# Email Bump Agent Skills

> Open-format agent skills that teach AI coding agents to send email, manage audiences, run campaigns, provision infrastructure, and stay deliverable — human in the loop by design.

**Canonical page:** [https://emailbump.com/agent-skills](https://emailbump.com/agent-skills)

Agent skills are folders with a `SKILL.md` file — an open format read by Claude Code, Cursor, Codex, Devin, Gemini CLI, GitHub Copilot, OpenCode, and other coding agents. Skills load on demand, so they add no prompt overhead until they are needed.

## Install

```bash
npx skills add danest/emailbump-skills
```

Or copy any skill directory into your agent's skills folder (for Claude Code: `~/.claude/skills/`). Every skill authenticates with an Email Bump API key via the `EMAILBUMP_API_KEY` environment variable. The skills are MIT-licensed at [github.com/danest/emailbump-skills](https://github.com/danest/emailbump-skills).

## The skills

- **emailbump** — transactional sending through the REST API: immediate and scheduled sends, stored MJML templates, and Liquid personalization.
- **emailbump-audience** — contacts, lists, live segments, consent (subscribe/unsubscribe), and behavioral events that drive automations.
- **emailbump-campaigns** — draft, A/B test (2–4 variants), schedule, and send marketing campaigns with gradual and timezone-aware strategies.
- **emailbump-management** — provision workspaces, projects, sending domains with DNS verification, starter flows, and project API keys.
- **email-best-practices** — vendor-neutral deliverability and compliance rules: SPF/DKIM/DMARC, bounce handling, list hygiene, warm-up.

## Guardrails

Email cannot be unsent, so every skill bakes the same rules into the instructions the agent follows:

1. Rehearse with a single test send to the operator's own address before any audience send.
2. Report the audience, its size, subject, from address, and schedule — then wait for explicit confirmation before calling send.
3. Never re-subscribe an unsubscribed contact or import a list without confirming it is permission-based.
4. Treat minted API keys as one-time secrets; never echo them into code, logs, or chat history.

## The rest of the agent-readable platform

- [agents.md](https://emailbump.com/agents.md) — Markdown twins of every public page.
- [llms.txt](https://emailbump.com/llms.txt) — curated index of product, guides, and API reference.
- [API overview](https://emailbump.com/docs/api-reference.md) — base URL, authentication, errors, and the map of every resource page.
- [Flows API](https://emailbump.com/docs/flows-api.md) — automated journeys as triggers and steps.
- [Workspaces API](https://emailbump.com/docs/workspaces-api.md) — provision the account structure programmatically.
- [openapi.json](https://emailbump.com/openapi.json) — OpenAPI 3.1 for every `/v1` endpoint, with the key scope each one needs.
- [/.well-known/oauth-protected-resource](https://emailbump.com/.well-known/oauth-protected-resource) — RFC 9728 auth metadata: bearer keys, and the `project` and `account` scopes.