162 Markdown documents, one cloud knowledge base, searchable by meaning from Claude Code, Cursor, or n8n — and by your teammates. One command deploys it. The payoff is that a written answer gets made once and reused, instead of re-litigated three times a quarter.
Built a skill Wednesday night, opened a different Mac Thursday morning, and it was gone. The full-mesh Syncthing setup that keeps the config behind my workflow automation software identical on 4 machines - .stignore rules for 360,000 files, an onboarding SOP, and the mistakes I made.
Your knowledge sits in dozens of folders and hundreds of chat threads, so the answers come back generic. Build a knowledge base on your file system instead: the four-layer architecture, a step-by-step build sequence, and a checklist to tick off as you go.
Software development, agent engineering, data and models
Three steps, install, API key, first Skill, and it is doing real work. Every stuck point in setup is one of three known failure modes, each with a one-line fix, plus the right first task to hand it on day one.
Week one was 60% frustration, 40% magic - and the frustration was avoidable. Ten specific mistakes with the actual terminal sessions, the three features that matter before anything else, and a copy-paste rule file. Written by someone using an AI assistant for business, not code.
Seventy tips tested over two months; seven survived. The rest got repeated because they sounded clever, not because anyone showed a before and after. Here are the seven that cut task time, the three that made an AI assistant for business slower, and a copy-paste cheat sheet.
Deleting 75% of an instruction file made the output better, not worse. That inversion is the whole argument. What each rewrite removed, what the final file still contains, and a starter template for business process automation without writing code.
Everyone says build a Skill for anything you do twice. Most workflows are not Skill-shaped. Eight built, five still running, three deleted: full SKILL.md source for the survivors, honest post-mortems on the rest, and the 5-gate test before the next one.
Twelve MCP servers tested over three months: five earned a permanent spot, seven got removed. The short list for non-programmers, with real usage frequency, an honest reason for each cut, and a copy-paste .mcp.json.
Syncthing, a Git repo with cron, or iCloud Drive plus one symlink - I ran all three across six months. Only one avoided becoming a part-time sysadmin job. Honest comparison, exact commands for the winner, a Gist backup plan, and when heavier workflow automation software is right.
When I rebuild one with AI agents, you get the write-up — including the parts that didn't work. No weekly roundup, no "5 tools you need."