What "never approves" means for an AI code review tool

Updated: 2026-08-10.

---

The most common question about an AI code review bot is the one you should ask first: does it approve my pull requests? The answer for Code Terrier is a permanent no. It is worth being precise about why, because the difference is not cosmetic.

What Code Terrier does and does not do

Code Terrier posts COMMENT reviews. It never posts APPROVE, never posts REQUEST_CHANGES, never merges, never pushes, and never commits. Its tools are read-only — it can read the diff and the changed files, and nothing else.

That is not a temporary posture. It is the product's design, in the same way "never auto-merges" is the design. The review agent grades the code, and a person keeps the authority to act on the grade.

Why that boundary matters

Approval is authority, and authority is a decision

When a bot approves a pull request, it is making a decision about your codebase — often the last decision before a merge. That decision has a cost when it is wrong, and the cost is paid by the people who trust the bot. A tool that can approve will eventually be relied on to approve, and reliance is exactly where a false-positive approval lands.

The reviewer should not grade its own work

The strongest reason is structural. An approval is the conclusion of a review. If the same model that found the findings also decides whether they are worth blocking on, the reviewer is grading its own work. That is a conflict of interest no prompt can fix. Code Terrier keeps the maker and the checker separate: the findings pass an independent verification stage before they are published, and the final call on merging stays with a person.

Quiet is a feature, not a bug

A bot that can approve needs to be right almost always. A bot that only comments can be right on the things it raises and silent on the rest — which is the honest posture for a tool that reviews diffs, not entire codebases. Code Terrier reports what it finds on the exact changed line, and says nothing on clean changes. "No issues found" is a correct answer, not a failure.

What this means for your review process

The loop is designed so humans keep the last word:

• trigger modes (auto, on-creation, manual) decide when reviews run;
• dispositions (Adopted, Skipped, Escalate) decide what happens to a finding;
• the merge button stays in the human's hand, always.

If you want a bot that closes the loop by itself — auto-approve, auto-merge, unattended — Code Terrier is not that tool. That is intentional. The teams that want a bot that reads the code and then waits for a person are the teams we built it for.

---

If you want a review loop that checks agent-written code and never takes over the merge decision, install Code Terrier and open a PR. It reviews in auto mode by default.