Skip to content

project-patterns.md

Latest commit

 

History

History
122 lines (89 loc) · 4.15 KB

project-patterns.md

File metadata and controls

122 lines (89 loc) · 4.15 KB
Raw

Project Patterns

Signal detection patterns for project type inference. Scan repo and match against these signals.

Frontend-Heavy

Primary signals (any 2+ → frontend project):

  • package.json with React, Vue, Angular, Svelte, Solid in dependencies
  • next.config.*, vite.config.*, webpack.config.*, turbo.json
  • tailwind.config.*, postcss.config.*
  • app/, pages/, components/, src/components/ directories
  • .tsx, .jsx, .vue, .svelte files > 50% of codebase
  • eslint, prettier configs

Secondary signals:

  • CSS-in-JS libraries (styled-components, emotion)
  • Storybook config (.storybook/)
  • Testing library (@testing-library/*)

Likely workflows: brainstorming, writing-plans, prototype, requesting-code-review

Backend-Heavy

Primary signals (any 2+ → backend project):

  • requirements.txt, go.mod, Cargo.toml, Gemfile, pom.xml, build.gradle
  • api/, routes/, handlers/, controllers/, services/, models/, middleware/
  • Database schemas: migrations/, prisma/, alembic/, *.sql files
  • docker-compose.yml with database services
  • ORM imports (SQLAlchemy, Prisma, GORM, Diesel, ActiveRecord)
  • Worker/queue systems: workers/, queues/, jobs/

Secondary signals:

  • gRPC proto files (*.proto)
  • GraphQL schemas (*.graphql)
  • Message queue configs (RabbitMQ, Kafka, Redis streams)

Likely workflows: test-driven-development, improve-codebase-architecture, verification-before-completion

Full-Stack

Primary signals (signals from both frontend + backend):

  • Monorepo with separate frontend/ + backend/ dirs
  • package.json + Python/Go/Rust backend files
  • Next.js with API routes + external backend

Likely workflows: tdd + prototype, subagent-driven-development (parallel frontend/backend work)

AI / Agent Project

Primary signals (any 2+ → AI agent project):

  • prompts/, agents/, tools/, memory/, skills/ directories
  • embeddings/, rag/, vector-store/, vectordb/
  • LLM SDK imports: openai, anthropic, @anthropic-ai/sdk, langchain, llamaindex
  • Agent framework: crewai, autogen, langgraph, semantic-kernel
  • .claude/ with substantial skill/hook configuration
  • Claude Code plugin or skill development

Secondary signals:

  • Token counting, context window management
  • Prompt templates (*.jinja, *.mustache)
  • Model routing / fallback logic
  • Streaming response handling
  • Function calling / tool use definitions

Likely workflows: brainstorming, writing-plans, systematic-debugging, writing-skills, claude-api

Production SaaS

Primary signals (3+ → production SaaS):

  • Payment integration: stripe, paddle, lemonsqueezy
  • Auth system: auth/, login/, sessions/, OAuth, JWT, bcrypt, next-auth
  • Multi-tenancy: tenants/, organizations/, workspaces/
  • Subscription management: plans/, billing/, quotas/
  • Admin dashboard: admin/, dashboard/
  • Email sending: sendgrid, resend, postmark
  • Production DB with backup configs
  • Monitoring: sentry, datadog, grafana, prometheus

Likely workflows: requesting-code-review, verification-before-completion, security-review

DevOps / Infrastructure

Primary signals:

  • .github/workflows/, .gitlab-ci.yml, Jenkinsfile
  • k8s/, helm/, terraform/, pulumi/, ansible/
  • Dockerfile, docker-compose.yml, docker-compose.*.yml
  • Cloud deploy configs: vercel.json, railway.toml, fly.toml, cloudbuild.yaml
  • Infrastructure as Code: *.tf, *.tfvars

Likely workflows: verification-before-completion (pre-deploy check)

Documentation Quality Heuristics

Weak documentation signals (any → documentation gap):

  • README.md < 200 chars
  • No docs/ directory
  • No architecture decision records (docs/adr/)
  • No CLAUDE.md or AGENTS.md
  • No API documentation
  • No onboarding / setup guide beyond "clone and run"

Strong documentation signals:

  • CLAUDE.md exists and is > 100 lines
  • docs/adr/ with multiple decisions
  • API reference docs (OpenAPI, Swagger)
  • Contributing guide

When documentation is weak: recommend grill-with-docs (if domain model exists) or writing-skills (to create CLAUDE.md).