Email Outreach Agent — AI skill for Claude Code

Email Outreach Agent

**Give your AI coding assistant a job: draft personalized cold emails — and don't stop until you've approved enough of them.**

Email Outreach Agent is a tiny, open, do-it-yourself email drafting worker with a hard safety design: **it never sends anything.** You install it once into [Claude Code](https://claude.com/claude-code), [OpenCode](https://opencode.ai), or [Codex CLI](https://github.com/openai/codex). From then on, you just say:

*"Draft 100 personalized emails based on my offer file — I'll review and approve each one."*

…and your AI goes to work: researching each person from public information, writing short plain-text emails grounded in something *real* about them, passing every draft through an honesty gate (spam-pattern check, length limits, cited evidence) — batch after batch. **You approve each one.** Export the approved set and send it from your own account, in your own voice.

No accounts. No servers. No SMTP credentials. No "AI autopilot" spamming strangers on your behalf — that failure mode is engineered out, not promised away.

---

What you get

✍️ Real drafts, not templates	Every email is written fresh around one real observation about the recipient. Fabricated flattery is rejected by contract.
🛑 Human approval gate	Drafts only count after YOU run approve. The engine cannot approve and cannot send — there is nothing to misuse.
🚫 Spam patterns blocked at the door	"Act now", fake urgency, guarantees, "Dear Sir/Madam" — the validator rejects them before a human ever sees them.
🎯 A target that survives restarts	Stored on disk (data/state.json), not chat memory. Close the laptop; say "continue".
📝 One editable offer file	What you sell, who deserves an email, what proof you may cite — one human-readable offer.md.
⏸️ A scheduler that waits for you	Hands-free mode pauses automatically when ~10 drafts await your review. The machine never outr