# Install Atom Eve Agent

> Use when setting up or installing an existing Atom Eve AI agent into a project, including scaffolding a new eve.dev (Eve) or Flue app, adding an agent with the atom-eve CLI, wiring credentials, and verifying it runs. Triggers include "set up an Atom Eve agent", "install the SEO/ads/QA agent", "add an agent to this project". To author a NEW agent for the registry, use author-atom-eve-agent instead.

- Skill: `elie222/install-atom-eve-agent` (Agent Skill)
- Install (CLI): `npx skillmds@latest add elie222/install-atom-eve-agent`
- Raw SKILL.md: https://api.skillmd.com/api/skills/elie222/install-atom-eve-agent/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Marketing & Growth
- Author: elie222 (https://skillmd.com/u/elie222)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/elie222/install-atom-eve-agent

---


# Install an Atom Eve agent

Atom Eve installs real agent source into the user's own repo, like shadcn but for
agents. It does not host agents or store credentials. Pick the agent first with
find-atom-eve-agent. For the always-current version of these steps, fetch
https://www.atomeve.dev/start.md.

## Install

Eve is the default target. Use Flue only if the user asks or the project already uses
`@flue/runtime` (an Eve project has an `agent/` directory).

- New project: `npx atom-eve create <name> --agent <agent>`, then `cd <name>`
- Existing Eve project: `npx atom-eve add <agent>`
- Flue: `npx atom-eve add <agent> --target flue`
- Slack: a two-way Slack channel is on by default (pass `--no-slack` to opt out); add
  `--deliver slack` to also post a scheduled run's report to Slack

## Wire credentials and config

Read the agent's `env` (secrets: name, description, url, optional) and `config`
(non-secret settings for the `<!-- project-config -->` block in `agent/instructions.md`)
from https://www.atomeve.dev/index.json. Those are the summaries; the walkthrough is the
agent's SETUP.md, installed into the project root (and shown on the agent page): how to
obtain each credential, a read-only verification call for it, and where to store the
value. Then follow the setup playbook in
https://www.atomeve.dev/start.md: inspect installed CLIs and existing env first, ask all
questions in one message up front, confirm before acting against a user account, and only
run read-only validation calls with consent. Never invent secret values. On Eve,
connections with auth `vercel-connect` are wired via Vercel Connect; `env`-auth
connections read project env vars. On Flue, use the project's runtime secrets.

## Runtime config (Eve)

Model calls route through the Vercel AI Gateway, so there is no model API key to set, but
the Vercel account or team must have Gateway access. Run `vercel link`, then
`vercel env pull`. Set `AGENT_MODEL` only to choose a different available model.

## Verify

Run `pnpm install`, `pnpm typecheck`, `pnpm build`. For a real smoke test that spends
tokens, run `npx eve dev`.

Atom Eve agents open PRs or post drafts for review. They do not auto-publish or take
destructive actions. If required setup is missing, stop and say exactly what to configure.