Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more

To tailor Claude Code to a project, use three files for three different jobs: CLAUDE.md for lasting project context, a skill’s SKILL.md for reusable task procedures, and .claude/settings.json for settings and hooks. They do not make Claude universally smarter; they give it relevant guidance and, in the case of hooks, can automate predictable actions.

These are practical choices, not an official Anthropic-mandated trio. Anthropic documents additional customization options, including path-scoped rules. Choose the mechanism that matches what you want Claude Code to remember or do.

1. Put project-wide context in CLAUDE.md

A root CLAUDE.md is a project briefing that Claude Code reads at session start. Use it for information that is useful across many tasks, such as how to build and test the project, how its directories are organized, a brief architecture overview, coding conventions, and team norms. Anthropic’s Help Center also describes a user-level file at ~/.claude/CLAUDE.md for context you want available across projects.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Keep the file focused: instructions that load routinely consume context even when a particular task does not need them. Anthropic’s June 18, 2026 guidance recommends keeping CLAUDE.md under 200 lines, assigning it an owner, and reviewing changes like code. The line count is a recommendation, not a product limit. Put detailed procedures elsewhere, and move guidance that applies only to certain paths into path-scoped rules or relevant subdirectory guidance.

For more on the file’s role and common locations, see Anthropic’s CLAUDE.md and context guide.

2. Put repeatable task procedures in a skill’s SKILL.md

A skill is a reusable workflow for a particular kind of work: for example, a release checklist or a code-review process. Its SKILL.md can hold the detailed steps, supporting material, and task-specific instructions that would clutter the project briefing.

Skills differ from always-loaded context. Claude Code can use a skill when it is invoked or when it identifies the task as relevant; the skill’s full instructions load when needed. Its name and description may be available at session start, while the complete procedure remains on demand. This makes a skill a better fit for a procedure used occasionally than for a project fact that should inform most sessions.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

3. Put configuration and hooks in settings.json

Settings files configure Claude Code behavior, including permissions and hooks. A project can register hooks in .claude/settings.json. Hooks are useful when an action should happen predictably—for instance, running a linter after edits or blocking a command before it executes. Unlike instructions that rely on Claude interpreting prose, command hooks can provide a deterministic trigger and action.

Anthropic’s hook guidance also distinguishes command hooks from HTTP and MCP tool hooks, and from prompt and agent hooks, which rely on Claude’s judgment. For safe hook scripts, Anthropic advises validating and sanitizing input, quoting shell variables, using absolute paths, and avoiding sensitive files such as .env and credentials. See the Claude Code hooks configuration guide for details.

Where the setting lives determines its scope: project settings can be shared with a repository, user settings apply across projects, and local project settings are intended for personal configuration that is not shared. Use the narrowest appropriate scope, especially when a setting or hook should not affect teammates.

Which file should you use?

Need Use How it applies
Context useful in most tasks CLAUDE.md Root guidance is loaded at session start.
A reusable workflow for particular tasks A skill’s SKILL.md The full procedure loads when invoked or judged relevant.
A constraint only for matching files Path-scoped rules Guidance is scoped to the paths it matches, rather than added to every session’s briefing.
A setting or repeatable automated action settings.json and, when appropriate, a hook Settings configure behavior; hooks can trigger predictable actions.

In practice, ask what should happen: if Claude should know it across most sessions, put it in CLAUDE.md; if it should follow a procedure for a particular kind of task, make a skill; if a rule applies only to certain files, use path-scoped rules; if an action needs a reliable trigger, configure a hook.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

What these files can—and cannot—do

Anthropic’s June 18, 2026 article describes multiple customization methods and their trade-offs, including when they load and how much context they use. These files help tailor Claude Code’s behavior to project conventions and task requirements; they are not evidence of a guaranteed improvement in coding quality or productivity. No universal performance gain follows from creating them. Keep instructions specific, choose the right scope, and use hooks for actions that should not depend solely on interpretation.

For Anthropic’s broader account of these customization options, read “Steering Claude Code: when to use CLAUDE.md, skills, hooks, rules, and subagents”.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.