Posts

An agent harness checklist: tools, memory, permissions, approval gates

What has to exist around the model before an agent is safe on real work: tools, skills, memory, permissions, budgets, approval gates and evals.

Your CLAUDE.md is probably costing you more than your prompts

My CLAUDE.md grew to 46KB of project history that Claude Code re-sent on every prompt. Here is what it cost me and how I cut it to under 1,000 words.

Why I built Releaslyy — the problem every sprint team ignores

After five years as a tech lead I kept writing the same release update by hand. Here is why I built a tool that writes it for QA, product and customers.