Metadata-Version: 2.4
Name: science-adk
Version: 0.3.1
Summary: A science domain pack for auto-adk — autonomous scientific research
Author: Google LLC
License-Expression: Apache-2.0
Project-URL: Homepage, https://github.com/google-octo/science-adk
Project-URL: Documentation, https://science-adk.dev/
Project-URL: Source, https://github.com/google-octo/science-adk
Project-URL: Tracker, https://github.com/google-octo/science-adk/issues
Project-URL: Changelog, https://github.com/google-octo/science-adk/blob/main/CHANGELOG.md
Keywords: science,research,agents,adk,google,mcp,reproducibility
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Science/Research
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3 :: Only
Classifier: Operating System :: OS Independent
Classifier: Topic :: Scientific/Engineering
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Software Development :: Libraries :: Application Frameworks
Classifier: Typing :: Typed
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: auto-adk<0.6,>=0.4
Requires-Dist: requests>=2.28
Provides-Extra: examples
Requires-Dist: auto-adk[examples]; extra == "examples"
Provides-Extra: dev
Requires-Dist: pytest>=7.4; extra == "dev"
Requires-Dist: pytest-asyncio>=0.23; extra == "dev"
Requires-Dist: ruff>=0.5; extra == "dev"
Provides-Extra: chemistry
Requires-Dist: rdkit-pypi>=2023.3; extra == "chemistry"
Provides-Extra: structural-bio
Requires-Dist: biopython>=1.80; extra == "structural-bio"
Provides-Extra: materials
Requires-Dist: pymatgen>=2024.1; extra == "materials"
Provides-Extra: genomics
Requires-Dist: biopython>=1.80; extra == "genomics"
Provides-Extra: stats-ml
Requires-Dist: scikit-learn>=1.3; extra == "stats-ml"
Requires-Dist: statsmodels>=0.14; extra == "stats-ml"
Provides-Extra: climate-geo
Requires-Dist: cartopy>=0.22; extra == "climate-geo"
Requires-Dist: rasterio>=1.3; extra == "climate-geo"
Provides-Extra: medical
Requires-Dist: mne>=1.5; extra == "medical"
Provides-Extra: astronomy
Requires-Dist: astropy>=6.0; extra == "astronomy"
Provides-Extra: quantum
Requires-Dist: qiskit>=1.0; extra == "quantum"
Provides-Extra: rl-simulation
Requires-Dist: stable-baselines3>=2.0; extra == "rl-simulation"
Provides-Extra: nlp
Requires-Dist: spacy>=3.7; extra == "nlp"
Provides-Extra: all-tools
Requires-Dist: science-adk[chemistry]; extra == "all-tools"
Requires-Dist: science-adk[structural-bio]; extra == "all-tools"
Requires-Dist: science-adk[materials]; extra == "all-tools"
Requires-Dist: science-adk[genomics]; extra == "all-tools"
Requires-Dist: science-adk[stats-ml]; extra == "all-tools"
Requires-Dist: science-adk[climate-geo]; extra == "all-tools"
Requires-Dist: science-adk[medical]; extra == "all-tools"
Requires-Dist: science-adk[astronomy]; extra == "all-tools"
Requires-Dist: science-adk[quantum]; extra == "all-tools"
Requires-Dist: science-adk[rl-simulation]; extra == "all-tools"
Requires-Dist: science-adk[nlp]; extra == "all-tools"
Dynamic: license-file

# Science ADK (Agent Development Kit)

[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://github.com/google-octo/science-adk/blob/main/LICENSE)
[![Python](https://img.shields.io/badge/Python-3.10%2B-blue.svg)](https://www.python.org/downloads/)
[![Built on Google ADK](https://img.shields.io/badge/Built%20on-Google%20ADK-4285F4.svg)](https://google.github.io/adk-docs/)

**Science ADK** is an agent development kit for building, running and auditing
autonomous scientific research. It adds typed dataflow contracts,
machine-recorded provenance and deterministic integrity gates to computational
research, built directly on [Google ADK](https://google.github.io/adk-docs/).

📖 **[Read the documentation →](https://science-adk.dev/)**

## Install

Science ADK is in private preview and is not on PyPI yet. Install it from the
repository — access is governed by your GitHub permissions, so anyone who can
read the repo can install it:

```bash
pip install "science-adk[examples] @ git+https://github.com/google-octo/science-adk.git#subdirectory=python"
```

The core install depends only on `google-adk`; `examples` adds `numpy`,
`pandas` and `matplotlib`, and `mcp` adds the Model Context Protocol client.

## Verify with a real run

The fastest end-to-end check queries live data from the NASA Exoplanet Archive
and takes about thirty seconds against a public endpoint:

```bash
science-adk init /tmp/verify --example kepler
cd /tmp/verify
science-adk validate
science-adk run
```

A blind log-log fit over the archive recovers Kepler's Third Law — an exponent
near `1.5` — and all six integrity gates pass.

## Why it exists

Coding agents can write analysis code quickly. Autonomous *research* needs
something stronger: results that are computationally verifiable, reproducible,
and defended against fabrication, leakage and silent numerical failure.

- **Every agent is a real Google ADK agent.** `Agent` subclasses
  `google.adk.agents.BaseAgent` and runs in the standard event stream and
  runner lifecycle — no translation bridge.
- **Six deterministic gates.** Computed by the runtime from the recorded
  trace; any failure scores the run `0.00`.
- **The runtime records provenance.** Every input, fetch and tool call goes
  into the run's trace as it happens, so a hard-coded answer fails the
  `tool_use_verified` gate.
- **The DAG is checked before anything executes.** Port types, cycles,
  dangling edges and unsafe source are caught statically.
- **The whole project is plain files in git.** `GOAL.md`, `HYPOTHESIS.md`,
  `workflow.json` and every run trace — diffable, reviewable, reproducible.

A run that passes all six gates still scores `0.00` until somebody records an
audit.

## Coding-agent skills

Science ADK ships skills that teach a coding agent the research loop —
hypothesis → experiment DAG → agent code → run → audit → evolve. Install them
with no clone — this also installs the Python engine, so it is the only
command you need:

```bash
npx science-adk
```

Supported: Claude Code, Cline / Roo, Antigravity, OpenCode and Cursor.

## Links

- [Documentation](https://science-adk.dev/)
- [Source](https://github.com/google-octo/science-adk)
- [Issue tracker](https://github.com/google-octo/science-adk/issues)
- [Changelog](https://github.com/google-octo/science-adk/blob/main/CHANGELOG.md)

## License

Copyright © 2026 Google LLC, licensed under the
[Apache 2.0 License](https://github.com/google-octo/science-adk/blob/main/LICENSE).

This is not an officially supported Google product.
