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.
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.
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.
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.
- 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.
Released under the MIT License — free to use, modify, and distribute. See LICENSE.