#developertooling

My skill evals scored exactly zero for as long as I had records. The lesson was not that the skills were bad. Only a broken instrument writes the same

Claude Code Save Conversation: Where Transcripts Live Claude Code hit a $1B annualized...

An MCP delegation layer that keeps Cursor or any MCP host as the orchestrator while local Ollama specialists handle narrow, context-heavy work.

How to write a CLAUDE.md that actually helps Claude Code: what to include, what to leave out, a real structure, and how to stop it from rotting.

Our verification gate let zero-change PRs pass because there were no files to check. The quality checks we built were never enforced.

Claude Code Errors: The Definitive Troubleshooting Reference (Every Code, Cause, and...

A 5-tier scoring system validated against 17 real repos replaced flat scoring that broke with real data. Content-aware matching fills the gaps.

100% line coverage means every line was executed, not that meaningful scenarios were tested. Quality checks fill the gap after 100%.

Given an implementation file, which test files are relevant? We tried stem matching, content grepping, and hybrid discovery. Each fails differently.

This post was originally published on graycloudarch.com. Three months ago I was re-explaining my...