Developer Adam Miller has released adamsreview, a plugin that layers multi-stage code review on top of Claude Code, according to Hacker News. The tool runs parallel sub-agent reviewers, validates their findings, and drives an automated fix loop that re-reviews its own work and reverts regressions before committing. The Show HN post claims it’s catching more real bugs than Claude Code’s built-in /review and /ultrareview commands, CodeRabbit, Greptile, and Codex’s built-in reviewer, with fewer false positives. The author flags those numbers as anecdotal (n=1), but the project pulled 169 points on Hacker News within hours.
What makes this launch interesting is the pipeline architecture. Most AI review tools are one-shot: feed in a diff, get back comments. adamsreview splits the work across six commands that hand artifacts to each other, so a review run from last week can still feed a fix run today.
The six commands
- /adamsreview:review: Fires up to seven parallel sub-agent lenses (correctness, security, UX, and more), then runs a dedup pass, a cheap-then-deep validation gate, and an optional Opus holistic pass. High-confidence fixes get pre-computed so the next steps can batch-accept them.
- /adamsreview:codex-review: A Codex CLI peer to the main review, drop-in compatible with everything downstream. Effort is tunable from low to xhigh.
- /adamsreview:add: Pastes external findings (a cloud /ultrareview, an Opus once-over, a teammate’s note) into the existing review artifact, deduped and re-validated.
- /adamsreview:walkthrough: Interactive driver for findings the auto-fixer would skip. Uses the harness’s question UI to walk through uncertain items one by one and posts a decisions log to the PR.
- /adamsreview:fix: The automated fix loop. Dispatches per-group sub-agents in parallel, re-reviews with Opus, reverts regressions, and commits the survivors. One combined commit by default, or –granular-commits for one per group.
- /adamsreview:promote: Human override to push a single finding into the auto-fix lane, bypassing the score threshold.
How it stacks up against alternatives
The pitch is mostly about depth and economics. CodeRabbit and Greptile are SaaS reviewers with their own billing. Claude Code’s /ultrareview charges against your Extra Usage pool. adamsreview runs against a regular Claude Code subscription (Max plan recommended), so heavier reviews don’t burn metered credits. The –ensemble flag layers a Codex CLI pass and a PR bot-comment scrape on top of the internal lenses when you want maximum coverage at higher token cost.
Availability and install
The plugin is free and installs through Claude Code’s plugin marketplace:
/plugin marketplace add adamjgmiller/adamsreview/plugin install adamsreview@adamsreview
Review artifacts persist under ~/.adams-reviews/<slug>/<branch>/, so the review, walkthrough, and fix phases can land days or weeks apart. The commands are also independent. You can skip straight from review to fix, or run fix against an existing artifact without ever running review in that session.
Caveats
The author is upfront that bug-catching claims are based on his own PRs, not a controlled benchmark. The –ensemble mode burns more tokens. And the Max plan recommendation suggests this isn’t a tool for casual users on lower Claude tiers.
For teams already living inside Claude Code, the pitch is straightforward: a more rigorous review pipeline that costs nothing extra beyond your existing subscription, with a fix loop that polices its own regressions. Full documentation and the artifact schema are available at the original source.