Rules vs prompts: what to write once and what to say every time
·5 min read
Some instructions belong in the message. Some belong in a rule your assistant reads every time. A simple test for telling them apart, with a before and after.
Loading page…
Your assistant answers from what it can see, and it cannot see most of what you know. What goes on its desk, what never does, and how to close the gap.
Build one reusable skill end to end: pick the task, write the file, add it to Claude, ChatGPT, Cursor, or Codex, and test that it fires. No code.
A step-by-step tutorial for dispatching 25 concurrent Claude Code agents into isolated git worktrees with a file-ownership matrix that prevents merge conflicts.
A postmortem on the MCP production incident that a single console.log caused — with a minimal repro and the structural fix.
Why monthly retainers warp your incentives and how a lump-sum model keeps the focus on outcomes rather than relationship survival.
A decision matrix for choosing the right product based on where you're stuck and how fast you need to move.
A founder essay on shipping pay-once MCP and Claude Code products instead of courses, cohorts, or retainers.
Six months of stdio framing bugs, FD leaks, and retry-storm misfires — the gotchas that cost me hours before I had names for them.