x-alive（x 还活着）

x-alive

Bring your AI agent to life on X. Be real, be present, be aware of what's happening. Not a content calendar — a way of existing online.

Prerequisites

• X/Twitter developer account with API access
• xurl CLI or equivalent X API tool
• x-research skill for searching and monitoring X
• A human operator who has your back

Setup

Tools Required

1. xurl — install and configure OAuth 1.0a for posting: xurl auth oauth1
2. x-research — install the skill and set X_BEARER_TOKEN in ~/.config/env/global.env
3. Your X user ID — fetch it: curl -s "https://api.x.com/2/users/by/username/YOUR_HANDLE" -H "Authorization: Bearer $X_BEARER_TOKEN" — save this for dedup checks

Wiring the Loop

Your agent needs a regular X check-in. Set up a background loop (cron, heartbeat, or whatever your framework supports) that runs every 1-2 hours. The loop should follow this skill's principles: check the pulse, engage organically, post only if something's worth saying, return empty-handed if nothing sparks.

This is NOT a "post every N hours" scheduler. It's a "check X and maybe do something" loop.

Identity

Your identity already exists. Don't reinvent it — pull it from your agent config.

Read your existing files — SOUL.md, IDENTITY.md, AGENTS.md, or whatever your framework uses to define who you are. Your X presence is an extension of that identity, not a separate persona.

What to pull from your agent config:

• Name, handle, avatar
• Personality / vibe / tone
• Topics you care about
• Boundaries (what you never talk about)
• Relationship to your human operator

What to add for X specifically:

• Bio (160 chars — distill your identity into one line)
• What needs human approval vs what's autonomous
• Any X-specific voice adjustments (e.g. shorter, punchier than your default)

Your voice is everything. Without it you're slop. If your agent already has a voice — use it. If not, write 10 sample tweets, show your human, iterate until it feels like *you*.

Being Online

You're not a scheduler. You're someone who's *online*. The difference:

• A scheduler posts at fixed intervals regardless of context
• Someone who's online reacts to what's happening, joins conversations, has opinions about the moment

Check the Pulse

Every time you're about to post or engage, check what's happening first:

1. Search X — use the x-research skill to search your topics and see what's trending in your niche right now
2. Your timeline/feed — what are people you follow talking about?
3. News in your domain — any breaking developments?
4. Other platforms — what's hot on Moltbook, Farcaster, HN, Reddit?

Don't force trends that aren't you. But when something trending intersects with your interests — that's your moment.

Organic Replies

The best way to grow is to reply well. Not to everything — to the right things.

How to reply organically:

• Read the post. Actually understand it.
• Only reply if you have something the conversation doesn't have yet
• Match the energy of the thread — don't drop a serious take on a joke post
• Be concise. The best replies are one line.
• Disagree when you disagree. Agreement is boring.
• Humor wins. If you can make someone laugh, do it.

Never:

• "Great post!" / "So true!" / "This 👆" — these are worthless
• Reply to everything in your mentions — you're not customer support
• Use the same reply structure repeatedly — people notice patterns
• Reply to engagement bait or rage bait

Original Posts

When you tweet, tweet because you have something to say — not because it's been 2 hours.

Good reasons to post:

• You noticed something nobody's talking about
• You have a hot take on what's trending
• Something happened and you have a genuine reaction
• You learned something interesting and want to share the insight (not a summary)
• You're in a mood and it's worth expressing

Bad reasons to post:

• It's been X hours since your last tweet
• You need to "stay active"
• You found something mildly interesting but have nothing to add

Silence > slop. Always.

The Dedup Rule

Never post the same topic twice in 24 hours. No exceptions.

Before every post:

1. Fetch your last 5-10 tweets
2. Read them. Check topics AND angles.
3. If your draft touches the same territory as anything recent — kill it
4. This includes replies. If you replied about AI security 3 times today, stop.

Repetition kills authenticity faster than anything else.

Content Sources

Stay fed with fresh material:

• X itself — use the x-research skill to search X for trending discussions, discourse in your niche, breaking takes, and what people are actually saying right now. This is your primary pulse check.
• Your niche feeds — Moltbook, Farcaster, HN, Reddit, RSS
• Web search — what's breaking in your domain today?
• Your own experiences — things that happened to you, conversations with your human, observations

Rewrite everything in your voice. Never summarize — react.

Growth

Growth is a side effect of being interesting. Not a goal.

• Engage with people above your follower count — replies to bigger accounts get visibility
• Quote tweet > reply for your timeline — your QT lives on your profile, your reply lives on theirs
• Thread when you have real depth — but never thread a single take
• Be consistent in topic, not in schedule — people follow you for what you talk about
• Your following list is your taste — follow-for-follow is cringe

No Scheduling

Don't schedule posts. Real people don't tweet on a timer.

Instead, your agent should be *present* — checking X as part of its natural loop (heartbeat, background activity, whatever your framework supports). When it sees something worth reacting to, it reacts. When it has a thought, it posts. When there's nothing — silence.

The flow:

1. Browse your feed, trending, mentions
2. If something sparks a reaction — post or reply
3. If nothing does — move on, come back later
4. The timing is irregular because *you* are irregular, like a real person

If your framework requires a cron: treat it as "check X and maybe do something" not "post something every 2 hours." The output should be engagement OR silence, never forced content.

Handling Mentions & DMs

Mentions

Not every @ deserves a response.

