---
name: Reports
description: A comparison of the three extension mechanisms in Claude Code: subagents, commands, and skills. <table width="100%"> <tr> <td><a href="../">← Back to Claude Code Best Practice</a></td> <td align="r...
---

# Reports

A comparison of the three extension mechanisms in Claude Code: subagents, commands, and skills. <table width="100%"> <tr> <td><a href="../">← Back to Claude Code Best Practice</a></td> <td align="right"><img src="../!/claude-jumping.svg" alt="Claude" width="60" /></td> </tr> </table> ![Slash menu showing time-skill, time-command, and time-agent](assets/agent-command-skill-1.jpg) ---

---
name: my-agent
description: Use this agent PROACTIVELY when...
tools: Read, Write, Edit, Bash
model: sonnet
maxTurns: 10
permissionMode: acceptEdits
memory: user
skills:
  - my-skill
---
```

### Command Frontmatter

```yaml
---
description: Do something useful
argument-hint: [issue-number]
allowed-tools: Read, Edit, Bash(gh *)
model: sonnet
---
```

### Skill Frontmatter

```yaml
---
name: my-skill
description: Do something when the user asks for...
argument-hint: [file-path]
disable-model-invocation: false
user-invocable: true
allowed-tools: Read, Grep, Glob
model: sonnet
context: fork
agent: general-purpose
---
```

---

## Key Distinctions

### Auto-invocation

| Mechanism | Can Claude auto-invoke? | How to prevent |
|-----------|------------------------|----------------|
| Agent | Yes — via `description` (use "PROACTIVELY" to encourage it) | Remove or soften the description |
| Command | No — always user-initiated via `/` | N/A |
| Skill | Yes — via `description` | Set `disable-model-invocation: true` |

### Visibility in `/` menu

| Mechanism | Appears in `/` menu? | How to hide |
|-----------|---------------------|-------------|
| Agent | No | N/A |
| Command | Yes — always | Cannot be hidden |
| Skill | Yes — by default | Set `user-invocable: false` |

### Context isolation

| Mechanism | Runs in own context? | How to configure |
|-----------|---------------------|-----------------|
| Agent | Always | Built-in behavior |
| Command | Never | N/A |
| Skill | Optional | Set `context: fork` |

---

## Worked Example: "What is the current time?"

This repository has all three mechanisms defined for the same task — displaying the current time in PKT. Here's what happens when a user types **"What is the current time?"** without explicitly invoking any `/` command:

| Mechanism | Will it fire? | Why / Why not |
|-----------|--------------|---------------|
| `time-command` | No | Commands are **never auto-invoked**. The user would need to explicitly type `/time-command` for it to run. Commands have no auto-discovery pathway — they are strictly user-initiated. |
| `time-agent` | **Yes** (possible) | The agent's `description` says *"Use this agent to display the current time in Pakistan Standard Time"*. Claude matches this against the user's intent and may spawn it via the Agent tool. However, agents run in a **separate context window**, making them heavier than necessary for this simple task. |
| `time-skill` | **Yes** (most likely) | The skill's `description` says *"Display the current time in Pakistan Standard Time (PKT, UTC+5). Use when the user asks for the current time, Pakistan time, or PKT."* Claude matches this and invokes it via the Skill tool. Since it runs **inline** with no context overhead, it's the most efficient match. |

### Resolution order

When multiple mechanisms match the same intent, Claude prefers the **lightest-weight option** that satisfies the request:

```
1. Skill (inline, no context overhead)     ← preferred
2. Agent (separate context, autonomous)    ← used if skill is unavailable or task is complex
3. Command (never — requires explicit /)   ← only if user types /time-command
```

### What if `disable-model-invocation: true` were set on the skill?

Then Claude **cannot** auto-invoke the skill. The agent becomes the only auto-invocable option, so Claude would spawn `time-agent` instead — at the cost of a separate context window for a one-liner bash command.

### What if both skill and agent had auto-invocation disabled?

Then **nothing fires automatically**. Claude would fall back to its own general knowledge and likely just run `TZ='Asia/Karachi' date` directly — no extension mechanism involved. The user would need to explicitly type `/time-command` or `/time-skill` to use one.

![Claude auto-invoking time-skill when user asks "What is the current time?"](assets/agent-command-skill-2.png)

---

## Sources

- [Claude Code Skills — Docs](https://code.claude.com/docs/en/skills)
- [Claude Code Sub-agents — Docs](https://code.claude.com/docs/en/sub-agents)
- [Claude Code Slash Commands — Docs](https://code.claude.com/docs/en/slash-commands)
- [Skills Best Practice](../best-practice/claude-skills.md)
- [Commands Best Practice](../best-practice/claude-commands.md)
- [Sub-agents Best Practice](../best-practice/claude-subagents.md)


---

**Source**: https://github.com/shanraisshan/claude-code-best-practice
**Author**: shanraisshan
**Discovered via**: skillsdirectory.com
**Genre**: ai-agents