Foundation

Foundation Skills

Open-source agent skills that do a feasibility consultant's job: size the market, benchmark competitors, build the financial study, tune the pricing, then write the business plan and the pitch deck. They run in your own AI agent (Claude, Codex, Gemini CLI, Cursor and others). They are built by the team behind Foundation.

▶ Foundation in 48 seconds: a sample café project from its costs to its business plan, and an AI assistant testing it.

Why this exists

Ask a general AI assistant whether your business idea works and it will write you something that sounds like a feasibility study. The problems are underneath:

• The math is guessed. Language models are not calculators. NPV, IRR, payback and break-even figures in an AI-written plan are often wrong, and nothing on the page tells you which ones.
• The assumptions are invisible. Where did "£38,000 for the fit-out" come from? A lender will ask, and "the AI said so" is not an answer.
• The parts don't agree. The team section says five staff, the payroll line pays four, and the market chapter uses a different number from the pitch deck.

A professional consultant avoids all three. They put a source behind every input, run the numbers through a real financial model, and keep one version of the truth that every document draws from. These skills teach an AI agent to work the same way.

How it works

The skills split the job the way it should be split:

• The agent does the analyst's work: interviewing the founder, researching costs, prices and competitors, choosing a sensible structure, tagging where every number came from, and writing.
• A deterministic engine does the math: Foundation is a feasibility-study web app with a tested calculation engine. It computes investment, operating costs, revenue ramps, profit, cash flow, NPV, IRR, payback, break-even and SaaS metrics from the inputs. When your agent is connected to Foundation, every financial figure in the output comes from that engine. The agent never computes one itself.

The skills also work without a Foundation account. The agent still does the research and structures every input with its source, and never invents financial results. It then hands you a foundation-project.json file. Import that file into Foundation (Projects → Import) to get the computed study.

The skills

All five follow the same rules, written down in CONVENTIONS.md:

• Every input carries a source tag: founder, quote, benchmark or estimate.
• The agent asks you only what it can't research, in one round.
• Writing follows a guide based on Wikipedia's Signs of AI writing: plain sentences, no hype, no em dashes.
• For brand templates or heavier design, the Word and PowerPoint files can go through your environment's own document tools as a second step, with a script that checks no number changed on the way.

Install

Any agent that supports skills (Claude Code, Codex, Gemini CLI, Cursor and others):

npx skills add mamarexa/foundation-skills

Or copy the skill folders you want by hand:

The Word and PowerPoint skills need Python with python-docx, python-pptx and matplotlib. These come preinstalled in Claude's code sandbox. Anywhere else, run pip install python-docx python-pptx matplotlib.

Connect Foundation (optional, recommended)

1. Create a free account at getfndtn.com.
2. Add https://getfndtn.com/api/mcp to your agent as a remote MCP server (in Claude: Settings → Connectors → Add custom connector). You'll sign in with your Foundation account.
3. Ask your agent for a feasibility study. It will find the Foundation tools and work on your real project, and the numbers will match what you see in the app.

Step-by-step setup for Claude, ChatGPT, Codex, Cursor, Gemini CLI, Antigravity and VS Code is in the Foundation docs.

Try it

"I want to open a specialty coffee shop near Leeds station, about 30 seats, opening next

spring. Is it viable, and how much do I need to raise?"

"Size the market for scheduling software for UK dental practices, and benchmark the main

competitors."

"Turn my Foundation project into a business plan for a bank loan, then a 12-slide deck."

Contributing

This repository is meant to get better with use. The most valuable contributions are:

• Benchmarks with sources: cost ratios, startup-cost ranges, wages, employer payroll costs, rents and tax rates for your industry or your country. Local numbers help founders outside the US and UK most.
• Corrections from practitioners: if you do this work for a living and a step is wrong or missing, tell us how it's done in practice.
• Test cases: a realistic business description and what a good study of it should contain.

Good contributions also flow back into the Foundation app, for example as better default categories and benchmark ranges. See CONTRIBUTING.md for the rules. The short version: every number needs a source you'd be comfortable showing a lender.

This repository and the skills/ folder in the Foundation app are kept in sync automatically in both directions, so a merged pull request here shows up there (as a reviewed PR, not a silent merge) without anyone copying files by hand.

Tests

pip install python-docx python-pptx matplotlib

python tests/test_scripts.py

The fixture project in tests/fixtures/ is a sample café. Its facts file is real output from Foundation's engine.

License

MIT. By contributing, you agree your contribution is licensed the same way, including its use in Foundation.