enlighter-content/project_51_claude_code/4_40004_creating_semi_deterministic_commands.html at 6cd6b8e53c1f9a88afb2b1a296d2dd0d3590a48a · hyperskill/enlighter-content · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
<!-- Enlighter Metainfo
{
  "id": 40004,
  "title": "Creating Semi-deterministic Commands",
  "next_button_title": "Let's create commands!"
}
-->

<h5>Creating Semi-deterministic Commands</h5>
<p>You've successfully set up the deterministic part of your workflow with hooks. Now, it's time to build the semi-deterministic part using custom commands. Custom slash commands allow you to give Claude specific instructions and context, empowering it to make intelligent decisions within a well-defined framework.</p>

<h5>Understanding Custom Commands</h5>
<p>Custom commands are markdown files that act as templates for prompts. They allow you to create reusable, context-aware instructions for Claude. This is the core of the semi-deterministic approach: you define the rules and the goals, and Claude uses its intelligence to figure out the best way to achieve them.</p>

<h5>Step 1: Create a Commands Folder</h5>
<p>First, we need a place to store our custom commands. Claude Code looks for a <code>commands</code> directory inside the <code>.claude</code> folder.</p>

<callout>Create a directory for your custom commands:

mkdir .claude/commands
</callout>

<h5>Step 2: Create Custom Slash Commands</h5>
<p>Now, let's ask Claude to create three essential commands for our web development workflow: one for adding features, one for reviewing code, and one for debugging.</p>

<callout label="Copy and paste to Claude Code">

Create three custom slash commands in the `.claude/commands/` directory:

1.  `/add-feature.md`: A command for adding new functionality. It should analyze the current code, suggest architectural solutions, and implement the feature, including tests.
2.  `/review-code.md`: A command for analyzing code quality. It should check for readability, performance, and security, then suggest improvements.
3.  `/debug-issue.md`: A command for finding and fixing bugs. It should analyze the code, help reproduce the problem, and suggest a solution.

Each command should be able to accept parameters through the `$ARGUMENTS` variable.
</callout>

<h5>Step 3: Create a Project Context File</h5>
<p>To help Claude make better decisions, we should provide it with high-level context about our project. The <code>CLAUDE.md</code> file in the root of your project is the perfect place for this.</p>

<callout label="Copy and paste to Claude Code">

Create a `CLAUDE.md` file in my project root. This file should describe the architecture of our Task Manager application. Include:
- A general description of the application's functionality.
- The file structure and the purpose of each file.
- The coding standards for this project.
- The principles that the AI should follow when working with the code.
</callout>

<alert>The CLAUDE.md file acts as a master prompt, providing Claude with a consistent set of guidelines for all its operations on your project.</alert>

<h5>Expected Results</h5>
<p>After completing this stage, your workflow will be significantly more powerful:</p>

<ul interactive>
  <li title="A .claude/commands/ directory with three custom commands"></li>
  <li title="A CLAUDE.md file in the project root"></li>
</ul>

<h5>What's Next</h5>
<p>You have now established a complete semi-deterministic workflow with both automated hooks and intelligent custom commands. In the next stage, we'll put this workflow into practice by developing new features for our Task Manager application.</p>