What Is This?

A collection of curated system prompts (we call them "profiles") that configure how AI coding tools behave for specific technology stacks.

Each profile is a Markdown file with battle-tested conventions, patterns, and guidelines for a given stack. When your AI reads it, it writes code the way a senior engineer on that stack would.

• 14 profiles: Java, Kotlin, Go, Python, Rust, Node.js, React, Vue, Android, iOS, Flutter, DevOps, Code Reviewer, Tech Consultant
• 14 knowledge bases: Design patterns, testing strategies, system design, code review guidelines, and more
• Memory architecture: Persistent context that survives across sessions
• OpenSpec integration: Spec-driven development workflow
• Lean CLAUDE.md: ~600 tokens pointing to profiles on demand, not 10k+ dumped upfront

These are not autonomous agents. They don't make decisions, don't run tools, don't loop. They're structured context that makes your existing AI tools (Claude Code, Cursor, Copilot, etc.) better at writing code for your stack.

Why Not Just Write Your Own CLAUDE.md?

You absolutely can. But most teams end up with one of two problems:

1. Too long (500+ lines, 10k+ tokens) — context stuffing that degrades model recall
2. Too generic ("write clean code, follow best practices") — no real signal

This playbook gives you:

• Stack-specific conventions curated per technology (not generic "best practices")
• Lean CLAUDE.md (~600 tokens) that points to detailed profiles on demand — following Boris Cherny's approach
• Knowledge bases the AI can pull from when it needs depth on patterns, testing, or system design
• Zero setup — npx eng-delivery-playbook and you're configured
• A starting point, not a straitjacket — customize from there

Quick Start

npx eng-delivery-playbook

This installs profiles, knowledge bases, and memory architecture into your project. Then:

eng-play switch java       # Switch to Java profile
eng-play switch py         # Use aliases
eng-play list              # Show all profiles

Spec-Driven Workflow

eng-play openspec start "
  context: We have a billing service.
  goals: Add usage-based billing API.
  requirements:
    - POST /api/billing/usage — record API calls per tenant
    - GET /api/billing/invoices/:tenant_id — list invoices
    - Pricing tiers: free (1k calls), pro (100k), enterprise (unlimited)"

The CLI auto-detects your stack, generates a structured prompt, and kicks off the OpenSpec workflow: PROPOSE > DESIGN > TASKS > IMPLEMENT > REVIEW > SHIP.

Profiles

Backend

| Profile | Stack | Alias | |---------|-------|-------| | Java | Spring Boot, JPA, Maven/Gradle | java | | Kotlin | Ktor, Spring, Coroutines | kt | | Go | stdlib, Gin/Chi, GORM/sqlx | go | | Python | FastAPI, Django, SQLAlchemy | py | | Rust | Actix/Axum, Tokio, SQLx | rs | | Node.js | Express, Fastify, NestJS, TypeScript | ts |

Frontend

| Profile | Stack | Alias | |---------|-------|-------| | React | React 19, Next.js 15, TanStack Query | next | | Vue | Vue 3, Nuxt 3, Pinia | nuxt |

Mobile

| Profile | Stack | Alias | |---------|-------|-------| | Android | Kotlin, Jetpack Compose, Hilt | droid | | iOS | Swift, SwiftUI, Combine | swift | | Flutter | Dart, Riverpod/Bloc, go_router | fl, dart |

Infrastructure & Review

| Profile | Purpose | Alias | |---------|---------|-------| | DevOps | K8s, Terraform, Docker, CI/CD | ops | | Reviewer | Tech-agnostic code review | review | | Consultant | Architecture advice (no code) | consult |

CLI Reference

eng-play                                      # Install (interactive)
eng-play switch <profile>                     # Switch profile
eng-play list                                 # List profiles
eng-play openspec start "<feature>"           # Start workflow (auto-detect)
eng-play openspec start "<feature>" <profile> # Start with specific profile
eng-play memory init                          # Initialize .AGENT/ memory
eng-play memory status                        # Show memory status

Auto-Detection

When you omit the profile from eng-play openspec start, the CLI scans your codebase:

| Signal | Profile | |--------|---------| | Cargo.toml | Rust | | go.mod | Go | | pom.xml, build.gradle | Java | | build.gradle.kts | Kotlin | | pyproject.toml, requirements.txt | Python | | pubspec.yaml | Flutter | | package.json + React/Next | React | | package.json + Vue/Nuxt | Vue | | Dockerfile, main.tf | DevOps |

Memory Architecture

The .AGENT/ directory provides persistent context across sessions with 5 memory types: working, procedural, semantic, episodic, and meta. Initialize with eng-play memory init.

See .AGENT/procedural_memory/PROFILES.md for details.

OpenSpec

OpenSpec provides spec-driven development — AI reads structured specs instead of guessing. Specs persist in the repo, enabling review of intent (spec deltas) alongside code diffs.

See profiles/knowledge/openspec.md for the full guide.

Examples

Coming soon: before/after examples and links to real PRs generated with this workflow.

If you've used this playbook in a project, open an issue — we'd love to feature it.

Changelog

3.0.0 (Breaking)

• Renamed "agents" to "profiles" — honest terminology, these are context configurations not autonomous agents
• Removed AI Metrics profile and knowledge base
• Rewrote README — shorter, honest positioning, no vanity badges
• Added CONTRIBUTING.md — proper contribution guide
• 14 profiles, 14 knowledge bases — curated per technology stack

2.1.x

• OpenSpec auto-install, structured prompt format

2.0.0

• eng-play CLI, auto-detect, lean CLAUDE.md

1.x

• Initial release through memory architecture

Contributing

See CONTRIBUTING.md.

License

MIT