Elias Stråvik

<start of post>

I’ve spent the last week obsessed with a tiny detail that most people ignore: agentic memory.

I was building a new workflow for /gtm-agent to handle inbound leads. The logic was sound, the prompts were sharp, and the API calls were hitting 200s every time.

But as I was testing it, I realized something frustrating. Every time the agent started a new session, it forgot everything we had discussed about our specific ICP nuances from the day before.

If you’re just doing one-off tasks, this doesn't matter. You just paste the context and move on.

But I wanted to build something that actually gets smarter the more your team uses it. A shared brain for your GTM operations that lives in your terminal but thinks like your best sales rep.

So I built a lightweight vector-sync for /gtmskills. It’s a simple way to let your local Claude or Codex instance pull from a shared "context-repo" that updates in real-time as you chat.

What’s really cool is that you don't need a complex database setup for this. It just uses a flat-file structure in a private GitHub repo that your agent can read and write to.

I’ve been running it for three days now, and the difference is night and day. I can ask @gtm-agent to "write a follow-up based on our new positioning" and it actually knows what the new positioning is without me telling it.

It’s 100% free and open source - I’ve just pushed the update to the main repo.

---

npx skills update eliasstravik/gtmskills

---

If you want to see how to hook this up to your team's Slack, I put a quick 2-min Loom in the comments showing the setup.

Lmk if you run into any issues setting up the repo permissions – happy to jump on a call to help!

Enjoy!

<end of post>