Lab 2A — The tool loop: build the app with files, shell, tests and git

Start from: app/ at tag lab01-done, with the Lab 0 setup loaded (see Working in app/).

Goal: watch the agent loop (model → tool call → result → model …) do real work. Claude Code writes files, runs the tests, reads the failures, fixes them and uses git like a developer would. This is the same flow as the Claude Code quickstart, applied to the app you’ll use for the rest of the course.

Start an interactive session in app/:

claude

Claude Code asks before it edits files or runs commands. Read each request, then approve it. Choosing “Yes, and don’t ask again” for python -m pytest is fine. Watch the ⏺ lines: each one is a tool call.

1. Build the app (Write + Bash + test-fix loop)

Build a small Python pet-store order calculator in this folder. Use the standard library only, plus pytest for tests.
- catalog.csv with header sku,name,price_cents,stock and 5 synthetic products P1..P5 (prices in integer cents)
- inventory.py: load the catalog into a dict keyed by sku
- pricing.py: compute an order total in integer cents from (sku, qty) lines, with a tax rate in percent, rounding half up
- cli.py: `python cli.py total P1:2 P3:1` prints the total like $12.34
- tests/ with pytest tests for inventory and pricing, including a rounding edge case and a test that loads the real catalog.csv and checks every price and stock is a non-negative integer
Run the tests and fix any failures before you finish.

Observe: the files it creates (Write), the python -m pytest command it runs, and whether it reads a failure and edits code again. That retry is the loop. Count how many tool calls it took.

2. Use git as a tool

Ask these one at a time:

What files have I changed?

Commit the app with a descriptive conventional-commit message.

Create a branch called feature/low-stock. On it, add a `python cli.py low-stock` command that lists products with stock below 5, with a test. Run the tests and commit.

Show me the last 5 commits, then merge feature/low-stock into main.

Observe: Claude Code runs git status, git diff, git switch -c, git log and git merge for you, and asks permission first. Check the result yourself in a second terminal: git log --oneline --graph.

3. Debug a regression with git history

Exit Claude Code (/exit). Now you introduce a regression, the way a careless teammate would, with a misleading commit message:

(Get-Content catalog.csv) -replace '^P2,([^,]+),\d+,', 'P2,$1,12.99,' | Set-Content catalog.csv
git commit -am "docs: tidy catalog formatting"

sed -i.bak -E 's/^P2,([^,]+),[0-9]+,/P2,\1,12.99,/' catalog.csv && rm catalog.csv.bak
git commit -am "docs: tidy catalog formatting"

Start claude again and ask:

The tests are failing on main. Use the git history to find which commit broke them and explain why, then fix it in a new commit. Do not rewrite history.

Observe: does it run the tests first, then git log/git show/git diff (or even git bisect) to find the commit, instead of guessing? Does it ignore the misleading “docs:” message and trust the diff?

Record

• How many tool calls each step took and which tools (Write, Edit, Bash, Read, Grep …).
• Whether a test failure ever fed back into a fix without your prompting.
• One thing you would not have let it do without a prompt.

Checkpoint

python -m pytest -q
git status            # should be clean
git tag lab02a-done

Your app will differ from other learners’. That’s fine, as long as tests pass and catalog.csv, pricing.py, inventory.py, cli.py and tests/ exist, because later labs rely on those names.