Shipwright

AI agents, skills, and MCPs for automated software development. A universal plugin framework that works as a Claude Code plugin marketplace, a GitHub Actions CI/CD automation source, and installs natively into Cursor, Codex, and Claude Code via the ship CLI.

What is Shipwright?

Shipwright provides a catalog of specialized AI agents, reusable skills, and MCP integrations that automate the full software development lifecycle -- from issue triage to production release:

• Triage issues to determine actionability and request clarification
• Architect solutions with tradeoff analysis and Mermaid diagrams
• Implement changes following project conventions and security best practices
• Write tests with comprehensive edge case and integration coverage
• Security review code against OWASP Top 10 and threat models
• Review PRs for quality, correctness, and adherence to standards
• Generate changelogs with semantic versioning and audience-aware release notes
• Analyze codebases by tracing runtime paths, dependency graphs, and architectural boundaries

Each agent is deeply opinionated about how to do its job well -- not just "write code," but "write code with security awareness, proper naming, error handling, and scalability in mind." Agents bundle domain-specific skills so they carry their expertise with them.

Three Ways to Use Shipwright

Path 1: Ship CLI (Universal)

Install agents, skills, and MCPs into any supported AI tool (Cursor, Claude Code, Codex). The CLI writes to each tool's native directory structure:

Tool	Agents	Skills
Claude Code	.claude/agents/*.md	.claude/skills/<name>/SKILL.md
Codex	.codex/agents/*.toml	.agents/skills/<name>/SKILL.md
Cursor	.cursor/agents/*.md	.cursor/skills/<name>/SKILL.md

# Install the CLI
go install github.com/CaptShanks/shipwright/cli/cmd/ship@latest

# Install a plugin locally (current project, all tools)
ship install architect-agent

# Install for a specific tool
ship install architect-agent --target cursor

# Install globally
ship install shipwright-full --global

# Browse the marketplace
ship search
ship info pr-reviewer

# Manage installed plugins
ship list
ship uninstall triage-agent
ship update

# Manage MCP servers
ship mcp list
ship mcp install context7
ship mcp install mcp-atlassian --target cursor --global
ship mcp remove context7

Path 2: Claude Code Plugin (Local Development)

Install agents and skills directly into Claude Code:

# Add the marketplace (one time)
/plugin marketplace add CaptShanks/shipwright

# Install individual agents (skills are bundled automatically)
/plugin install security-reviewer@shipwright
/plugin install implementer-agent@shipwright

# Or install everything
/plugin install shipwright-full@shipwright

Path 3: GitHub Actions (CI/CD Automation)

Use reusable workflows to automate the issue-to-PR pipeline. Supports both Claude Code (via OAuth) and Codex (via API key):

# .github/workflows/ai-triage.yml
name: AI Triage
on:
  issues:
    types: [opened]

jobs:
  triage:
    uses: CaptShanks/shipwright/.github/workflows/ai-triage.yml@main
    with:
      ai-provider: claude
      project-context-path: .github/shipwright/project-context.md
    secrets:
      CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}

Reusable workflows available: ai-triage.yml, ai-implement.yml, ai-pr-review.yml.

Available Plugins

Agents (8)

Plugin	Description	Bundled Skills
triage-agent	Analyzes issues for actionability, requests clarification, classifies for downstream agents	--
architect-agent	Designs solution architecture with tradeoff analysis and implementation plans	security-awareness, scalability-resilience
implementer-agent	Implements solutions following architecture plans and project conventions	security-awareness, scalability-resilience, code-quality-fundamentals
test-engineer	Writes unit and integration tests with edge case coverage	code-quality-fundamentals, test-patterns
security-reviewer	Reviews code for vulnerabilities using OWASP knowledge and threat modeling	security-awareness, security-owasp
pr-reviewer	Reviews PRs for quality, correctness, and security with actionable feedback	security-awareness, code-quality-fundamentals, code-review-standards
changelog-generator	Produces semantically versioned, audience-aware changelogs from git history and PRs	--
codebase-analyzer	Produces structured analyses of runtime paths, architectural boundaries, and dependency graphs	--

Skills (10)

Skill	Description
security-awareness	Input validation, secret handling, least privilege, and defensive coding
code-quality-fundamentals	Naming conventions, SOLID principles, error handling, and code organization
scalability-resilience	Graceful degradation, idempotency, retry patterns, and failure domain analysis
security-owasp	OWASP Top 10 review guidance with concrete code patterns and severity bands
test-patterns	Testing strategies, mocking, fixture management, and the testing pyramid
code-review-standards	Code review methodology, severity classification, and actionable feedback
semantic-versioning	Breaking change detection, semver decision rules, and version bump classification
architecture-patterns	Architectural style identification, practiced vs aspirational architecture analysis
go-skills	Go development idioms, patterns, and conventions
terraform-development	Terraform/HCL module architecture, state management, provider patterns, and IaC security

Skills are standalone and usable by any agent or independently. Use the additional-skills input in CI/CD workflows to inject language/framework-specific skills, or add them to an agent's frontmatter for local development.

MCPs (4)

MCP	Description	Auth Required
context7	Up-to-date library and framework documentation	None
serena	Semantic code navigation and symbol-level editing	None
mcp-atlassian	Jira + Confluence integration	5 env vars
bitbucket-cloud	Bitbucket Cloud PR and repo management	3 env vars

Bundle

Plugin	Contents
shipwright-full	All 8 agents + all 10 skills + all 4 MCPs

Integrating with Your Repo

1. Create .github/shipwright/project-context.md describing your project (see Knowledge Base Format)
2. Add thin caller workflows (see Getting Started)
3. Optionally activate language/framework skills via additional-skills in workflows or agent frontmatter locally

Creating Custom Agents and Skills

• Creating Agents -- agent frontmatter, persona structure, and skill dependencies
• Creating Skills -- follows the Agent Skills specification

License

MIT