▶ quest log

The Four .claude/ Files That Matter: CLAUDE.md, Rules, settings.json, Hooks

Claude Code / Skills◆ Intermediate◆ Advanced@raycfusaved MAR 30

◇ Could this help me?

Yes. These four layers map to real Claude Code features, and two of them (settings.json allow/deny permissions and hooks that block dangerous bash or gate completion on passing tests) genuinely harden a workflow rather than just speed it up, so adopting them improves safety. Two accuracy notes when applying it: the under-20-lines figure is best read as keep it focused, not a hard cap, and path-scoped rules in Claude Code are typically achieved with nested CLAUDE.md files or imports rather than an arbitrary rules folder. This is a general Claude Code workflow update, not project-specific.

✦ Walkthrough+10 xp / step

CLAUDE.md: put your build commands, architecture decisions, naming conventions, and gotchas here, since it is read before every session. Keep it short and focused (the creator suggests roughly under 20 lines) so nothing important gets ignored.
Rules folder: instead of one giant instruction file, split rules by concern (code style in one file, testing standards in another, API conventions in another) and scope them by file path so a given rule set only loads when working in the relevant folder.
settings.json: define the permission and safety rules. Specify what can run without asking, what is blocked entirely, and what requires confirmation first. A safe default is to allow build scripts and deny anything that touches .env files or runs destructive commands.
Hooks: scripts that fire automatically at specific points. Examples: auto-format every time a file finishes being edited, block dangerous bash commands before they execute, and run the test suite so the assistant cannot report done until tests pass.

0 / 4 steps · adopt for +40

$ suggested_prompt+40xp

Review this project's .claude/ setup and propose improvements across four areas: (1) trim CLAUDE.md to focused build commands, architecture decisions, naming conventions, and gotchas; (2) split instructions by concern (code style, testing, API conventions) and scope them by file path using nested CLAUDE.md files or imports; (3) draft a settings.json permissions block that allows safe build scripts, requires confirmation for sensitive actions, and denies anything touching .env or destructive commands; (4) suggest hooks to auto-format after edits, block dangerous bash before it runs, and run tests before reporting done. Show me each proposed file and the exact permission and hook rules before applying anything, and do not enable any hook or permission change until I approve it.

The Four .claude/ Files That Matter: CLAUDE.md, Rules, settings.json, Hooks

More from Claude Code / Skills

Self-Improving Agent Skills From an Open Source Optimizer

Chaining Plan, Goal, Background Agents, and Review

A Claude skill that heals your other skills after each session