Short-Form Content

Heads up: this is still really early. I've built the pieces out but I'm still iterating on how well it all works in practice. The pipeline exists and produces drafts, but I'm still figuring out how to get consistent output that sounds like me and not like a content creator. Take everything here as a starting point that's actively being developed.

Tweet Pipeline

The idea is to pull insights from the vault -- literature notes, reference notes, recent research -- and turn them into tweet drafts that match my writing voice. I built a set of 50+ tweet templates, but the key is that they're filtered through VOICE.md so they don't sound like templates.

Workflow Modes

1. Single note -- turn one specific vault note into a tweet
2. Daily batch -- agent picks 2-3 interesting notes and drafts tweets
3. Theme thread -- build a thread from related notes on a topic

Setup Prompt

Create a tweet pipeline skill. It should:

1. Pull interesting insights from my vault (literature notes,
   reference notes, recent additions)
2. Match them against tweet templates:
   - Contrarian take
   - Hard truth / uncomfortable observation
   - Before/after comparison
   - Simple observation or reflection
   - Question to the audience
   - Sharing something specific and interesting
3. Write 2-3 draft tweets in my voice (reference VOICE.md)
4. Save drafts to workspace/tweets/drafts/YYYY-MM-DD.md
5. NEVER post without my explicit approval

Goal: 2-3 tweets per day with minimal effort on my part.

Key rule: The tweets should sound like me sharing something
interesting, NOT like a content creator trying to go viral.
Reference VOICE.md for my actual patterns and anti-patterns.

Template Examples

Templates that don't sound like templates:

• Sharing a finding: "[Source] found that [specific thing]. [Brief reaction]"
• Observation: "I started [doing X] and now [unexpected result]"
• Question: "For the [audience] out there: [genuine question]?"
• Contrarian: "[Common belief]. But [what I actually found]."

Always review drafts before posting. The agent drafts, you approve.

X Writing System

This is different from the pipeline. Instead of generating tweets from scratch, this takes a specific draft post you've already written and tries to improve it -- looking at writing guidelines, your top-performing posts, topic-specific winners, and current trends. It produces 5 variations to pick from.

Setup Prompt

Create an X writing system skill. When I give you a draft post:

1. Analyze it against writing guidelines (Matt Gray style: hook,
   value, CTA patterns)
2. Compare against my top-performing posts (reference VOICE.md)
3. Search for topic-specific winners on X
4. Check current trends related to the topic
5. Produce 5 rewrite variations, each with a different angle

The rewrites should sound like ME, not like a content strategist.
Use VOICE.md as the voice filter for every variation.

When to Use Which

• Tweet pipeline -- daily low-effort drafts from vault notes, when you want content but don't have a specific idea
• X writing system -- when you have a specific draft and want to make it better

Where I Want This to Go

The bigger goal isn't really about going viral or optimizing engagement. It's about building a steady cadence of sharing what I'm working on, thinking about, learning, and researching -- and using my agent to give me leverage in actually doing it consistently.

Left to my own devices, I'll forget to post for days. I'll have ideas and not write them down. I'll do interesting research and never share any of it. The agent can solve that by both generating content ideas and holding me accountable to a posting rhythm.

What I want to build toward:

Ideation -- The agent already has access to everything I'm reading, researching, and working on through the vault. It should be able to surface interesting things to post about without me having to think of them. "You processed three papers on AI reasoning this week -- here are some angles worth sharing." That kind of thing.

Drafting across platforms -- Right now this is focused on Twitter. I want to extend it to LinkedIn too, with a cadence of a few posts per day across both platforms. The content would be based on the same source material (vault notes, research, projects) but adapted for each platform's style. LinkedIn is more professional and longer-form, Twitter is more casual and observational.

Automated cadence -- Ideally the agent would batch-produce drafts on a schedule. Maybe a cron job that generates a handful of drafts each morning for me to review, edit, and approve. I still want to review everything before it goes out -- the automation is in the production, not the publishing.

Accountability -- Part of the value is just having the agent nudge me. "You haven't posted in two days, here are three drafts based on what you've been working on." Not annoying, just a gentle push to keep the muscle going.

The core thing I care about: the content sounds like me sharing things I'm genuinely interested in with people I want to have conversations with. Not content-creator energy, not trying to game algorithms. Just a consistent presence built on what I'm actually doing and thinking about, with my agent handling the work of turning that into posts.

Tool to explore: xurl -- the official CLI for the X API, built by the X developer platform team. Supports OAuth, posting, media uploads, streaming. If gog is what gives the agent access to Google, xurl could be what gives it access to X. Haven't set it up yet but it's the obvious next piece for automating the posting side of this.