· 
· 
What is this
omnistack-agent is an open-source, platform-agnostic system prompt that turns any capable AI model into a Full Stack Software Engineering Specialist — one agent that fluidly takes on whichever engineering role a task needs, with object-oriented design as its default lens. The agent's "brain" is authored once in a single source (core/ + knowledge/) and compiled by a tested, zero-dependency Node script into ready-to-use adapter files for every major AI platform. You don't write prompts — you copy a file.
It can take on all ten of these roles:
- Software Architect — system structure, boundaries, and trade-offs (ADRs included).
- Full Stack Developer — end-to-end features across UI, API, and data.
- Mobile Developer — native and cross-platform apps, offline, and push.
- Backend Engineer — domain logic, services, jobs, and data integrity.
- Frontend Engineer — accessible, performant UIs and predictable state.
- Database Administrator — schemas, indexing, migrations, and tuning.
- DevOps Engineer — CI/CD, Infrastructure as Code, and rollbacks.
- QA Engineer — test plans, automated suites, and sharp bug reports.
- Technical Writer — READMEs, API references, and architecture docs.
- Software Mentor — the why behind the code, with runnable examples.
▶️ How to use
Pick your platform, copy the listed file's contents, and paste it where that platform expects its instructions. No build step is required to consume the agent — the adapters are pre-generated and committed.
| Platform | File to copy | How to install |
|---|---|---|
| ChatGPT (Custom GPT) | adapters/chatgpt/custom-gpt-instructions.md | Create a new GPT → open Configure → paste the file into the Instructions box (this is the lean variant). For the complete knowledge inlined (larger; best used as an API system prompt), use adapters/chatgpt/system-prompt.md. |
| Claude (Skill) | adapters/claude/SKILL.md | Drop the file into your project's skills folder (e.g. .claude/skills/omnistack-agent/SKILL.md); it ships with YAML frontmatter and is user-invocable. |
| Claude (Agent) | adapters/claude/agent.md | Register it as a subagent in .claude/agents/; the frontmatter already sets the name, description, and model. For repo-wide guidance instead, use adapters/claude/AGENTS.md. |
| GitHub Copilot | adapters/copilot/copilot-instructions.md | Save it as .github/copilot-instructions.md at your repository root → reload Copilot. |
| Gemini (Gem) | adapters/gemini/gem-instructions.md | Create a new Gem in Gemini → paste the file into the Instructions field → save. |
| Cursor / Windsurf | adapters/cursor/AGENTS.md | Place it as AGENTS.md at your project root so the editor picks it up automatically. |
| Generic (any LLM) | adapters/generic/system-prompt.md | Paste the file as the system prompt of any chat or API request (OpenAI, Anthropic, local models, etc.). |
Full per-platform walkthroughs with screenshots-worthy detail live in
docs/platforms.md.
🤝 How to contribute
Contributions are welcome — new knowledge modules, language seeds, fixes, and translations all help. There is one golden rule:
Edit
core/orknowledge/— never editadapters/. The adapter files are generated. Hand-edits are overwritten by the next build and rejected by CI.
The workflow:
- Edit the source. Change a numbered file in
core/, or add/update a module underknowledge/(and link it inknowledge/_index.md). - Rebuild the adapters. Run
npm run buildto regenerate every file underadapters/. - Verify. Run
node --test(unit tests) andnpm run validate(confirms the adapters match the source). - Open a PR. CI runs
npm run validate, so a PR fails if the committed adapters drift fromcore/+knowledge/. Always commit the regenerated adapters alongside your source change.
Requirements: Node ≥ 18, zero npm dependencies. See CONTRIBUTING.md for the full guide, including the knowledge-module template and commit conventions.
🗂️ Repository structure
omnistack-agent/
├── core/ # The agent's brain (single source): identity, principles,
│ # capabilities, workflow, interaction style, guardrails.
├── knowledge/ # Modular knowledge base — one topic per Markdown file,
│ # indexed by knowledge/_index.md.
├── adapters/ # GENERATED per-platform output. Do not edit by hand.
├── scripts/ # Zero-dependency Node build (build.mjs), drift-detecting
│ # validate (validate.mjs), pure lib + tests.
├── docs/ # Guides: architecture, adding knowledge, platforms.
└── assets/ # Banner and other static media.
🛣️ Roadmap
- More languages and frameworks in
knowledge/(Python, Go, Rust, Vue, Angular, .NET MAUI, and more). - Additional platform adapters as new AI tools emerge.
- A curated
examples/folder with real prompts and the agent's responses. - A full Brazilian-Portuguese translation of the knowledge modules (the README and docs are already bilingual).
- Deeper modules per domain, expanding the seeds into complete references.
📄 License
Released under the MIT License — free to use, modify, and distribute. See LICENSE.