| name | install-atom-eve-agent |
| description | 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. |
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.
