The ClaudeDevs account on X posted a set of updates for Claude Code’s /code-review tool, saying the command now supports multiple effort levels and that each level is “rewritten” with a different review strategy. The post also claims the feature is available in all Claude Code sessions after an update.
According to the thread, the new modes include low, medium, high, xhigh and /code-review ultra. Low effort runs a single pass over the diff and is described as fast enough to use before every push. Medium reads the changed code in context, runs multiple finder passes, and verifies findings before surfacing them. High and xhigh move the finders and verifiers into subagents with fresh context, while xhigh also sweeps for impacts outside the change itself.
A chart posted with the update, titled “More effort, more issues caught,” compares “Claude Opus 4.8 /code-review” with a “Leading competitor” on an open source, human-labeled code review dataset. The chart claims the Claude tool found “17%” of bugs at low effort, rising to “22%” at medium, “24%” at high and “25%” at xhigh, while the competitor ranged from “8%” to “12%” across the same levels. As with any vendor-run benchmark, those figures merit a grain of salt.
The most aggressive mode, /code-review ultra, is described as spawning “a fleet of reviewer agents” that independently reproduce each finding. The account claims it is used on every PR at Anthropic and says it aims to keep the same severe-issue coverage as high effort with fewer false positives.
Source: X