Where does all the AI context actually live? The answer is simpler than you'd expect: files. A PARA-based system with progressive disclosure that keeps Claude operating under 40% of its context window.
The official Ralph loop plugin isn't true Ralph. Here's how I built a plugin that installs project-specific configuration via a skill, supporting both same-session and fresh-context modes for autonomous TDD workflows.
Understanding the philosophy of skills-over-agents doesn't tell you how to implement it. Here's how to build composable AI capabilities that don't duplicate code everywhere.
I have a /daily-brief command that runs 18 steps. Today it skipped two steps without me noticing. Here's how Claude Code's built-in TodoWrite tool fixes this with the checklist manifesto pattern.
Most AI agent projects fail—not because models aren't smart enough, but because developers keep solving the wrong problem. The shift from building agents to building skills changes everything.