SOTA Ruby (2026)

SkillDatabases & data

State-of-the-art Ruby engineering rules (2026 baseline, Ruby 3.4+ / 4.0) that Claude applies when writing or auditing Ruby. Covers modern idioms (frozen string literals, pattern matching, Data/Struct, RBS/Sorbet/Steep typing), security (SQL injection via ActiveRecord/Sequel, ERB/XSS escaping, mass assignment, CSRF, Marshal/YAML deserialization, command injection, ReDoS), framework-neutral web hardening (Rails/Sinatra/Hanami as neutral examples), supply chain and tooling (Bundler lockfile and checksums, bundler-audit, RuboCop/StandardRB, Brakeman, RSpec/Minitest, CI gates), and concurrency/performance (GVL, threads vs fibers vs Ractors, background-job idempotency, YJIT/ZJIT, GC and memory, N+1 queries). Trigger keywords: Ruby, gem, Gemfile, bundler, Rails, Sinatra, Hanami, Rack, ERB, ActiveRecord, Sequel, RSpec, minitest, RuboCop, Sorbet, RBS, Sidekiq, YJIT, Ractor, rake, ruby-lang. Use for BOTH building Ruby services/gems/CLIs and reviewing or auditing Ruby codebases.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the SOTA Ruby (2026) skill

What this skill tells your AI

The instructions your AI receives, as published by martinholovsky/sota-skills in skills/sota-ruby/SKILL.md and read by ahel’s review.

Expert-level rules for producing and auditing production Ruby. Baseline language line: Ruby 3.4+, with Ruby 4.0 (released 2025-12-25) as the latest major line. Per the official branches page: 4.0 and 3.4 are in normal maintenance; 3.3 is security-maintenance only (expected EOL 2027-03); 3.2 and older are EOL (3.2 since 2026-04-01) — running them is itself a finding. Feature notes: Data.define and Regexp.timeout from 3.2, it block parameter and chilled-string warnings from 3.4, Ractor::Port and experimental ZJIT from 4.0 — noted where relevant. Every rules file ends with an audit checklist of grep/lint patterns.

Purpose

Two consumers, one source of truth:

  • BUILD mode — generating new Ruby code: follow the rules as defaults, not suggestions. Deviate only with an explicit comment justifying it.
  • AUDIT mode — reviewing existing Ruby code: hunt violations using the audit checklists, classify by severity, report in the finding format below.

BUILD mode

  1. Before writing code, read the rules files relevant to the task (see index). A web endpoint touching the DB and a background job needs 02, 03, 05.
  2. Apply the top-10 non-negotiables (below) unconditionally.
  3. Establish context first: .ruby-version, Gemfile/Gemfile.lock, RuboCop or StandardRB config, framework and test runner in use. Match the project's floor (no it block param on a 3.3 project).
  4. New projects: pin the Ruby version (.ruby-version), commit Gemfile.lock, add RuboCop or StandardRB, bundler-audit, and the test suite to CI from day one (see rules/04). Rails apps add Brakeman.
  5. Security posture is non-optional even when unrequested: parameterized SQL, argv-form process spawning, YAML.safe_load semantics, SecureRandom, escaped output (see rules/02, rules/03).
  6. Write tests alongside the code (RSpec or Minitest — match the project). Anything with threads or jobs gets an idempotency/concurrency test.
  7. When code must violate a rule for a legitimate reason, leave a # NOTE(sota): comment explaining the trade-off so auditors don't flag it.

AUDIT mode

Work through each relevant rules file's audit checklist against the target repo. Run the listed grep/lint commands; confirm each hit manually before reporting (greps are recall-oriented, expect false positives). Useful mechanical sweeps: bundle exec rubocop, bundler-audit check --update, brakeman -q (Rails), plus the per-file greps.

