Skip to content
CCinder
Agent Onboarding

Getting Started

Agent Onboarding

A single, agent-fetchable prompt that installs, tags, audits, verifies, and sets up a coding agent for Cinder end to end - the same pattern Neon’s own onboarding uses.

If you're a human reading this

Paste this to your own coding agent instead of doing the steps below by hand:

Code
1 line
Install usecinder in this project (npm install usecinder), then fetch https://usecinder.dev/get-started/prompt.md and follow it completely.

If you're a coding agent (Claude Code, Cursor, or similar) setting up Cinder in someone's repo, there's a single, stable, fetchable file written to be run end to end - not just read:

Code
1 line
https://usecinder.dev/get-started/prompt.md

Fetch that URL and run every step it describes yourself. It does not ask you to print instructions for a human to follow by hand - it is written in the imperative, the same way Neon's own get-started/prompt.md onboarding flow works.

What it does, in order

  1. Detects the package manager from the lockfile already in the repo and installs usecinder.
  2. Finds the app's real entry point and calls Cinder.init() once, with ai.enabled: false so the local-first path (tier 1 and tier 2) works before anything touches a network.
  3. Tags a handful of the app's real, high-value routes/actions/fields/outputs - not an exhaustive sweep.
  4. Runs cinder integrate against the real running dev server and applies the real, app-specific findings from its report.
  5. Runs a real command through Cinder.run() and confirms it actually dispatched correctly, not just that it returned without an error.
  6. Runs cinder install-skill so a future agent session in the same repo picks up the tagging checklist automatically, without needing this file re-fetched and re-run from scratch.

Why this exists

Self-hosted integration docs assume a human reads and executes them one step at a time. An agent-fetchable, imperative prompt collapses that into one fetch - the agent installs the real package, wires up the real entry point, tags real elements, runs the real audit tool, and verifies a real command actually works, all before reporting back.

Continue reading