Good
Good: source-backed Claude Code project onboarding guidance with a clear trigger, scoped commands or tools, concrete verification, and no duplicated generic advice.
Claude Code courseware
A focused workbench for drafting Claude memory, skills, and subagents before learners use them with Claude Code.
CLAUDE.md at session start; keep this file concise, specific, and under 200 lines..claude/skills/<skill-name>/SKILL.md..claude/rules/*.md with paths frontmatter when the rule only applies to part of the repo.| Material | Starter value |
|---|---|
| Memory location | ./CLAUDE.md for root project memory, or ./.claude/CLAUDE.md when the team keeps Claude config together |
| Claude Code behavior | Plan mode guidance, memory boundaries, rules, skills, permissions, or verification conventions |
| First files to inspect | README files, package manifests, CI config, docs, and existing scripts |
| Common Claude Code task | install dependencies, run the right checks, and make a small safe change |
| Main risk | too much generic guidance hides the few project facts Claude should load every session |
.claude/rules/ so they load only when matching files are relevant.pnpm installpnpm run lintpnpm run typecheckpnpm run testpnpm run build/memory to inspect loaded project memory when behavior does not match this file..claude/rules/testing.md for test conventions that only matter near test files..claude/rules/frontend.md for UI conventions that only apply under frontend paths..claude/skills/<skill-name>/SKILL.md for repeatable workflows with inputs, steps, and outputs.CLAUDE.local.md and gitignore it.CLAUDE.local.md, not in this shared file.~/.claude/CLAUDE.md.Ask Claude Code: "Summarize the CLAUDE.md, CLAUDE.local.md, and .claude/rules files you loaded for this repo. What would you do before a Claude Code project onboarding change?"
template guide
Use these notes as source-backed guidance for the selected artifact. The editor only flags clear errors and warnings; this section stays advisory.
For Claude Code project onboarding, gather README files, package manifests, CI config, docs, and existing scripts.
Use it to help Claude Code sessions entering a shared project install dependencies, run the right checks, and make a small safe change.Choose whether this shared project memory belongs at `./CLAUDE.md` or `./.claude/CLAUDE.md`, then keep it focused on facts Claude should know every session.
Commands, verification expectations, architecture landmarks, repeated corrections, and critical warnings.Keep only the facts Claude should load every session: commands, architecture landmarks, repeat corrections, and hard constraints.
Under 200 lines, concrete commands, no generic advice like 'write clean code'.Move path-specific conventions to `.claude/rules/` and repeatable procedures to `.claude/skills/` instead of bloating CLAUDE.md.
Frontend styling rules in `.claude/rules/frontend.md`; release workflow in `.claude/skills/release/SKILL.md`.Separate personal sandbox URLs, machine-specific shortcuts, and private preferences into CLAUDE.local.md or user-level memory.
Add `CLAUDE.local.md` to `.gitignore`; do not commit private tokens or local-only paths.Keep a prompt that asks Claude Code to summarize which memory files loaded and which instructions apply to a sample edit.
"Which CLAUDE.md, CLAUDE.local.md, and .claude/rules files shaped your plan for this change?"Good: source-backed Claude Code project onboarding guidance with a clear trigger, scoped commands or tools, concrete verification, and no duplicated generic advice.
Bad: generic advice such as "be helpful", broad tool grants, no trigger, no source boundaries, and no way to verify the result.
Explains CLAUDE.md scope, concise instruction writing, load order, local memory, and rules.
Recommends concise project memory, clear verification, skills for reusable workflows, and subagents for focused investigation.
Maps where Claude Code reads project memory, rules, skills, subagents, hooks, settings, and local-only files.
Use this artifact with a short Claude Code project onboarding sample. Ask the agent to install dependencies, run the right checks, and make a small safe change, then report which instruction or reference shaped the result.