Agent System Components


  ┌─ Claude Code CLI ──────┐  ┌─ File System ──────────┐  ┌─ Execution ──────────┐
  │                        │  │                        │  │                      │
  │  Slash Commands        │  │  ~/.claude/commands/   │  │  Direct prompt       │
  │  /command-name     ────┼──▶  (.md files)       ────┼──▶  injection           │
  │                        │  │                        │  │                      │
  │  Agents                │  │  ~/.claude/agents/     │  │  Autonomous          │
  │  Task tool spawns  ────┼──▶  (.md personas)    ────┼──▶  sub-agent           │
  │                        │  │                        │  │                      │
  │  Skill Packs           │  │  Skill directories     │  │  Multi-tool          │
  │  Multi-step        ────┼──▶  (SKILL.md + Tools/)───┼──▶  orchestration       │
  │                        │  │                        │  │                      │
  └────────────────────────┘  └────────────────────────┘  └──────────┬───────────┘
                                                                     │
                                                                     ▼
                                                          ┌─────────────────────┐
                                                          │ Python Scripts /    │
                                                          │ APIs / Vault        │
                                                          └─────────────────────┘

Agents

Agents are .md files in ~/.claude/agents/ that define specialized personas with specific expertise, tools, and behavior patterns.

How they work: Claude Code’s Task tool spawns sub-agents that run autonomously. Each agent definition tells the sub-agent what it knows, what tools to use, and how to behave.

Example categories:

Category	Examples
Quality	Bar-raiser (Amazon-style review), quality orchestrator
Content	Writer, content analyzer, learning teacher
Infrastructure	Calendar manager, data orchestrator
Specialist	Researcher, coder, operator, coach

Creating a New Agent

Add a .md file to ~/.claude/agents/:


# Agent: agent-name
 
Brief description of expertise and purpose.
 
## Capabilities
- What this agent can do
- What tools it uses
 
## Behavior
- How it should interact
- Quality standards
 
## Context
- Domain knowledge
- Reference files to read

Commands (Slash Commands)

Commands are .md files in ~/.claude/commands/ that define reusable workflows triggered via /command-name.

How they work: Type /command-name in Claude Code. The .md file’s content becomes the prompt, often with embedded bash scripts and multi-step instructions.

Core Commands (Available to All)

Command	What It Does
`/contxt`	Load full system context, route to domain
`/contxt [domain]`	Load domain-specific context (e.g., `/contxt coding`)
`/resume`	Resume where you left off after interruption
`/audit`	Run vault health check (deterministic)

Example Workflow Commands

Command	What It Does
`/weekly-review`	Generate weekly review from data sources
`/deploy`	Commit, push, deploy to hosting
`/ship`	Commit + push current project
`/verify-changes`	Verify code changes work correctly
`/dev-cleanup`	Kill orphaned processes, clean lock files

Creating a New Command


# Command Name
 
Description of what this command does.
 
## Steps
 
1. Step one instructions
2. Step two instructions
 
$BASH
echo "Shell commands go here"
$ENDBASH

Skill Packs

Skills are directories containing multiple files that implement complex, multi-step workflows.

Each skill typically contains:

SKILL.md — definition and entry point
Tools/ — CLI tools (TypeScript or Python)
Workflows/ — multi-step workflow definitions

Example skills: Browser automation, file organization, prompt engineering, voice conversations, weekly reviews.