# xBot

AI-powered Twitter/X bot that generates tweets via the Anthropic Claude API and posts them using stealth browser automation through Kernel.sh (residential proxies + CAPTCHA solving).

## Key Features
- AI tweet generation (claude-sonnet-4-6, max 280 tokens)
- Multi-account support (credentials keyed by account name in .env)
- Configurable scheduling (daily or random interval posting)
- Tweet history tracking to avoid repetition
- Link injection (URLs added post-generation, never sent to AI)
- Anti-detection: keyboard.type() with delays, adaptive login flow

## Tech Stack
Node.js 18+ (ES modules), @onkernel/sdk, dotenv

## Project Structure
- `bot.js` — Core logic: login to X, generate tweet via Claude, post it
- `scheduler.js` — Picks random times within a window, invokes bot.js via child_process
- `package.json` — Dependencies & scripts
- `.env` — Credentials (not tracked)
- `prompts.json` — Prompt configs with account mapping, schedule, and link placeholders (not tracked)
- `history-<prompt-name>.json` — Per-prompt tweet history (auto-generated)

## Entry Points
- `node bot.js <prompt-name>` — Post a single tweet immediately
- `node scheduler.js <prompt-name>` — Schedule random posts for the day

## External Services
| Service | Purpose | Config Key |
|---------|---------|------------|
| **Kernel.sh** | Stealth browser automation (residential proxies, CAPTCHA solving, Playwright) | `KERNEL` env var |
| **Anthropic Claude API** | Tweet generation | `ANTHROPIC` env var |
| **X/Twitter** | Target platform for posting | Per-account: `{PREFIX}_USER`, `{PREFIX}_PW`, `{PREFIX}_EMAIL`, `{PREFIX}_PHONE` |

Multi-account credentials use uppercase prefix pattern in .env (e.g., MYACCOUNT_USER, MYACCOUNT_PW).

## Deployment
Deployed on AWS EC2 in **us-west-2**:
- Instance: `i-088b711cac69e6ac9`
- Name: `xbot`
- Type: `t4g.nano` (ARM-based)
- Public IP: `44.249.48.3`
- AWS account: `457667483187`
- Cron: `0 16 * * *` (4pm UTC / before 8am PST window), one job per prompt

## Guidelines
- All changes must preserve stealth/anti-detection behavior and multi-account architecture
- bot.js is the core; scheduler.js orchestrates timing
- Config lives in .env + prompts.json
- All project memory/context stays in this file (or a `CLAUDE-memory/` folder in the repo) — never save to the external `~/.claude/projects/.../memory/` system