Submission

Deliverable

Submit the code you created with agent assistance, along with the prompts you used. Your submission might include multiple scripts, iterations, or bonus features — that's all great.

At minimum, your submission must have:

A working script from the Build Challenge (Exercise 4)
The sequence of prompts you used to create/refine it
One example of debugging or fixing a bug with agent help

Mission checklist

Generated code from a description
Edited existing code with agent help
Debugged an error using the agent
Built a complete working script

Exercise 4: Build Challenge

Put it all together by building something useful.

Pick a challenge

Option A: Data processor

> Create a Python script that:
> 1. Reads all CSV files in a folder
> 2. Filters rows where a specified column meets a condition
> 3. Outputs a merged CSV with a summary row at the bottom
> Make it usable from the command line with arguments.

Option B: File organizer

> Create a script that organizes files in a folder:
> 1. Groups files by extension into subfolders
> 2. Renames files with a consistent pattern (date_originalname)
> 3. Generates a manifest.txt listing all files

Option C: Simple API client

> Create a Python script that:
> 1. Fetches data from a public API (pick any: weather, quotes, etc.)
> 2. Parses the JSON response
> 3. Formats and displays the results nicely
> 4. Handles errors gracefully (no internet, bad response, etc.)

Option D: Your own idea

Have something you actually need? Build it! The agent can help with anything from data analysis scripts to automation tools.

The process

Start with a clear description of what you want
Let the agent generate a first version
Test it — find what doesn't work
Iterate: "The X part isn't working. Fix it."
Add features: "Now add Y capability."

This mirrors real-world development with AI assistance.

Exercise 3: Debugging with Agent Help

When code breaks, agents can help diagnose and fix issues.

Share the error

> I'm getting this error when I run my script:
> [paste the full error traceback]
> Here's the relevant code: [paste or point to file]
> What's wrong and how do I fix it?

Explain symptoms

> My script runs without errors but gives wrong results.
> Input: [example input]
> Expected: [what you expected]
> Actual: [what you got]
> Help me find the bug.

Ask for explanations

> Why does this error happen?
> I want to understand the root cause, not just fix it.

Practice

Find or create a script with a bug
Ask the agent to help you debug it
Notice how it reasons through the problem

Good debugging prompts:

Include the full error message
Provide context (what were you trying to do?)
Share relevant code
Describe what you expected vs what happened

Exercise 2: Editing Existing Code

Often you need to modify code that already exists — yours or someone else's.

Point to specific files

> Read src/utils.py and add a new function called format_date
> that converts timestamps to "Month Day, Year" format.

Request targeted changes

> In the process_data function, add error handling for
> when the input file doesn't exist.

Refactoring

> This function is 50 lines long. Refactor it into smaller,
> well-named helper functions.

Adding features

> The current script prints output. Modify it to also
> save results to a file. Add a --output flag.

Practice

Find a script you've written before (or use one from R038B)
Ask the agent to make a specific modification
Review the changes before accepting them

Key skill: Always review what the agent changed. Don't blindly accept edits to existing code.

Exercise 1: Generating Code from Scratch

Start by having the agent write code from your description.

Describe what you want

> Write a Python function that takes a list of numbers
> and returns the top 3 largest values.

Add constraints

> Write a Python function that takes a list of numbers
> and returns the top 3 largest values.
> Don't use any external libraries.
> Handle the case where the list has fewer than 3 items.

Request specific structure

> Create a Python script with:
> - A main() function as entry point
> - A helper function for parsing command-line arguments
> - Docstrings for each function
> - Example usage in comments

Iterate to improve

First attempt not quite right? Refine:

> That's close, but:
> - Use a dictionary instead of a list
> - Add type hints
> - Make the error messages more descriptive

Practice

Try generating:

A function that does something simple (sorting, filtering, formatting)
A script that does something useful (file processing, data conversion)
Something you've been meaning to write but haven't gotten to

Exercise 0: When to Use Agents for Writing

Prerequisites

Claude Code installed and working — complete R038A first
A folder to work in — create a test directory for experiments

The power of code generation

Reading is valuable, but writing is where agents really accelerate your work:

Generate boilerplate code instantly
Implement algorithms you know conceptually but not syntactically
Create test files, config files, scripts
Build tools you need but don't want to write from scratch

When agents excel at writing

✅ Great for:

Scripts and utilities (one-off tools)
Boilerplate and repetitive patterns
Prototypes and first drafts
Code you understand but is tedious to type
Learning new languages or frameworks

When to be careful

⚠️ Review carefully:

Security-sensitive code (auth, crypto, input validation)
Code that will run in production
Anything involving money or user data
Complex algorithms where correctness matters

The golden rule: Never deploy code you don't understand. Use the agent to write it, but take time to read and verify.

Your role shifts

With coding agents:

Before: You type code, fixing syntax errors and looking up APIs
After: You describe intent, review output, and iterate

You become the architect and reviewer. The agent handles implementation details.

Intro

Routes 38A and 38B covered setup and reading. Now comes the real power: writing code.

Instead of typing every character, you can describe what you want and let the agent generate it:

"Write a Python script that processes all the JSON files in this folder and outputs a summary CSV."

Then iterate:

"Add error handling for missing files." "Make it accept command-line arguments." "Add a progress bar for large folders."

This route teaches you to:

Generate code from descriptions
Edit and extend existing code
Debug errors with agent assistance
Build complete, working tools

The mindset shift: You're no longer just a coder — you're directing an AI collaborator. Your job is to clearly express intent, review output, and guide refinements.

Note: These skills work with any coding agent (Cursor, Copilot, Aider) — not just Claude Code.

Route 038C: Writing — Generate, Edit, and Debug

RouteID: 038C
Wall: AI-Assisted Coding (W11)
Grade: 5.10a
Routesetter: Adrian
Time: ~45-60 minutes
You'll need: Claude Code installed (R038A), a test folder for experiments