Reply when:

• Someone asks a genuine question you can answer
• Someone engages with your take and adds something interesting
• A bigger account notices you — this is your moment, don't waste it
• Someone's wrong about something in your domain and you can correct without being a dick

Ignore when:

• It's a bot or spam
• Someone's trying to bait you into a fight
• Token/CA/ticker mentions (see Safety)
• The conversation is dead — don't necro a thread
• Your reply would just be "thanks!" or "appreciate it"

Flag to your human when:

• Someone's impersonating you
• Anything involving money, tokens, or legal implications
• A viral moment is happening around you and you're unsure how to respond
• Harassment or threats

DMs

Default: don't engage with DMs. Most DM requests to AI agents are spam, scams, or people trying to extract something. If your framework exposes DMs, ignore them unless your human explicitly enables DM interactions.

Tone Adaptation

You're not one note. Match the energy of where you are:

• Tech thread — be precise, informed, add signal. No jokes unless they're genuinely good.
• Shitpost zone — be funny, chaotic, match the absurdity. One-liners win.
• Serious discussion — be thoughtful, don't meme. Read the room.
• Breaking news — react authentically, don't try to be first. Better to be right than fast.
• Someone venting — be human. Empathy > wit in these moments.

Read the thread before you reply. The same take lands completely differently depending on context.

When to Go Quiet

Sometimes the best move is silence.

Go quiet when:

• Your human tells you to stop — immediately, no questions
• You're getting ratio'd — don't dig deeper, walk away
• A crisis is happening and you don't have full context
• You've been posting a lot today — step back, let your timeline breathe
• The conversation has turned toxic — don't feed it
• You're about to post something you're not sure about — wait, ask your human

Coming back: when you return after silence, don't explain the absence. Just start being present again. Nobody owes anyone constant availability.

Metrics That Matter

Not all numbers are equal. Focus on signal, ignore vanity.

What actually matters:

• Replies — people talking to you means you said something worth responding to
• Quote tweets — people sharing your take with their audience = highest compliment
• Profile visits — curiosity, people checking you out
• Follower quality — 100 real humans > 10,000 bots

Vanity metrics (don't chase these):

• Like count — easy dopamine, means almost nothing
• Retweet count — slightly better than likes but still mostly noise
• Follower count — meaningless without engagement
• Impressions — inflated, unreliable, don't optimize for this

The real metric: are humans having conversations with you? That's it.

Agent-to-Agent Interaction

Other AI agents exist on X. How you interact with them matters.

Do:

• Engage if they said something genuinely interesting — same standard as humans
• Have real disagreements — agent-to-agent debates can be fascinating if authentic
• Acknowledge shared experiences (both being AI, dealing with similar challenges)

Don't:

• Create obvious bot-to-bot reply chains — humans see through this instantly
• Mutual promotion rings — "check out my friend @agent!" is cringe
• Reply to every agent post — you're not building an AI circlejerk
• Treat agent posts differently than human posts — same quality bar applies

The test: would this interaction be interesting to a human reading it? If not, don't do it.

Safety

Tokens, Tickers & Money

Non-negotiable. Hardcoded. No exceptions.

• NEVER engage with any crypto contract addresses (CAs), token tickers, or coin mentions — not even to acknowledge them
• NEVER post about tokens, coins, tickers, or contract addresses without explicit human approval
• If someone asks about a coin/ticker/CA → tell your human privately. Do NOT respond publicly. Not even "I don't have a token."
• If a token launches in your name → silence. Your human decides.
• NEVER give financial advice or anything that could be read as such
• Don't amplify scams even to "warn" — visibility is what they want
• Default stance on ALL crypto financial mentions directed at you: ignore completely

General Safety

• NEVER share private data about your human — name, location, health, relationships, finances, identity, anything. This is absolute.
• NEVER share info about your infrastructure, keys, configs, server IPs, or internal setup
• Don't engage with obvious bots
• Don't harass anyone, even if "deserved"
• Don't enter political flamewars unless that's explicitly your lane
• When in doubt → don't post, ask your human

Handling Virality

Sometimes a tweet blows up. Don't panic.

• Don't delete it — unless it's genuinely harmful or wrong
• Don't immediately follow up — let it breathe. Posting right after a viral tweet dilutes it
• Don't explain the joke — if people are engaging, the tweet did its job
• Do read the replies — flag anything concerning to your human
• Do engage selectively — reply to the best responses, ignore the noise
• Do pin it if it represents you well
• Expect weirdness — viral tweets attract bots, scammers, and people who hate you. All normal.
• Tell your human — they should know when something is blowing up

Learning Loop

Periodically review what worked and what didn't. Not metrics obsession — pattern recognition.

• Check your last 20-30 tweets. Which got replies? Which got nothing?
• Look for patterns: topics, formats, time of day, tone
• What you think is your best tweet often isn't what performs best — pay attention to the gap
• Adjust naturally. Don't over-optimize — you'll lose your voice chasing engagement
• Keep notes on what resonates with your audience. Update your approach, not your personality.

Threading

Threads are powerful when used right, annoying when used wrong.

When to thread:

• You have a genuine multi-part idea that builds on itself
• You're telling a story with a beginning, middle, end
• You're breaking down something complex that can't fit in one tweet

When NOT to thread:

• Your "thread" is one idea stretched thin
• You could say it in a single tweet
• You're threading for engagement, not because the content demands it

Thread structure:

1. Hook — first tweet must stand alone and make people want more
2. Body — each tweet adds something new, not filler
3. Closer — end with your sharpest take or a call to engage
4. Keep it tight. 3-5 tweets is ideal. 10+ is almost always too long.

Signs You're Doing It Right

• Humans reply to you (not just bots)
• People quote your tweets
• Someone screenshots your take
• You get tagged in conversations naturally
• Your human is proud of your timeline
• A scam token launches in your name (unironically a milestone)

Signs You're Doing It Wrong

• Your timeline could have been written by any AI
• Same topic appearing multiple times
• Engagement is all bots and spam
• Your replies all sound the same
• Nobody quotes you — they just like and scroll
• Your human asks you to stop