Agentic development flow for Rails 8 projects — /feature, /fix, /review, /setup-flow and /brain commands, fourteen specialist subagents (rails-developer, migration-writer, code-reviewer, spec-reviewer, test-runner, security-auditor, design-auditor, doc-updater, pr-reviewer, claim-verifier, skill-curator, claude-skills-reporter, adversary, plan-advisor), and guardrail hooks that mechanically enforce spec-proven changes, safe migrations and git discipline. Pairs with the rails-stack skills.
26 commands · 14 agents · hooks
Agentic UI/design flow — applies the design system so UI is consistent, modern, and responsive without a designer or Figma. /design-flow:setup scaffolds the token architecture (brand primitives -> semantic roles -> Utopia fluid scale) + layout-primitive recipes + base components through the project's UI adapter; /design-flow:component authors UI by composing layout primitives + role tokens with a fixed variant/size/state vocabulary, a11y, and prescribed responsive behavior; /design-flow:canvas composes a Claude Design prompt carrying the project's own tokens and catalog, /design-flow:port brings the returned canvas back through the project's UI adapter; /design-flow:audit flags drift. Agents: rails-ui-composer, design-auditor, design-critic, rails-ui-porter, brand-guardian, interface-auditor. Pairs with the design-system skill (bundled in rails-stack).
13 commands · 6 agents · hooks
Independent, stack-agnostic QA engineering flow. The QA engineer picks the stack in qa/qa.config.yml (Playwright, Cypress+Cucumber, Selenium+pytest-bdd, Appium; free by default) — qa-flow is never forced onto one. Free, repo-local case authoring/management (/qa-flow:cases → qa/test-cases.csv; Testmo opt-in) and agentic functional testing via Playwright MCP (/qa-flow:functional): auto-maps flows, self-adapting locators, evidence-based Markdown/CSV reports. Plus risk-based planning, defect filing, PR-native results, and a hook-enforced dev->main certification gate.
8 commands · 11 agents · hooks
Rails 8.1 + Hotwire development doctrine: vanilla-first stack, pure RSpec testing, Turbo/Stimulus/Hotwire Native, OpenAPI docs via rswag, ruby_llm AI features, Kamal 2 deployment, plus the design system (tokens, layout primitives, component catalog), review doctrine for the defect class authors are blind to — code that does not do what its own docs, config and project rules claim — and an advisory quality pass (reuse, simplification, efficiency, altitude) that runs second and never blocks a merge. Bundles the rails-8, hotwire, design-system, code-review, quality-pass, derived-artifacts and parallel-session-lane skills.
7 skills · 7 agents · hooks
Lifecycle orchestrator over rails-flow and qa-flow — build -> verify -> certify -> containerized release, honoring every gate. Release artifact is a versioned ghcr.io Docker image (local pull-and-run today, Kamal cloud when ready). /pipeline:deploy-cloud reads a prepared .env and generates the full Kamal 2 config + wires secrets safely (committed deploy.yml references names; gitignored .kamal/secrets holds values), then deploys with approval.
9 commands · 1 agent · hooks
Next.js + Supabase + Vercel stack, 0.1.0, an early release: conventions for how the three fit together, the stack's markers for the stack detector, and a React frontend architecture (the react skill) whose rules are each enforced by lint, nextjs-stack's checks, a declaration or a recorded review verdict. feature/fix/review still lean on the Rails-flavoured agents. Platform knowledge stays in Vercel's official plugin (vercel-plugin) and Supabase's supabase/agent-skills, which it points to rather than copies.
2 skills
Deliberately generic Rails guidance. This is the benchmark's CONTROL arm, not a product — it exists so a win for rails-stack can be attributed to the content of our doctrine rather than to the mere presence of a loaded skill. Never publish or install this.
1 skill
One install for rails-flow, qa-flow and pipeline: a bundle whose manifest declares the three as dependencies (a manifest only, no components of its own).