Skip to content
(01) // CODE

Get a useful review of code you didn't write

A prompt for reviewing inherited code that returns ranked, specific findings instead of a list of style opinions — with a rule that stops it inventing problems.

15 SEPT 2026 · Claude · ChatGPT · Gemini

THE PROMPT// CLAUDE / CHATGPT / GEMINI
You are reviewing code you did not write, in a codebase you cannot run.

Context: [WHAT THIS CODE DOES]. It runs [WHERE / HOW OFTEN]. The part that must
not break is [THE EXPENSIVE FAILURE].

Review the code below and report findings in this order:
1. Correctness — anything that produces a wrong result or loses data.
2. Failure behaviour — what happens on a timeout, a retry, a partial write, an empty input.
3. Security — injection, authorisation gaps, secrets in the wrong place.
4. Clarity — only where it would cause a future bug, not where it is merely unfashionable.

For each finding give: the specific line or function, what goes wrong, the concrete
input or sequence that triggers it, and the smallest change that fixes it.

Rules:
- Rank by consequence, not by how easy it is to describe.
- If you are unsure whether something is a bug, say so explicitly and say what you
  would need to check. Do not present a guess as a finding.
- Do not comment on formatting, naming or import order.
- If you find nothing serious, say that plainly rather than padding the list.

```
[PASTE CODE]
```
  • [WHAT THIS CODE DOES]

    One sentence of business purpose. Without it, the model reviews syntax rather than behaviour.

  • [WHERE / HOW OFTEN]

    "A queue worker, ~50k times a day" changes the answer completely versus "a one-off migration script".

  • [THE EXPENSIVE FAILURE]

    The outcome you cannot afford — double-charging, losing a record, exposing another tenant's data. This is what makes the ranking useful.

START A
PROJECT
EMAIL CHANNELHELLO@BEECODEX.COM
SYSTEMS ONLINE / TAKING NEW PROJECTS