RLS-HDR-2 Custom CSP via config.content_security_policy block.
• RLS-HDR-3 secure_headers gem or Rails defaults for X-Frame-Options, X-Content-Type-Options.

Brakeman

• RLS-BR-1 Brakeman run in CI. brakeman --no-pager -A should pass with high/critical confidence findings = 0.
• RLS-BR-2 Brakeman ignores (config/brakeman.ignore) reviewed; false positives documented.

Dependencies

• RLS-DEP-1 bundle update recent; Gemfile.lock current.
• RLS-DEP-2 bundle audit clean.
• RLS-DEP-3 Rails version current LTS; old versions unpatched.

Phase 4: Triage

Critical: params.permit! on user model; YAML.load(user_input); raw SQL with string interpolation; CSRF disabled globally.

Phase 5: Report

Use ../_shared/findings-schema.md. Prefix IDs with RLS-.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: hlsitechio
• Source: hlsitechio/claude-skills-security
• License: MIT

Install and usage instructions live in the source repository linked above.