ai-engineering-os
v2.0.0
Published
Model-independent engineering workflow installer for Codex, Claude Code, Cursor, Cline, Aider and OpenCode
Readme
AI Engineering OS
AI Engineering OS turns AI-assisted coding into a scoped, verifiable engineering workflow. It installs a small routing layer plus task-specific engineering guidance for Codex, Claude Code, Cursor, Cline, Aider, and OpenCode.
Supported product stack
- Mobile: React Native, Expo, and TypeScript
- Web: React, Next.js, TypeScript, and Tailwind CSS
- Backend: Python and FastAPI
- Data: PostgreSQL and Supabase
- Storage: Google Cloud Storage
- API verification: automated tests and Postman
The installer detects only these technologies. Existing repository conventions take precedence over greenfield defaults.
What it does
- Classifies work as lean, standard, or full
- Loads only the pipeline, roles, presets, and gates required by the task
- Detects applications and framework versions in monorepos
- Preserves repository rules and unrelated user changes
- Requires evidence before tests or quality gates can be reported as passed
- Keeps API consumers, automated tests, and Postman contracts aligned
Install
Run once from the repository root:
npx ai-engineering-os@latest init .Reload the editor or agent session, then describe the task normally:
Inspect this project and add profile photo upload.Adaptive context
The installed tool instruction initially loads only .ai/router.md.
lean: text, local styling, renames, explanations, or one-file behavior-preserving editsstandard: bug fixes, refactors, endpoints, screens, and normal featuresfull: authentication, personal data, migrations, storage, production, release, or cross-application work
The router expands context only when the task requires it.
Monorepos
Install once at the repository root. The detector scans up to four directory levels for supported manifests and entry points, then records every detected application in .ai/activeProfile.md.
product/
├── AGENTS.md
├── .ai/
├── products/controlCenter/
├── clients/fieldExperience/
└── services/coreApi/Directory names are not used as technology evidence.
Optional growth pack
Executive, growth, and conversion guidance is intentionally excluded from the default engineering workflow.
npx ai-engineering-os@latest init . --packs=growthCommands
npx ai-engineering-os@latest inspect .
npx ai-engineering-os@latest init . --tools=cursor
npx ai-engineering-os@latest init . --packs=growth
npx ai-engineering-os@latest init . --force
npx ai-engineering-os@latest doctor .inspect: reports detected applications without writing filesinit: installs the workflow--tools: limits editor and agent connections--packs: installs optional guidance packs--dryRun: previews writes--force: refreshes managed files and removes legacy layout entriesdoctor: checks the installed core and tool connections
Installed structure
.ai/
├── router.md
├── orchestrator.md
├── activeProfile.md
├── projectProfile.md
├── projectOverrides.md
├── governance/
├── orchestration/
├── projectManagement/
├── design/
├── development/
├── testing/
├── security/
├── operations/
└── documentation/Package documentation and unused adapter guides are not copied into the target repository.
Principles
- Repository evidence before assumptions
- The smallest complete change
- Verification proportional to risk
- No passed claim without evidence
- camelCase for new identifiers, with PascalCase where language conventions require it
- No source-code comments; intent is expressed through naming and structure
- Product-specific design instead of generic AI aesthetics
Documentation
- Quick start
- CLI behavior
- Package manifest
- Task router
- Orchestrator
- Default product profile
- Behavioral evaluations
