Session 4: Managing Work with GitHub

What We’ll Do Today

Today is the final session. We’ll build the “operational framework” for continuously applying Claude Code in your day-to-day work.

Today’s Goal

Create your own repository → Configure CLAUDE.md → Build the Issue → AI execution → PR workflow

GitHub Fundamentals

Four Terms to Learn Today

What Was Happening in Session 0

Let’s revisit the Session 0 experience using today’s terminology:

1. You created an Issue (task ticket)
       |
2. The AI agent created a branch (draft space)
       |
3. The AI worked in the draft space (research, file creation)
       |
4. The AI created a PR (review request)
       |
5. You reviewed and gave feedback on the deliverable

In Session 0, it may have seemed like “magic,” but these five steps were actually happening behind the scenes. Today, you’ll build this system yourself.

Hands-On 1: Creating Your Own Repository

Step 1: Create the Repository

1. Log into github.com
2. Click the ”+” button in the top right → Select “New repository”
3. Fill in the following:

4. Click “Create repository”

Private vs. Public

• Private: Only invited people can access it (recommended for business use)
• Public: Anyone on the internet can view it

Step 2: Create a CLAUDE.md

CLAUDE.md is a “work manual” for the AI agent. When executing tasks, the AI reads this file and follows its rules.

1. On the repository’s main page, click “Add file” → “Create new file”
2. Enter CLAUDE.md as the filename (note the uppercase letters)
3. Use the template below as a starting point and customize it for your work

CLAUDE.md Template

# Project Settings

You are a work assistant for the [X] department.
Execute tasks assigned via GitHub Issues and save deliverables in the output/ folder.

## Work Rules
- Create deliverables in Markdown format
- Use the filename format `YYYY-MM-DD-task-summary.md`
- Always cite sources in research
- Do not handle confidential information
- Create a Pull Request when finished, referencing the Issue

## Areas of Expertise
- Research and information gathering
- Drafting documents
- Data organization and summarization
- Structuring meeting notes

4. Click “Commit changes” at the bottom of the page

Customization Tips

• Replace “[X] department” with your actual department name
• Add your own commonly requested tasks under “Areas of Expertise”
• Add department-specific rules under “Work Rules” (e.g., “Use formal language,” “Keep under 2 pages”)

Step 3: Create an Issue and Execute with Claude Code

1. Go to the “Issues” tab in the repository → Click “New issue”
2. Write your task and click “Submit new issue”
3. Launch Claude Code in the terminal and execute based on the Issue
4. Return to GitHub and check the “Pull requests” tab to confirm a PR was created

Three Team Operation Patterns

Pattern 1: Individual Use

Your own repository → You create Issues → AI executes → You review

Best for:

• Weekly business report drafts
• Meeting notes cleanup
• Information gathering and research

Benefits: Start at your own pace. Failures only affect you.

Pattern 2: Shared Team Repository

Shared team repository → Team members create Issues → AI executes → Team reviews

Best for:

• Building a team knowledge base
• Distributing project research tasks
• Creating and updating internal manuals

Benefits: Deliverables can be shared and reviewed across the team. Quality management happens naturally.

Pattern 3: Cross-Departmental Projects

Project repository → Each department creates Issues → AI executes → Stakeholders review

Best for:

• Preparation tasks across departments for a new product launch
• Company-wide event planning and coordination
• Cross-departmental research projects

Issue Templates

Writing Issues from scratch every time is tedious. Templates let you prepare a format in advance.

Template Example: Research Request

## Research Topic
(Enter the topic to research)

## Background and Purpose
(Why is this research needed?)

## What You Want to Know
- (Specific question 1)
- (Specific question 2)
- (Specific question 3)

## Deliverable Format
- [ ] Report (narrative format)
- [ ] Comparison table
- [ ] Bullet-point summary

## Target Audience
(Who will read this?)

## Expected Length
(Estimated page count or word count)

Template Example: Document Creation Request

## Document Title
(Title of the document to create)

## Purpose
(What this document will be used for)

## Content to Include
- (Element 1)
- (Element 2)
- (Element 3)

## Format and Tone
- Format: Report / Presentation outline / Proposal
- Tone: Formal / Casual
- Length: approximately X pages

## Target Audience
(Who is this document for?)

## Reference Information
(Any relevant URLs or information)

Security and Risk Management

Data That’s Okay to Share with AI — and Data That’s Not

Rule of thumb: “Would it be okay if this content were published on the internet?” If the answer is no, consult your manager before sharing it with the AI.

Always Have Humans Check AI Deliverables

AI deliverables may contain the following types of errors:

• Numerical errors: Statistics and financial figures may be inaccurate
• Outdated information: Content may not reflect the very latest developments
• Context gaps: Company-specific circumstances may not be properly considered

PR review is a quality management mechanism. The AI creates the deliverable, and a human verifies it before it’s considered “complete.” Make this part of your routine.

What Is Prompt Injection?

External documents may contain hidden instructions for the AI embedded within them.

Countermeasures:

• Don’t blindly copy-paste external documents into Issues
• If AI output seems unusual, review the input
• Add “Verify facts before using external data” to your CLAUDE.md

Comprehensive Exercise

Imagine your actual work and experience the full workflow.

Steps

1. Choose a work scenario

   • Pick one task from your work that you’d like to delegate to the AI agent
   • Example: “Create an industry trends report for next month’s department meeting”

2. Create an Issue

   • Use a template to create an Issue in your repository

3. Execute with Claude Code

   • Use Claude Code from the terminal to execute the Issue’s task

4. Review the PR

   • Review the deliverable content
   • Write feedback as comments

5. Reflect

   • Identify what went well and what you’d like to improve

Full Series Retrospective

What We Learned Across Five Sessions

My Action Plan

What I’ll Do Next Week (1 item)

What I’ll Do Within 1 Month (1–2 items)

What I Want to Share with My Team

Self-Study Resources

Today’s Key Takeaways