Reviews

AICoder gates every change through a review pipeline before you ship. When a coding session exits cleanly, the reviews run automatically — but you can also run them on demand from the task drawer.

The built-in checks

• DRY — flags repeated or copy-pasted code.
• Docs — flags missing or stale documentation.
• Security — a dedicated, validated security pass with a medium-severity gate.
• Correctness — domain-specific sanity checks.

Reviews run on Haiku to keep them cheap — see Model routing.

Reading findings

Each review produces a list of findings. A finding has:

• Severity — error, warning, or note.
• Title and detail — what's wrong and why.
• Location — the file and line it points at.

Findings open in a full-screen viewer rather than expanding inside a card: the list of findings sits on the left (worst first, dismissed ones last) and the selected finding is shown in full on the right. Open it from the queue card's findings line, from a finished check in the task drawer, or from a finding in the diff sidebar.

• ↑ / ↓ move between findings, Esc closes the viewer.
• The card itself only shows the count and severity mix, so the queue stays scannable.

Acting on findings

For each finding you can:

• Dismiss it — mark it "no change required" (a false positive won't block shipping).
• Fix with AI — re-run the agent with the outstanding findings injected into its prompt so it addresses them, then reviews run again.
• View the diff — see the findings alongside the actual code changes.

Approved vs needs work

• Green (approved) — no outstanding findings; the work is ready to ship.
• Needs work — findings remain; use Fix with AI or dismiss them, then re-review.

Custom checks

Define your own review checks in Settings — for example a project convention or a security rule specific to your codebase. Custom checks run alongside the built-in ones.