Skip to content

Get started

These pages take you from an empty machine to an audited scientific result. The first four are the path; the last is for Antigravity specifically.

  • 1. Installation

    Install the Python package and check that the CLI works. Needs Python 3.10 or newer, and nothing else.

    Install

  • 2. Quickstart

    Scaffold the bundled Kepler experiment and recover Kepler's Third Law from live NASA data in about thirty seconds.

    Quickstart

  • 3. Your first project

    The full loop on a question of your own: goal, tools, hypothesis, workflow, agents, validate, run, audit, evolve.

    First project

  • 4. Coding-agent skills

    Install the seven research skills so Claude Code, Cline, Cursor, OpenCode or Antigravity can drive the loop for you.

    Skills

  • In Antigravity

    The Antigravity-specific path: install the plugin, confirm the skills loaded, and hand it a research question.

    Antigravity

What you are about to build

A Science ADK project is an ordinary git repository with an opinionated shape:

my-research/
├── science.toml              # project config: tools, run limits
├── GOAL.md                   # the question, the metric, the threshold
├── tools/                    # plain Python functions, auto-registered
├── research/
│   └── 001-my-hypothesis/
│       ├── HYPOTHESIS.md     # the testable claim and its rationale
│       ├── workflow.json     # the typed DAG
│       ├── agents/           # one Python file per node
│       └── runs/             # every execution, with its full trace
└── LEARNINGS.md              # append-only empirical memory

Nothing is hidden in a database. Every claim the framework makes about a run is backed by a file you can open, diff and review.

The loop

graph LR
    Goal[GOAL.md<br/><i>the question</i>] --> Hyp[HYPOTHESIS.md<br/><i>a testable claim</i>]
    Hyp --> Flow[workflow.json<br/><i>the typed DAG</i>]
    Flow --> Val[validate<br/><i>static checks</i>]
    Val --> Run[run<br/><i>trace + provenance</i>]
    Run --> Score[score<br/><i>deterministic gates</i>]
    Score --> Audit[audit<br/><i>three pillars</i>]
    Audit --> Learn[learn<br/><i>recorded insight</i>]
    Learn --> Hyp

Each arrow is one CLI command. The CLI reference lists all of them.