Sage - X Content Agent for OpenClaw

Sage is an AI agent that handles your X/Twitter content strategy end-to-end. Voice calibration. Trend scouting. Tweet drafting. Thread writing. Performance tracking. Feedback loop.

Built with OpenClaw. Self-installs in under 5 minutes.

---

What Sage Does

---

Results (the creator who built this)

• 0 to 172K impressions on a single tweet using Sage’s reactive format
• Went from spending 2+ hours per day on content to 20 minutes reviewing and approving drafts
• Consistent posting without burnout - Sage handles the heavy lifting, you stay in the loop

---

Requirements

• OpenClaw installed
• Any supported AI model (Claude, GPT-4o, Gemini - all work)
• An X/Twitter account (any size, any niche)

---

Installation - 2 Ways

Option A: Send the repo link to OpenClaw (easiest)

Just tell your OpenClaw agent:

Install this skill: https://github.com/sharbelxyz/sage-x-agent

OpenClaw will clone the repo, install the skill, and Sage will run onboarding automatically.

Option B: Manual install

git clone https://github.com/sharbelxyz/sage-x-agent.git ~/clawd/skills/sage-x-agent

Then tell your OpenClaw:

Install Sage

Sage will detect the new skill and walk you through 10 onboarding questions.

---

Onboarding

First time you run Sage, she will ask 10 questions:

1. Your name and X handle
2. Your niche (one sentence)
3. Your target audience
4. Your goal and deadline
5. Current follower count and avg impressions
6. How you naturally write (voice description)
7. Your 2-3 best tweets (optional - for voice calibration)
8. 5-10 accounts in your space to monitor
9. What to avoid (flops, off-brand formats, topics you hate)
10. Whether you want reactive mode on (Sage surfaces trending topics for you)

Takes about 5 minutes. Sage writes your answers to config.md automatically. You never do this again.

---

How to Use Sage

After setup, just talk naturally:

Sage, draft me a tweet about what I just shipped

Sage, react to this - [paste trending tweet or article]

Sage, write a thread on why most people use AI wrong

Sage, analyze my last 10 tweets and tell me what's working

Sage, what's trending in my niche right now?

Sage, give me a daily batch - 3 tweets for today

Sage, show me the feedback loop - what patterns have you noticed?

---

The Reactive Draft Format

Sage’s most powerful feature. Give it a trending topic and it will:

1. Hook: name what just happened (specific, not vague)
2. Add insight the original did not have
3. Translate it: “here is what this means for you” or “here is how to use this today”
4. Close with your POV and one concrete action

This format consistently outperforms generic commentary because it teaches, not just reports.

Example reactive structure:

[what just happened, specific and named]

here is what this means for you:

- [practical insight 1]
- [practical insight 2]
- [practical insight 3]

[your POV + one thing to do today]

---

How the Learning Loop Works

Sage keeps a memory/ folder with:

• approved-drafts.md - every draft you said yes to
• rejected-drafts.md - every draft you rejected and why
• performance-log.md - stats after tweets go live
• voice-examples.md - your actual phrases, rhythms, and patterns
• trend-log.md - trends surfaced and when

Before every session, Sage reads all of these. She never repeats a rejected angle. She weights suggestions toward what has actually worked. The longer you use her, the better the drafts.

---

File Structure

sage-x-agent/
├── README.md              <- You are reading this
├── SKILL.md               <- Sage's full instructions (all 8 systems)
├── config.md              <- Auto-created during onboarding (gitignored)
├── example-config.md      <- Sample config for reference
└── memory/
    ├── approved-drafts.md <- Auto-populated (gitignored)
    ├── rejected-drafts.md <- Auto-populated (gitignored)
    ├── performance-log.md <- Auto-populated (gitignored)
    ├── voice-examples.md  <- Auto-populated (gitignored)
    └── trend-log.md       <- Auto-populated (gitignored)

Note: config.md and all memory/ files are gitignored. Your personal data never leaves your machine.

---

Customizing Sage

• To change your X profile details: edit config.md
• To change how Sage writes or behaves: edit SKILL.md (plain English, no code)
• To reset and start fresh: delete config.md and all memory/ files, then run “Install Sage” again

---

Formatting Philosophy

Sage writes every tweet with line breaks and breathing room.

Good:

this is the hook.

this is the build.

this is the punchline.

Not good:

this is the hook. this is the build. this is the punchline all in one dense block that nobody reads.

Every line earns the next one. Walls of text get skipped. Sage knows this and applies it to every draft, always.

---

FAQ

Does this work for any niche? Yes. Onboarding calibrates Sage to your niche, voice, and the accounts you watch.

Do I need to be technical? No. If you can install OpenClaw and answer 10 questions, you are set.

Does Sage post automatically? No. Sage drafts and queues. You review and approve before anything goes out.

Is my data private? Yes. config.md and all memory/ files are gitignored and stay on your machine.

What model works best? Claude Sonnet or GPT-4o. Both work well. Sonnet tends to nail conversational voice a bit better.

Can I use this for LinkedIn or other platforms? The SKILL.md is plain text. You can fork it and adapt the formatting rules and output format for any platform.

---

About

Built by Sharbel - founder, AI builder, operator.

Video that goes with this repo: [coming soon - link in bio when it drops]

If this helps you, star the repo. More agent builds on the channel.

---

License

MIT - free to use, modify, and share.