Brandon Cardoso

Claude Code

Notes seedling created edit
ai
On this page ▼

Key Concepts

Skills and Commands

.claude/skills/ .claude/commands/

  • skills are stored workflows targeting a specific scope, shorthand for prompts structured for a specific task
    • e.g. a “refactor/clean” skill goes through and refactors all Code
  • skills are a broader workflow definition
  • commands are skills that are executed by a slash command
  • commands are quick executable prompts

Hooks

.claude/hooks/

  • trigger based automations that fire on specific events
  • restricted to tool calls and lifecycle events

Hook Types

  • PreToolUse - before a tool executes*
  • PostToolUse - after a tool has finished executing
  • UserPromptSubmit - when user sends a message
  • Stop - when Claude finishes responding
  • PreCompact - before context is compacted
  • Notification - on permission request

Subagents

~/.claude/agents/

  • processes that are orchestrated by Claude to delegate tasks with limited scopes
  • can run in background or foreground
  • a subagents can be assigned a subset of skills, and sandboxed with specific tool access

Rules and Memory

  • define rules and best practices that Claude should always follow

Two approaches:

  • single .claude/CLAUDE.md file with everything, user or project level
  • multiple .claude/rules/<...>.md files grouped by concern: security, testing, coding-style, etc.

Example Rules

  • never commit console.log
  • follow SOLID principles
  • always test code before deployment

Model Context Protocol (MCP)

  • an MCP connects Claude to external services
  • prompt driven wrapper around an API, allowing more flexibility

CRITICAL - Context Window Management

  • keep unused MCPs disabled
  • performance/context usage degrades rapidly with many enabled

Plugins

  • packaged tools for easy installation
  • can contain skills, MCP, hooks, tools, or some combination

Tips and Tricks

/fork - forks a conversation to do non-overlapping tasks in parallel instead of queuing messages

git worktrees - for overlapping parallel Claude instances without conflicts. Each worktree is an independent checkout

Useful commands

  • /rewind - go back to previous state
  • /statusline - customize the Claude status line (branch name, context %, todos, etc)
  • /compact - manually trigger context compaction

Context and Memory Management

everything claude code - structured claude code configs for full product development