#046Guide4 min

Verify that a vulnerability is gone

Separate a new scan from replay evidence, passing tests, and proof of the corrected deployment.

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.

01Decide what would demonstrate the repair

Verification starts with the original finding, not with the overall appearance of a new report. Identify the request or code path that demonstrated the weakness and the expected protected behavior. If another account could read a private record, the useful question is whether that same unauthorized request is now denied while the owner still has access. A changed heading, a hidden button, or an empty report does not answer that question on its own. Preserve the original reproduction details without copying sensitive data into shared notes.

Also identify the version you intend to verify. A fix can exist in a branch while the deployed application still runs older code. Hackaiz distinguishes a prepared change, a deployed but unverified change, a verified repair, a regression, and an inconclusive outcome. These distinctions concern what the evidence establishes. They are not grades for the developer and should not be replaced with a simple “done” because a change has been merged.

02Use the correction lifecycle

Open Fixes and use Applied to find changes that still need attention after application. The full status Applied · verify means the change has been applied but verification remains outstanding. Verified is a separate state. Open Change details and validation to inspect the available context and use the linked finding to compare the correction with the original issue. The quick filters help locate work; they do not perform the verification merely because you select them.

For GitHub proposals, human approval remains part of the process. Hackaiz prepares a draft pull request after approval for publication, and does not merge it automatically. Even a human merge is not equivalent to proof that the deployed application contains the merge commit. Keep review, merge, deployment, and security verification separate when reporting progress to colleagues. This prevents an applied change from being counted as a demonstrated repair before the relevant environment is tested.

03Look for three complementary pieces of evidence

The correction classifier requires the original test to replay as resolved, the corrected commit to match deployment, and continuous-integration tests to pass before it classifies the repair as fixed and verified. Continuous integration is the project’s automated test process. It provides regression evidence alongside the security scenario; it is not interchangeable with that scenario. Deployment evidence links a repository revision to the running system rather than assuming that the latest branch is already live.

A replay that still finds the weakness is treated as a regression. A failed automated test also prevents a successful verification, even when the specific replay appears resolved. If tests are waiting, unavailable, or absent, the classifier does not turn that gap into success. An inconclusive replay remains inconclusive. Read these outcomes literally: unavailable evidence calls for investigation, not a claim that the application is protected or that the original finding was necessarily mistaken.

04Retest within the authorized scope

When you run additional website or API analysis, prove domain control first using _hackaiz in DNS, the HTTPS verification file at /.well-known/hackaiz-verification.txt, or the project AI creating that file. Use the same intended environment and comparable test permissions. Testing a development hostname cannot establish what a production hostname serves. A timeout, redirect, or missing baseline can stop a scenario from establishing anything useful; do not relabel those outcomes as a successful block.

A rescan alone never verifies a correction. It can help discover whether a finding is still reported, but formal correction verification needs the linked replay, tests, and deployment evidence. Analyses and verification work consume credits; deselected checks do not run and are not billed. Record precisely which scenario was resolved and which areas were not tested. Even a verified repair describes that repair and its tested version, not an exhaustive security guarantee for the entire application.

Frequently asked questions

Was this article helpful?

Related articles