Rails Agent Skills MCP server

Rails Agent Skills

A catalog of 28 Rails skills and 9 personas that teach agents how to test, implement, and review Rails work. Process skills (TDD gates, YARD, DDD, service objects) live in ruby-core-skills. Install both.

Write the test → run it → confirm it fails for the right reason → implement → confirm it passes

flowchart TB
  subgraph thisRepo[rails-agent-skills]
    atomics[28 atomics]
    personas[9 personas]
  end
  core[ruby-core-skills]
  thisRepo --> core

Also in the same ecosystem: hanakai-yaku, agnostic-planning-skills, agent-mcp-runtime, ruby-skill-bench.

Daily loop

flowchart LR
  A[Task] --> B[load-context]
  B --> C[plan-tests]
  C --> D[write-tests RED]
  D --> E[implement GREEN]
  E --> F[docs + review]
  F --> G[PR]

Name the persona when you want the whole chain: tdd, bug-fix, graphql, migration, background-job, engine, review, quality, setup.

See docs/persona-guide.md and docs/personas/.

Catalog

Core skills (install separately): triage-bug, respond-to-review, write-yard-docs, create-service-object, integrate-api-client, define-domain-language, model-domain, skill-router, and the process skills.

Full list: docs/reference/skill-catalog.md. Gaps: docs/reference/gaps.md.

Install

There is no root SKILL.md. Each folder under skills/ is its own skill, so the CLI can prompt for all or one.

# picker: all skills, or a subset
npx skills add igmarin/ruby-core-skills
npx skills add igmarin/rails-agent-skills

# all skills, skip prompts
npx skills add igmarin/ruby-core-skills --skill '*'
npx skills add igmarin/rails-agent-skills --skill '*'

# one skill
npx skills add igmarin/rails-agent-skills --skill load-context

Or with GitHub CLI v2.90.0+ (gh skill):

gh skill install igmarin/ruby-core-skills
gh skill install igmarin/rails-agent-skills
gh skill install igmarin/rails-agent-skills code-review --scope project

A PromptScript global-install warning from skills.sh is harmless. The skills still install for other hosts.

Docs

Contributing

• Artifacts in English unless the user asks otherwise.
• Keep the tests-gate rule on every code-producing skill.
• description is when + triggers (≤ 600 chars). Procedure stays in the body.
• Keep public docs in sync with directory.json.

Thanks to Mumo Carlos (@mumoc).

Agent composition and compatibility

Shortened. The full README is on GitHub.