Fundamentals
Tier 1

Fundamentals

The working parts of Claude Code: context, files as state, CLAUDE.md, tools, MCP, skills, hooks, settings, git, logging and cost. Ends with your first agent.

13 articles71 min total

Start reading →

How it works

  1. Sessions and context windowsWhat a session holds, why the context window fills, what compaction does, and when to start fresh instead.5 min read
  2. Files as state, not chat memoryAnything that must outlive a session goes in a file. The layout that lets an agent system remember its own work.5 min read
  3. CLAUDE.md: teaching the agent your rulesGlobal vs project CLAUDE.md, what belongs in each, and how to keep it small enough that the agent follows it.5 min read
  4. Tools 101Read, Edit, Write, Bash and the rest: how the agent picks a tool, what a result looks like, and why tool use is the signal.5 min read
  5. Prompting Claude Code wellGoals, not steps. Surgical scope. The context to hand over, and the requests that reliably produce a mess.5 min read

Extending Claude Code

  1. MCP: giving the agent new sensesWhat the Model Context Protocol is, how to add a server, and how to choose between an MCP server, a CLI and a skill.6 min read
  2. Skills: packaged know-how the agent loads on demandWhat a skill is, how SKILL.md works, when a skill beats a prompt or a rule, and how to build and test one.5 min read
  3. Hooks: rules the model cannot forgetPreToolUse, Stop and the other hook events, why infrastructure beats instructions, and three real hooks worth copying.6 min read
  4. settings.json and permissions configThe settings hierarchy, allow and deny rules, env vars, hooks wiring, and what goes in user, project and local files.5 min read

Working habits

  1. Git discipline for agentsBranch first, commit small, never force-push, and why parallel agents on one checkout will eventually wreck each other's work.6 min read
  2. Logging and receipts: making agent work inspectableNo artifact, no accountability. One-line ship logs, daily rollups, run ledgers and receipts that say what really happened.6 min read
  3. Cost and model choiceSubscription vs API billing, tiering models by work type, and how to read cost without fooling yourself.6 min read

Capstone

  1. Build your first agentA small agent that runs on a schedule: a definition, rules, one skill, a memory file, and a hard rule that it drafts but never sends.6 min read