brodyautomates

Inbox Agent — AI skill for Claude Code

AI community

An AI that reads your inbox, drafts every reply against a playbook you own, and puts each one on your phone with a send button.

How to install Inbox Agent

This entry records only its repository, not the path inside it, so there is no exact command to give. Open brodyautomates/inbox-agent and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Inbox Agent does

An AI that reads your inbox, drafts every reply against a playbook you own, and puts each one on your phone with a send button. Manual gate first, autopilot when it has earned it. Runs 24/7 on a $5 server.

Alternatives in AI

Daymon — Daymon puts your favorite AI to work 24/7. It schedules, remembers, and orchestrates your own virtual team 372 ★
Unofficial Claude Code Prompt Playbook — Unofficial playbook for production-grade LLM system prompt architecture, derived from local analysis of Claude 258 ★
Gh PR Review — GitHub CLI extension that adds full inline PR review comment support — view, navigate, reply to, and resolve r 174 ★

README

inbox-agent

An AI that reads one inbox, writes every reply against a playbook you own, and puts each one on your phone with a send button. Runs 24/7 on a five dollar server, a Mac you leave on, or a Windows PC. About an hour to set up. Nothing to buy except the Claude subscription you probably already have.

Start with a manual send gate: nothing goes without your tap. Move to assisted, then autopilot, once it has earned it. The gate is in code, not in a prompt.

Inbox  →  Playbook  →  Draft in Gmail  →  Your phone (Telegram)  →  [gate]  →  Sent

What it does

Polls the inbox every two minutes. New email, it decides whether it is worth a reply.
Writes the reply against playbook/SKILL.md: your prices, your rules, your stages, your voice. Creates it as a Gmail draft, threaded, to the sender only.
Posts the draft to Telegram with SEND, SKIP, CHAIN. Reply to it in plain text ("make it 9k", "warmer opener") and it rewrites.
One live draft per thread. A follow-up from them retires the old draft and writes a fresh one.
Three modes: manual, assisted, autopilot. Hard stops that hold in every mode. A stop file that beats everything.
Works with Claude Code headless (your Claude subscription) or the Anthropic API. One config flag.

What it never does

Sends to anyone who did not email you first. There is no cold email path in this code.
Lets the model choose a recipient. It is copied from the original message.
Sends from the wrong address. The Gmail token is checked against your config on every command.
Sends twice. A draft is marked done before the send fires.
Sends anything while ~/.config/inbox-agent/PAUSED exists.

Sixty minute path

Telegram bot. Message @BotFather, /newbot, keep the token. Message your new bot once. Get your own id:
```
curl -s "https://api.telegram.org/bot/getUpdates" | grep -o '"id":[0-9]*' | head -1
```
Google. Four minutes of clicks, and press Publish. [docs/GOOG