Describe how to use Claude.

• Getting Claude Code to do work for you with /loop and skills + python scripts
• claude-code
• Agent SDK overview
• Common workflows

CLAUDE.md — persistent instructions

• ./CLAUDE.md (project, shared via git), ./CLAUDE.local.md (personal, gitignore it), ~/.claude/CLAUDE.md (user-wide, all projects)

• Generate one automatically with /init — it inspects your codebase and drafts build commands/conventions

• Claude works with two types of configuration files

  • Rules - act as persistent system instructions that the AI must follow across all prompt iterations in a project.
  • Skills - modular directories containing instructions, helper scripts, and resources.
    • The agent will only load a skill's contents dynamically when the task is relevant, keeping the main context window. clean.

• Gemini works with two levels of configuration files

  • Global -
    • rules - ~/.claude/CLAUDE.md
    • skills - ~/.claude/skills//SKILL.md
      • Structure — each skill is a directory:
        • my-skill/
          • SKILL.md (required)
          • reference.md (optional supporting docs)
          • scripts/
            • helper.sh
  • Project specific -
    • rules - ./CLAUDE.md
    • skills - .claude/skills//SKILL.md

  SKILL.md frontmatter:

  name: my-skill description: What this does, and when to use it # required — drives auto-invocation disable-model-invocation: true # optional: make it user-only (typed /my-skill) allowed-tools: Bash Read # optional: pre-approve tools