Claude Code Memory, Rules, Skills, and CI
Put stable guidance where its scope is true, and executable constraints where failure is unacceptable. Design project and user instruction hierarchy without context bloat. Choose CLAUDE.md, path rules, Skills, commands, agents, hooks, and settings by purpose. Author and distribute a real multi-file SKILL.md package with narrow tool grants. Use plan, direct execution, and bounded subagents with explicit obstacle reports. Configure headless Claude Code for reproducible CI evidence. Prevent stale memory, broad permissions, and hidden local configuration from controlling team work. A team keeps every instruction in one root CLAUDE.md: architecture history, formatting, database rules, deployment steps, personal preferences, commands, and examples for six languages. It is copied into every task. Developers add private overrides. CI has a different configuration. One command assumes write access. A broad hook reformats unrelated files. The instructions say "always run every test," so a small docs edit triggers a 40-minute suite. When the agent ignores a safety rule, the team adds more bold text. The problem is not insufficient instruction. The problem is scope, precedence, progressive disclosure, and confusing guidance with enforcement. Product note, verified 2026-08-09: custom commands have been merged into Skills. Files under .claude/commands/ remain compatible, while .claude/skills/ /SKILL.md is the preferred package for new workflows. Exact fields, precedence, and product availability can change. Verify the current Claude Code documentation before implementation. The…
Claude Code Memory, Rules, Skills, and CI: Put stable guidance where its scope is true, and executable constraints where failure is unacceptable.
This free lesson is part of the AI Engineering from Scratch curriculum. Read the full explanation, run the lesson code, and verify the result in the interactive reader or from the repository source.
Browse the complete course catalog or open this lesson on GitHub.