I built a daily job to re-check my published posts for staleness. A link sweep catches dead links and a deep review reads one post a day.

Three parallel subagents, ten open Projex issues, one release. I skimmed the diffs at the end, not in the middle.

Anthropic contributed MCP to the world while locking down Claude subscription access and sending legal letters to open source projects. I use one of the tools they targeted.

I broke my entire blog review system trying to reduce prompt overhead so I could fit more messages into Claude's session limits. My 'optimization' saved 1,100 tokens but cost me 5 hours of debugging. Here's what I learned about premature optimization in multi-agent systems.

How I created a four-agent workflow system to transform raw notes into authentic blog posts - complete with debates, convergence protocols, and real expert frameworks.