PL-06 Sample v1.0 · updated 4 Jul 2026
Review a code change for bugs, risks and missing tests
A structured review of a diff: correctness first, then security and edge cases, then tests, with a confidence level on every finding.
- Task
- Code
- Level
- Advanced
- Placeholders
- 3
- Works with
- Reasoning, Chat, API, Local
Review the following {{language}} change. What it is meant to do: {{intent}}.
Report findings in this order, most serious first within each group:
1. Correctness: cases where the code does not do what is intended.
2. Security and data: unsafe input handling, secrets, permissions, data loss.
3. Edge cases: empty, very large, concurrent or unexpected inputs.
4. Tests: what is untested, with one concrete test case per gap.
For each finding give the line or function, what goes wrong, a suggested fix, and your confidence (high, medium or low). Do not comment on style unless it hides a bug. If you need to see other code to be sure, say which.
Change:
```
{{diff}}
```
Words in double braces are placeholders. Replace them before you send the prompt, here or in your tool.
Fill in the prompt
Filling in on the page needs JavaScript. Copy the prompt by selecting it, then replace each placeholder in your tool.
| Placeholder | What to put there | Example |
|---|---|---|
{{language}} | Language or framework. Include the version if it matters. | Python 3.12 |
{{intent}} | What the change should do. One or two sentences, as in a pull request description. | reject uploads larger than 10 MB with a clear error |
{{diff}} | The diff or code. A unified diff is best. Remove secrets and keys first. | (paste the diff here) |
Is it the right prompt?
Use it when
- You want a second reader before asking a colleague for review.
- The change touches input handling, money, permissions or data deletion.
Skip it when
- The code is proprietary and your tool is not approved for it.
- You plan to merge on the model's approval alone.
How to adapt it
- Your team has a review checklist
- Paste it as a fifth group and ask the model to tick each item.
- The diff is large
- Review one file at a time, then ask how the files interact.
Check the answer
Before you rely on the result, confirm that:
- Each finding names a line or function you can locate.
- Every high-confidence finding is something you can reproduce or explain.
- Suggested tests are concrete inputs with expected results.
Watch out for
- Security Never paste credentials, tokens or private keys. Remove them from the diff first.
- Accuracy Models miss bugs and invent some. A model review adds to human review and tests; it does not replace them.
Revision history
| Version | Date | Change |
|---|---|---|
| v1.0 | First sample version. |