@appliqation/create-agents

An interactive installer for the Appliqation standalone agent family — Autotest, Autopilot, Scriptgen, Defect Fix, Heal Selector, Visual Regression, Explorer, PR Raise, and Dashboard. It answers a few questions and scaffolds a local folder with exactly the agents you picked, a working .env, and ready-to-run terminal/CI commands.

Nothing here replaces the manual install path (npm init -y && npm install @appliqation/autotest) — it's the same steps, just asked as questions instead of typed by hand, plus a generated README.md/CI workflow tailored to what you chose.

Use it

npx @appliqation/create-agents

or, using npm's own initializer convention:

npm create @appliqation/agents

What it asks

1. What to call the suite — becomes the folder name and the generated package.json's name.
2. Which LLM — Anthropic, OpenAI, DeepSeek, or GLM. Prompts for the right API key, and for DeepSeek/GLM, the model ID too (this family has no default model for either, on purpose — see each agent's own config/env.ts).
3. Which agents — a checklist, each with a one-paragraph description of what it does and what installing it actually commits you to (write access to your repo, to Appliqation, or to GitHub). Pick any combination: testing only, testing plus defect-fixing, everything except PR Raise, all nine. Picking PR Raise asks for a GITHUB_TOKEN; picking Dashboard asks where its audit trail should live (a local JSONL file or MongoDB).
4. Confirm, then it writes the folder, runs npm install for exactly the packages you picked, and installs Chromium only if a browser-driving agent is among them.

What you get

Inside the new folder:

• package.json — the agents you picked, nothing else.
• .env — filled in with what you entered, ready to use.
• .env.example — the same variable names with secrets blanked out; the one to commit and to mirror into your CI secrets.
• README.md — one section per agent with its real CLI invocation forms, plus a copy-paste GitHub Actions workflow already wired to the agents and env vars you chose.

Why a separate installer, not a flag on Autopilot

The nine agents are meant to be picked up independently — appliqation-autotest alone is a complete, safe starting point, per the learning-order docs. This tool's only job is the one-time folder setup every one of them needs (a .env, the right packages, the right CI wiring); it has no opinion on your test cases and never runs an agent itself.

Development

git clone https://github.com/appliqation/agents-installer.git
cd agents-installer
npm install
npm run dev            # runs the wizard via tsx, no build needed
npm run typecheck
npm test

src/scaffold.ts holds the pure, fully-tested builders (.env, README.md, the CI YAML); src/cli.ts is the thin interactive wrapper around them (prompts-based, not unit-tested directly, mirroring how every sibling agent keeps its own Commander wiring out of test scope).

License

MIT — see LICENSE.