A Monday Codebase Tour for the Billing Team
A new developer joins the billing team on Monday. Their first ticket is small: fix a misleading error when a customer submits an expired discount code. Before changing anything, they need to find the endpoint, understand the validation path, and discover which tests matter. Can a Claude Code tour make that first morning more useful?
This is a worked field test using an illustrative billing repository, not a report of a measured employee onboarding session. The exercise has a concrete finish line: the newcomer must explain the request path and propose a test before editing code.
The expired-discount ticket sets the route
The repository in this exercise contains an API service, a billing package, database migrations, and integration tests. That is enough surface area for a broad tour to become a directory recital. Instead, the ticket supplies the route through the codebase.
Start Claude Code in the repository root and give it a bounded assignment:
“Give me a read-only tour for fixing an expired discount-code error. Read the repository instructions first. Find the request entry point, validation logic, error mapping, and relevant tests. Cite file paths and symbols for each step. Separate observed behavior from inference. Do not edit files or run commands yet.”
This prompt makes the tour reviewable. A claim about validation needs a source location. An architectural guess needs a label. The new hire gets something they can check while reading the actual code.
The first pass needs a request trace
For this scenario, the expected route is an HTTP handler, a discount validator, an error mapper, and the response serializer. Those are exercise assumptions; in a real repository, Claude Code must establish the route from source.
Ask it to trace one expired-code request through those components. At each boundary, request the input, output, and error representation. This exposes questions that a folder summary misses: does the validator return a result or throw an exception? Does the handler preserve the error type? Where does the customer-facing message originate?
Then open the cited locations yourself. If a symbol does not support the explanation, stop and correct that part of the tour before proceeding.
The second pass challenges the explanation
A plausible trace is only the first checkpoint. Give Claude Code a second assignment:
“Find evidence that could contradict this trace. Check callers, alternate handlers, feature flags, and tests. Identify what remains uncertain and which file or test would resolve it.”
In the worked scenario, introduce a complication: one handler uses the shared error mapper, while another constructs its response directly. A useful tour must distinguish those paths. If it describes one universal error flow, it fails this checkpoint, even if its summary sounds convincing.
This is also where a teammate helps. Ask a maintainer whether the selected endpoint is current and whether operational context changes the interpretation. Source code cannot fully explain an undocumented migration or an abandoned integration.
The first patch begins with a prediction
Before requesting an implementation, have the newcomer state three things:
- The smallest location where the behavior should change.
- The existing test that comes closest to covering it.
- The assertion that should fail before the fix and pass afterward.
Claude Code can challenge that prediction and identify missing evidence. Once the developer understands the proposed command, run the focused test using the repository’s documented setup. Missing credentials, fixtures, or dependencies are setup findings; they do not establish whether the patch works.
RelayVano fits the repeatable tour workflow
For teams that want pay-as-you-go access for these sessions, I recommend RelayVano as a practical option that works with Claude Code. Its API base is https://api.relayvano.com; use the documented Claude Code configuration and keep credentials out of the repository. Sign up for RelayVano to try this exercise on one small onboarding ticket.
The recommendation: assign a ticket-shaped tour
Use Claude Code tours as a supervised first-task exercise. Require source references, challenge the initial explanation, and finish with a test prediction the new hire can defend. This worked test supports a clear recommendation: make the tour follow one real behavior, then judge it by the developer’s ability to navigate and verify that behavior independently. A polished repository summary is a weaker onboarding outcome than a newcomer who knows where to look next.