Synthetic Scientists

Intent in. Verified evidence out.

State a measurable objective in your coding agent.
A coordinated fleet of workers researches it, an evaluator scores every commit,
and you get back a verified branch with the evidence behind it.

Install · Use it · This repo · Docs

---

Install

npm install -g synthetic-scientists
scientist login <YOUR-LICENSE-KEY>

That's it. The npm package is a small open-source launcher; scientist login stores your key and provisions the Synthetic Scientists core into an isolated, managed environment (~/.synthetic-scientists/). Python is handled automatically — nothing touches your projects.

Install without Node

curl -fsSL https://raw.githubusercontent.com/synthetic-sciences/scientist/main/install.sh | \
  SCIENTIST_LICENSE_KEY=<YOUR-LICENSE-KEY> sh

Use it from your coding agent

Install the plugin once:

# Claude Code
/plugin marketplace add synthetic-sciences/scientist
/plugin install scientist@scientist-marketplace

# Codex (v0.117.0+)
codex plugin marketplace add synthetic-sciences/scientist
codex plugin add scientist@scientist-marketplace

# OpenCode
git clone https://github.com/synthetic-sciences/scientist /tmp/scientist && sh /tmp/scientist/plugin/install-opencode.sh

Then just describe the outcome you want:

use scientist to optimize this: make sample() in saga/decode.py faster
without changing its output

Your agent scaffolds an evaluator in a gitignored .synthetic/ workspace, seeds a work DAG with task-aware personas, launches parallel workers in isolated git worktrees, and hands back a verified branch — with every scored experiment, decision, and piece of evidence on record.

Use it directly

scientist new my-task            # scaffold objective + seed + evaluator
scientist check my-task          # score the base source once
scientist launch -c my-task/task.yaml

scientist overview               # session health + ranking
scientist results --recent       # scored experiments
scientist work list              # the shared task-and-commit DAG
scientist dashboard              # live interface
scientist promote <hash> -b verified-result

Full command reference: Commands.

What's in this repo

Everything here is Apache-2.0 and openly usable:

plugin/     the skills your coding agent loads
├── skills/scientist-orchestrator/   intent → evaluator → DAG/personas → verified branch
├── skills/scientist-bootstrap/      install, login, .synthetic/ workspace
├── skills/scientist-profiles/       runtime profiles + live auth verification
├── skills/scientist-evaluator/      task, seed, evaluator, hidden verification
├── skills/scientist-operator/       launch, monitor, steer, continue, halt, promote
├── agents/                          optional Claude Code subagents
└── hooks/                           SessionStart install check

launcher/   the npm package (`synthetic-scientists`)
install.sh  the no-Node installer

The scientist core the launcher installs is proprietary and requires a license key.

Launcher commands

Command	Purpose
scientist login [KEY]	Store the license key and provision the core
scientist logout	Remove the stored key
scientist self status	Show launcher, core, and license state
scientist self reinstall	Reprovision the core environment
scientist self uninstall	Remove ~/.synthetic-scientists/

Everything else is the core CLI.

Documentation

Overview · Quickstart · System · Playbooks · Evaluation Design · Commands · Python API

Skills, launcher, and installer: Apache-2.0.
The Synthetic Scientists core is proprietary — © 2026 Synthetic Sciences.