Owning software · 6 min

Codebase audit checklist: what we check and why

Published
Reading time
6 min

Short answer: a codebase audit reads six areas of your software: the code, what it costs to run, who holds the keys, security and data, how it ships, and whether your usage numbers are true. For each one there are a few questions you can ask your developers this week, and a few answers that should worry you. This is the list we work from, with the red flags we look for.

This list is for founders and CTOs who own software today. Buying a company instead? Our SaaS due diligence checklist is for the weeks before the letter of intent.

How to use the list

Ask the questions in writing, and ask to see the evidence with each answer: a list, an invoice, a screen share. A slow or vague answer is an answer too. Most red flags aren't emergencies. They're costs you haven't priced yet.

We score each area on the same four steps:

  • Healthy: a new developer could work here safely next week.
  • Watch: known issues, nothing urgent, fix as you go.
  • Risk: will cause real cost or an incident within months if left.
  • Critical: can lose data, money, access or users soon. Fix first.

Code and architecture

What we check: how the code is structured, which parts are tested, what it depends on, and where a change is most likely to break something. We also check whether the data model still fits what the business does. When it doesn't, that's a real reason to rebuild.

Questions to ask:

  • Is all the code in repositories the company owns, including the infrastructure scripts?
  • Which parts would you rather not touch?
  • What runs automatically before a release?
  • Is the framework, the language or the database out of support, or close to it?

Red flags:

  • Code that only exists on a server, or sits in a contractor's account.
  • Billing, login or data import on the "rather not touch" list.
  • Nothing automated around payments or login.
  • A major upgrade that's overdue.

Running costs

What we check: every recurring charge, matched to the bank, and which ones grow with use. We read the invoices, not a summary.

Questions to ask:

  • What did the product cost to run last month, line by line?
  • Which bills grow with usage: AI models, SMS, email, storage, bandwidth?
  • Does each of those have a spending limit and an alert?
  • Is anything running on free credits or an old plan that's going to end?

Red flags:

  • Costs on someone's personal card that never reach the accounts.
  • A usage-priced service with no cap and no alert.

On 23 September 2026 we ran the check on GarageHub, the car maintenance app we build and run. All its infrastructure cost about €50 to 60 a month. The cost to watch was AI. There was a daily budget per user in the code, but no monthly limit or alert at the model providers. A bug, a runaway loop or an abused account would have shown up on the invoice, not on a dashboard. The fix is half a day to a day of work.

Keys and ownership

What we check: every account the business depends on, who controls it, and how it moves if that person leaves.

Questions to ask:

  • Whose name is on the domain, DNS, hosting, code repository, App Store, Google Play, payments, email sending, analytics and AI provider accounts?
  • Which developer account do the mobile apps sit in?
  • Where are the app signing keys kept?

Red flags:

  • An account on someone's personal email or card.
  • Apps in a personal developer account.
  • Signing keys that only exist on one person's laptop.
  • A list of accounts that takes more than a day to produce.

None of this is about code quality, and it's the area that hurts most when someone leaves. A tidy business answers it in minutes.

Security and data

What we check: where personal data lives, who can reach it, which secrets are in the code, and whether the backups can actually be restored.

Questions to ask:

  • Has anyone searched the code history for keys and passwords?
  • When was a backup last restored, and how long did it take?
  • Who has production access today, people and services both?
  • What happens to a user's data when they delete their account?

Red flags:

  • Live keys in the history, and no plan to change them.
  • Backups that exist but have never been restored.
  • Former contractors who still have access.

A backup nobody has restored is a hope, not a backup. GarageHub's are encrypted, kept off-site and restored automatically every week to prove they work.

Delivery

What we check: how code reaches production, how often, and how a bad release gets undone.

Questions to ask:

  • Could someone other than the usual person ship a small change today, from the written steps?
  • How often did it ship in the last three months?
  • If a release goes wrong, how do you roll it back?
  • If coding agents write some of the code, who reviews it before it ships?

Red flags:

  • Only one person can release, or the steps live in their head.
  • No releases in months.
  • No way back from a bad release other than fixing it live.

Product and usage

What we check: the main user flows, the errors, the usage numbers and what they really measure. We re-run the queries behind the numbers you make decisions on.

Questions to ask:

  • Where does each number on your dashboard come from, and can someone re-run it?
  • What breaks most often?
  • Who answers customers, and how much of their week does it take?

Red flags:

  • A figure nobody can reproduce.
  • No error tracking, so nobody knows what breaks.
  • Support time that isn't counted anywhere.

This is the one that caught us out. GarageHub's dashboard said 17% of new users came back on day 2. Re-running the query gave 31%. A "last seen" field added in late July was empty for older users, so every earlier signup counted as never returning. We'd been making roadmap calls against a number that was far too low. The fault was in the measurement, not the product.

What comes after the list

A checklist tells you where to look. A report tells you what to do. Ours ends with one of three verdicts: keep, fix or rebuild. Then comes a plan for the first 90 days, with the effort for each step. GarageHub's verdict was Keep, with 35 to 76 senior days of work in the plan.

If you'd rather we ran the list, that's our codebase audit, the Reality Check. It takes 3 to 5 working days and costs from €1,800, fixed before we start. If we carry on together within 60 days, the full fee comes off the work. You can read the whole GarageHub report first and see what you'd get.

For your CTO

What to have ready, and what a good report should show:

  • Access: read-only access to the code, the hosting and the bills, plus one call with whoever knows the history.
  • Tests: GarageHub runs more than 4,400 API unit tests in under a minute, plus 110 integration tests against a real database, covering payments and login.
  • CI: about 60 automated checks enforce authorization, data ownership and compatibility with app versions already in the stores.
  • Deletion: deleting an account removes photos, payment customer data and analytics profiles after a 14-day grace period.
  • Strengths: the report had 56 findings, 15 of them strengths. A good audit tells you what to leave alone, not only what to fix.
  • Limits: ask any auditor what they didn't check. Ours lists it: no penetration test (secrets were searched with targeted patterns), user flows read from the code rather than clicked through on devices, no App Store or Play Console vitals, and usage figures re-run once, on the day of the check.

Already have software?

A written verdict in 3 to 5 working days, from €1,800: keep, fix or rebuild, and what to do first. Credited if we carry on together within 60 days.