Code review
How /review checks a change for vulnerabilities, bugs and best practice. Each review agent's model and instructions live on the Agents screen.
Review agents
How the agents that read the change are run.
Review agents at once
How many review agents run together.
Challenging findings
Every finding is argued against before you see it.
Challenge threshold, challenge each finding separately, check suggested fixes against
Which findings survive, and how suggested patches are checked.
Scanners
Dependency and secret scanning, run before the agents start.
Scanners, refresh the vulnerability database daily
osv-scanner and Betterleaks: where they come from, and how fresh their advisories are.
Defaults and config keys
| Setting | Default | config.toml |
|---|---|---|
| Review agents at once | 1 (1–4) | [review] parts_at_once |
| Challenge threshold | 8 of 10 | [review] threshold |
| Challenge each finding separately | Together | [review] challenge_each |
| Check suggested fixes against | Head commit (or Working tree) | [review] patch_base = head_commit, working_tree |
| Scanners | Koda's copies (or From PATH) | [review] scanners = managed, path |
| Refresh the vulnerability database daily | Daily | [review] refresh_daily |
With Review agents at once at 1, the review's agents run one after another in a fixed order. So one review never sends your provider more than one request at a time. Scanners and advisories shows the scanners' versions and advisory freshness, with Refresh now. The review always checks uncommitted changes against the files on disk. See Code review.