Claude Code project structure explained: CLAUDE.md, commands, skills and agents
Aiden_Tech_Ai · x · 2026-10-24
A widely shared infographic breaks down the recommended Claude Code project layout and what each file does:
- CLAUDE.md: auto-loaded every session for project overview, tech stack and coding conventions; CLAUDE.local.md holds personal overrides kept out of git
- .mcp.json: MCP server integrations (GitHub, JIRA, Slack, databases), shared across the team via git
- settings.json / settings.local.json: permissions, hooks (PreToolUse/PostToolUse for auto-linting and branch protection)
- commands/.md: custom slash commands with $ARGUMENTS support, great for PR reviews, deploys and scaffolding
- skills/<name>/SKILL.md: auto-triggered by task context or invoked via /skill-name, loaded on demand to keep context light
- agents/.md: specialized sub-agents with isolated context, own tools and permissions, invoked with @agent-name
More from coding & agent
- Disappointed with Codex, developer has it write a handoff file for Claude to take over — lxfater · 2026-10-24
- terrio v0.5.0 ships pure Rust geospatial engine with no GDAL, adds SQL and GRIB support — giswqs · 2026-10-24
- DeepSeek Flash-1017 pitfalls: broken budget counting, poor ambiguity handling, 'stingy' mode — karminski3 · 2026-10-24
- When AI agents disagree, who decides? Why the orchestrator shouldn't quietly become the decision-maker — Environmental_Fly371 · 2026-10-24
- A buyer's checklist: 12 questions to ask before adopting an MCP server vendor — pournasserian · 2026-10-24
- Three principles for personal agents: usage-based billing, bot hierarchies, cross-device sharing — sujingshen · 2026-10-24