X/Twitter Engagement Workflow

Most "Twitter automation" tools fall into auto-DM-spam-or-quit. The useful version is different: the agent reads, summarises, drafts, and monitors — but you remain the publisher. Saves an hour a day, doesn't sound like a bot.

What it does

Daily timeline summary: what your follows posted today, ranked by signal
Draft replies for important threads (including yours) — you approve or edit before posting
Monitor for keywords (your product name, your competitors, mentions of your work) — alert + draft when opportunities appear
Pull-quote extraction: when you save a long thread, agent extracts the meat as a single retweetable card

What you'll need

An x-tools plugin with X API v2 credentials
A defined "your voice" sample (10–20 tweets you wrote, stored in memory)
Your follow-list and keyword-watch list
Cron

Setup

1. Build the x-tools plugin

Send to Flowly

Create a plugin called "x-tools" using X API v2. Tools: - x_timeline(window: int = 6) -> recent tweets from my follows - x_search(query: str, limit: int = 50) -> matching tweets - x_user_recent(handle: str, limit: int = 20) -> a user's recent tweets - x_post(text: str) -> post a tweet (RETURNS preview only — do NOT actually post; surface the text + ask user to confirm) - x_reply_draft(tweet_id: str, draft: str) -> reply preview, not posted Read X_BEARER_TOKEN from env. The post and reply tools should NEVER actually post — they should return a JSON object describing what would be posted, and let me copy-paste manually.

2. Capture your voice

Send to Flowly

Remember my X voice: Sample tweets (paste 10-20 of yours that you'd be proud of): "Most plugin systems fail because they let plugins run in-process. By the time you find the bug it's already crashed prod." "... Tone notes: - Direct, opinionated - Specific examples > generic claims - Self-deprecating about my own past mistakes - Avoid emojis, hashtags, exclamation marks - 1-2 sentences per tweet, threading rare

3. Daily timeline summary

Send to Flowly

Cron "x-summary" daily at 10 AM: 1. x_timeline(window=12) — get last 12h of follows' tweets 2. Score each by: - Engagement (likes, replies — relative to author's baseline) - Topical fit (against my interests in memory) - Conversation potential (asks a question, takes a position) 3. Group top 10 into: - Worth a reply (drafts included via x_reply_draft) - Worth reading - Worth a retweet/quote 4. Send as Telegram message. Each item: tweet snippet, author, score, action.

4. Keyword watch

Send to Flowly

Cron "x-watch" every 2 hours: For each keyword in my watch list: 1. x_search(<keyword>, limit=30) 2. Filter to tweets within last 2 hours, no spam (low-engagement burst account) 3. Highlight ones where I might add value (my expertise overlaps) 4. Draft a reply via x_reply_draft Send anything found to Telegram. Each: tweet, author, draft reply.

Tips

Never auto-post. Auto-posted replies sound like a bot, even when carefully prompted. Draft + human approval keeps voice authentic.
Voice samples matter. Without good samples, drafts are generic AI slop. With them, drafts pass for yours after light editing.
Don't watch your own mentions obsessively. Once-an-hour is fine. More than that is anxiety, not strategy.
Skip negative threads. When the timeline is full of dunking, the agent should flag and skip — engaging in flame wars hurts you, not the bot.
API limits are real. X API rate limits hit fast. Tune cron frequencies down if you hit them. Better to miss a tweet than to get rate-limited at the wrong moment.
Posting from the agent crosses an ethical line for some. Disclosure is your call but generally — if a reply is 100% the agent, mark it. If it's heavy edit, no need.