#042Guide3 min

Use AI assistance to understand a fix

Ask focused questions about a finding, its proposed change and the tests it needs.

Hackaiz team · updated 21/11/2026

Illustration: two developers review a change before approving it.
Illustration: two developers review a change before approving it. Illustrative photo, not a screenshot.

01Start from the finding you want to understand

The Hackaiz Pilot can help explain a finding and the next action. Open the relevant finding before opening the assistance panel so that your question is attached to the item you are reviewing. The interface offers “Explain this finding”, “How do I fix it?” and “Is it proven?” on finding pages. These are useful starting points for different questions, not interchangeable confirmations that a proposed solution is correct.

Read the original observation first. Identify the affected file, rule or component and what the report says was observed. Then ask about the part you do not understand: why an access rule is too broad, why a secret belongs on the server, or what a dependency change might affect. Never paste live passwords, access tokens or personal records into a question to make it more detailed.

02Get the right level of explanation

Type a focused question and use “Send”. For example: “Explain why this change adds a server-side permission check instead of only hiding the button.” This asks for the security boundary behind a fix rather than another copy of the proposed code. For a database rule, ask which user should be able to read or change which rows, and which users should be refused.

After an answer, “Simpler” requests a plain-language explanation and “Technical details” requests a more technical one. Use them to move between the intended behavior and the implementation. A useful follow-up is: “Separate the report’s evidence from assumptions in your answer.” The assistant’s interpretation should be checked against the finding; clearer wording does not create new proof or mean an unperformed check has been executed.

03Ask for tests, not just replacement code

Ask the assistant to explain what changes, what stays unchanged and how a reviewer could test both. For an authorization fix, request a positive case where an allowed user succeeds and a negative case where another user is refused. For a secret exposure, ask why removing the value from a source file is not the same as invalidating the exposed credential. These are review questions to discuss with the person responsible for the system.

Compare the response with the actual proposal and your application’s expected behavior. If the assistant mentions a file, permission or test that you cannot find, ask it to identify the supporting evidence rather than accepting the detail as fact. The Pilot does not replace repository review, application tests or an authorized follow-up analysis. It also does not silently turn advice into an approved production change.

04Keep approval and verification separate

When a correction is available, inspect its preview and the finding before authorizing publication on GitHub. Fixes are draft pull requests reviewed and approved by a human; Hackaiz never merges them automatically. Use assistance to clarify a change you are considering, not as a reason to bypass the reviewer. A successful explanation is not the same event as a successful test.

AI assistance uses credits. If “Not enough credits for this action” appears, another question cannot complete until eligible credits are available. A temporary service error is different from a credit shortage. If you need a person, use “Talk to a human”; the request is handed to the team and the reply appears in Help. Include the specific point you cannot validate, without sending credentials or private customer data.

Frequently asked questions

Was this article helpful?

Related articles