Great nugget here for those building agents
Too many skills can eat your context
Claude Code's /plugin Stats tab shows which of your skills Claude has never used and what each one costs you in context. Didn't know this existed
Skills sit in context on every turn and past 1% of the context window Claude Code starts cutting skill descriptions
That means Claude loses the keywords it uses to match a plain request and stops picking that skill on its own starting with the least-used first (from its skills docs)
run:
/plugin stats
to turn one off without uninstalling, in ~/.claude/settings.json:
"skillOverrides": { "skill-name": "off" }
Nice, @kentcdodds
I just aced the Roadster test lab on Kody: kentcdodds.kody.run/packages…
🧍”Agents are here. Dashboards are dead!”
🤖”I’ve built your dashboard in Hubspot.”
What a time to be alive
Dammit, @posthog is sooo good with AI.
It's night and day making dashboards in hubspot.
and the debugging, is chef's kiss.
ai native all the way!
Always be tryin’
The #1 marketing lesson from Monaco:
Monaco has our first marketing hire starting today. I’ve been doing marketing for the company with <10% of my time. I don’t have much marketing experience prior to Monaco (I’ve been in sales most of my career). My #1 lesson: just try things.
The animation below represents my experience perfectly. The data isn’t mine, but anecdotally seems correct. 9 out of 10 startups never try something new. Most that do, fail and quit trying new things. The top 1% just keep shooting. Keep shooting.
Alex Rapp retweeted
Starting to think the "Super” El Niño predicted this winter is just big ski pass controlling the media
Final day of @clerk Growth Shop: measurement and testing.
As onboarding moves into the CLI, traditional web attribution only tells part of the story.
We’ve begun to map a new funnel & started evals with @gauge_sh & @2027dev to understand performance & find where agents get stuck.
This frames the attribution challenge clearly without feeling uncertain & gives us forward momentum for measurement framework.
Day two of @clerk growth shop pushed agent tooling forward:
- Started mapping every agent entry point to a consistent CLI path
- Refined CLI vs. MCP roles & dashboard coverage
- Surfaced ~250 API commands agents often miss
- Moved CLI-first Next.js quickstart closer to merge















