Cold memory: specs, MCP tools, and on-demand context retrieval
How subsystem specs and MCP retrieval tools handle architectural knowledge too large for hot memory — and why stale specs are worse than no specs.
Claude-Code
Domain specialist skills: teaching AI to think like your senior dev
What specialist skills are, why the 50% domain knowledge rule matters, and how waaseyaa’s spec-backed orchestration keeps AI consistent across a 29-package PHP monorepo.
Writing a CLAUDE.md that actually works
How to structure your CLAUDE.md as a routing layer so AI agents always know where to look.
Why AI agents lose their minds in complex codebases
Token limits aren’t the real problem with AI in large codebases — inconsistent context is. Here’s what breaks and why a three-tier architecture fixes it.
Building a Codebase Cleanup Skill for Claude Code
How to write reusable Claude Code skills, featuring a codebase cleanup skill I built and ran on this very blog.