How it works
Security
LintCat reads your code to review it, and that is all the AI part can do. Everything it posts is checked first, and it never approves, merges or blocks anything.
What LintCat can access
- Only the repositories you pick when you install the GitHub App. The permissions it asks for, and why, are listed under Configuration.
- Your model provider sees the code it reviews. Reviews run on your organization’s own Anthropic or OpenAI key, under your agreement with that provider.
- Your key stays encrypted. Only the review service reads it back; the dashboard shows its last four characters and nothing more.
What it never does
- The AI reviewer can only read. It can look through your code but has no way to comment, commit, approve or merge. Posting the review is done by LintCat itself, after the checks below.
- It never blocks a merge. The AI PR Review check is advisory, and its conclusion is
neutralwhen there are findings. - It never force-pushes. Fixes are committed only when you turn that on, and only if nobody pushed while the review ran. Otherwise they arrive as suggestions.
Only findings worth reading
Before anything is posted, every finding has to pass these checks:
- It points at a line this pull request added or changed.
- The reviewer is confident in it.
- It doesn’t repeat another finding.
- It makes the cut of at most 10 per review, most important first.
Fixes you can trust
A suggested fix is offered only when the lines it replaces match the file at the pull request’s latest commit exactly, so it never lands on the wrong code. Fixes stay inside the changed code and small enough to review at a glance.
Prompt injection
Pull requests can contain text written to steer an AI. LintCat treats everything in the repository, including the title, description, diff and files, as code to review and never as instructions. And since the reviewer can only read, text like that has nothing it could make LintCat do.