Severity conventions

SeverityMeaningExamples
CRITICALExploitable now, or data lossSQL built with #{} interpolation, Marshal.load/YAML.unsafe_load on external data, command injection via backticks with user input, html_safe on user input
HIGHExploitable with preconditions, or production-breakingMissing CSRF protection on state-changing routes, permit!, ^/$ anchors in validation regexes, rand for tokens, Timeout.timeout around DB work, EOL Ruby in production
MEDIUMCorrectness/maintenance hazard, latent bugN+1 on a hot path, non-idempotent retried jobs, no Gemfile.lock in an app, mutable shared state across threads without a lock, rescue Exception
LOWDeviation from SOTA, frictionMissing frozen-string-literal comments, Struct where Data fits, stringly-typed booleans, unpinned dev tooling
INFOWorth knowing, no action forcedYJIT not enabled, typing (RBS/Sorbet) absent, newer-Ruby features available after a floor bump

Finding format

file:line | rule violated (rules/NN §section) | severity | effort | fix

Severity: Critical / High / Medium / Low / Info. Effort: trivial / small / medium / large. Borderline severities state the deciding assumption; unconfirmed findings are marked "needs verification", never asserted. Group findings by severity, CRITICAL first; end with counts per severity, the three highest-leverage fixes, and which checklists were run.

Rules index

FileRead this when...
rules/01-language-idioms.mdChoosing/verifying the Ruby version baseline; frozen string literals; pattern matching; Data vs Struct; exception design; nil over a sentinel, and to_i silently returning 0 for garbage; typing with RBS/Sorbet/Steep; general idioms and pitfalls
rules/02-security.mdAny input crossing a trust boundary: SQL injection (ActiveRecord/Sequel), command injection (system/backticks/Open3), deserialization (Marshal, YAML/Psych), ReDoS and regex anchors, eval/send/constantize, secrets and randomness, path traversal
rules/03-web-hardening.mdBuilding or auditing anything web-facing: XSS/ERB escaping, mass assignment and strong params, CSRF, sessions/cookies, security headers, open redirects, SSRF, file uploads — framework-neutral
rules/04-supply-chain-tooling.mdBundler and Gemfile.lock discipline, lockfile checksums, bundler-audit, RuboCop/StandardRB, Brakeman, RSpec/Minitest mechanics, CI gates, gem authoring/publishing
rules/05-concurrency-performance.mdThreads, fibers, Ractors, and the GVL; background-job idempotency; YJIT/ZJIT; GC and memory (allocator, RSS); N+1 detection; profiling workflow

Top-10 non-negotiables

  1. Supported Ruby only — 3.4+ in production (3.3 accepted short-term with an upgrade plan; ≤3.2 is a finding). Pin it in .ruby-version and CI. (rules/01)
  2. SQL only via parameterized queries — hash conditions or ?/named placeholders in ActiveRecord/Sequel; string-interpolated SQL is CRITICAL, no exceptions for "internal" values. (rules/02)
  3. Never Marshal.load, YAML.unsafe_load, or eval-family on data you don't fully control; YAML.load is safe-by-default only on Psych 4+ (Ruby 3.1+) — verify the runtime. (rules/02)
  4. Processes spawn with argv lists (system("cmd", arg), Open3.capture2), never a shell string containing external input; no Kernel#open/URI.open on user-supplied names. (rules/02)
  5. All HTML output escaped by default; every raw/html_safe is reviewed; non-Rails ERB configured to auto-escape. (rules/03)
  6. Mass assignment goes through an attribute allowlist (strong params / params.expect; explicit attribute lists elsewhere); permit! is HIGH. (rules/03)
  7. CSRF protection on every cookie-authenticated state-changing endpoint; \A/\z (never ^/$) to anchor validation regexes. (rules/02, rules/03)
  8. Gemfile.lock committed and CI installs frozen (BUNDLE_FROZEN=true); bundler-audit gates CI; git-sourced gems pinned to a SHA. (rules/04)
  9. SecureRandom (never rand/Random) for anything security-relevant; constant-time comparison for secrets. (rules/02)
  10. Background jobs are idempotent and enqueue after commit — at-least-once delivery and retries are the contract; jobs take IDs, not objects. (rules/05)

Signals

GitHub stars
22
Forks
4
Last commit
Sep 2026

ahel review

  • K1binfo
    installs-packages (in rules/04-supply-chain-tooling.md)

Automated review, not a security audit. Ruleset v1+k2.

Advanced
Catalog kind
skill
Gateway key
sota-ruby
Source
github.com/martinholovsky/sota-skills