Featured
A layered CLAUDE.md, tight permissions, hooks, skills and scoped MCP servers: the setup that keeps Claude Code safe and cheap across dozens of repositories.
13 min
Topic 06 / 08 · 03 articles
About this topic
121 words
AI coding agents are now part of how serious software gets built, but most advice about them is either hype or a list of prompts. What matters in practice is the environment around the agent: the instructions it reads, the permissions it has, the tools it can call and the review process its output goes through.
Here I document how I set up and operate AI-assisted engineering on a large multi-repository codebase — project instruction files, permission allow-lists, hooks, reusable skills, subagents, MCP servers and memory — and what I have learned about keeping quality high while moving much faster than I could alone.
These are working notes from daily use, updated as the tools change, with the configuration files included.
Featured
A layered CLAUDE.md, tight permissions, hooks, skills and scoped MCP servers: the setup that keeps Claude Code safe and cheap across dozens of repositories.
13 min
· 11 min
Thirty well-chosen cases, deterministic checks first and a model judge last, run several times per case. A practical eval setup a single developer can maintain.
11 min
· 11 min
Hook exit codes that do not block, rules that never load, skills that abort and memory that stays on one machine: the failure modes and the fix for each.
11 min
· 13 min
A layered CLAUDE.md, tight permissions, hooks, skills and scoped MCP servers: the setup that keeps Claude Code safe and cheap across dozens of repositories.
13 min