Plan Before You Edit in Claude Code: Expand Features with plan Mode

01Define the change scope first, even for a small-looking edit

The current summarize.py reads expenses.csv and prints monthly amount totals. This time, we will extend it without removing the existing behavior so that category totals for each month are shown only when the `--by-category` option is provided. Even when adding just one feature, letting AI edit the file immediately can leave details such as output format, whether existing behavior is preserved, and sort order to the AI. So first, fix what must stay the same and what alone should change.

The key point here is the criterion for preventing regression, not the new feature itself. Even after adding the option, the existing output of `python summarize.py expenses.csv` must not change. You need to verify the old execution result as well as the result of the new option.

02Start in plan mode so files are not edited yet

Claude Code can be started in plan mode with `claude --permission-mode plan`. This mode is used for read-focused analysis of the code and task scope and for reviewing a plan. For beginners, it is easier to understand the change by getting a plan first rather than immediately widening edit permissions while existing code is present.

cd vibe-expenses
claude --permission-mode plan

After the session starts, first ask it to read the current summarize.py and expenses.csv and explain what should change. In this request, explicitly say not to edit yet. If the plan is not satisfactory, you can revise the requirements before any file is touched.

Read summarize.py and explain its current behavior first.
Then write only a change plan for adding the `--by-category` option. Do not edit any files yet.
The default run `python summarize.py expenses.csv` must keep exactly the same output.
The option run `python summarize.py expenses.csv --by-category` should add category totals for each month.
Use only the Python standard library and do not modify expenses.csv.
For 2026-09, the expected category totals are food 20500, transport 2900, supplies 3200.

03Look for missing requirements before judging the implementation approach

When you receive a plan, first check whether every requirement is reflected rather than whether the proposed code sounds elegant. In particular, verify that the existing output is preserved, what happens when the option is absent, the category sort order, handling of an invalid option, and whether any external package is added. If the plan is vague, ask one more question before implementation.

[Editorial example · not an actual response]
1. Keep the current CSV reading and monthly total calculation structure.
2. Add a data structure that also accumulates category totals by month.
3. Check whether the second argument is `--by-category` to decide whether the option is enabled.
4. In the default run, print only the existing monthly totals.
5. In the option run, print sorted category names and totals on the lines after each monthly total.
6. After implementation, run both the default and option cases and compare them with the expected values.

• Does the output of the existing default run remain unchanged?
• Are category results added only when the option is present?
• Does the plan avoid requiring any new external package installation?
• Does it avoid modifying expenses.csv or changing it to a new format?
• Does the plan include the commands needed to compare results with the expected values?

The plan above is an editorial example for explanation, not an actual Claude Code response. In a real session, the model may suggest a different implementation order, so do not memorize the wording of the plan as the answer. Check whether it matches your requirements.

04Implement after reviewing the plan, then inspect the completed code

After confirming that the plan matches the requirements, proceed with implementation. In Claude Code, actual file edits and command execution may require approval depending on the environment and permission settings. The code below is the completed example used to verify the results in this article. It was not copied from an actual Claude Code response; it was written to match the requirements in this article and run separately in the Python execution environment in GPT chat.

import csv
import sys
from collections import defaultdict

def read_totals(path):
    monthly = defaultdict(int)
    by_category = defaultdict(lambda: defaultdict(int))

    with open(path, newline="", encoding="utf-8") as f:
        reader = csv.DictReader(f)
        for row in reader:
            month = row["date"][:7]
            amount = int(row["amount"])
            monthly[month] += amount
            by_category[month][row["category"]] += amount

    return monthly, by_category

def main():
    if len(sys.argv) not in (2, 3):
        print("Usage: python summarize.py expenses.csv [--by-category]")
        raise SystemExit(2)

    if len(sys.argv) == 3 and sys.argv[2] != "--by-category":
        print("Usage: python summarize.py expenses.csv [--by-category]")
        raise SystemExit(2)

    monthly, by_category = read_totals(sys.argv[1])
    show_category = len(sys.argv) == 3

    for month in sorted(monthly):
        print(f"{month} = {monthly[month]}")
        if show_category:
            for category in sorted(by_category[month]):
                print(f"  {category} = {by_category[month][category]}")

if __name__ == "__main__":
    main()

This code uses only the csv module and collections.defaultdict. In the default run it prints only monthly totals, and only when `--by-category` is exactly the second argument does it print sorted category totals under each monthly total. If another option is provided or the number of arguments is invalid, it prints the usage message and exits.

05Recheck the existing run as well as the new feature

After adding the feature, verify both executions. First run without the option to confirm that the existing result is preserved, then add `--by-category` to check the new feature. The Python code in this article was actually run with the Python tool in GPT chat and matched the expected output below exactly.

python summarize.py expenses.csv
python summarize.py expenses.csv --by-category

Expected output for default run:
2026-09 = 26600
2026-10 = 9800

Expected output for --by-category:
2026-09 = 26600
  food = 20500
  supplies = 3200
  transport = 2900
2026-10 = 9800
  food = 9800

For month 9, categories are printed in alphabetical order: food, supplies, transport. The totals themselves all match the reference values defined in the first article of the series: food 20500, transport 2900, supplies 3200. Month 10 has only one category, food, so 9800 is printed.

06A short checklist before approving implementation

The purpose of plan mode is not to receive a long plan from AI, but to create decision points for a person before edits are made. If the following questions are answered clearly, approve implementation; if even one is vague, revise the plan first.

In the next article, we will record the state before and after a change like this with git. Even though Claude Code has checkpoints, git is a separate version-control mechanism. Learning to read the changed files and diff yourself makes it clearer what the AI actually changed.

What to check yourself

Actually ran the virtual expenses.csv and the summarize.py in this article as files in the Python execution environment of the GPT chat · Claude Code was not actually run

• Check that the default run prints exactly 2026-09 = 26600 and 2026-10 = 9800
• Check that the `--by-category` run prints exactly 2026-09 food = 20500, supplies = 3200, transport = 2900
• Check that the `--by-category` run prints 2026-10 food = 9800
• Check that both runs complete with exit code 0
• Check that the code uses only the Python standard library and does not modify expenses.csv
• Check that the Claude Code plan response example is labeled '[Editorial example · not an actual response]'