Configuration · Instructions: memories, skills, subagents

Instructions are the text that tells your agent what to do. Berth lays them out so you can see what is loaded, from which scope, and how files chain together.

Memories

CLAUDE.md and AGENTS.md are durable instructions the agent reads at the start of work. Berth shows their scope (user / project / enterprise) and resolves the @path import chain — including broken links.

Skills

A Skill packages a reusable procedure in a SKILL.md (plus optional scripts). Claude loads only the name/description until the skill is relevant — "progressive disclosure" — so long procedures cost almost nothing until needed.

Subagents

A subagent is a specialized assistant with its own context window, system prompt, and allowed tools. The main agent delegates focused work to it and gets back a summary — keeping the main conversation clean.

Sources

  1. 01
    Claude Code — Memory

    Primary source; CLAUDE.md scopes and imports.

    https://code.claude.com/docs/en/memory

  2. 02
    Claude Code — Skills

    Primary source; SKILL.md structure and progressive disclosure.

    https://code.claude.com/docs/en/skills

  3. 03
    Claude Code — Subagents

    Primary source; isolated specialized assistants.

    https://code.claude.com/docs/en/sub-agents