Code review for AI-generated code from Claude Code, Cursor and Copilot
Solemly reviews every pull request the same way whether a person or an AI dev tool wrote it: with the repository's related files, structure and instruction files as context, a security pass, and a severity on each finding. It spots AI-written pull requests and tracks how they fare, and each finding can be handed straight back to Claude Code, Cursor or Codex to fix.
Checked against the product on 2026-11-03.
Why AI-written code needs the repo, not just the diff
Code from Claude Code, Cursor, Copilot or Codex is usually plausible and well formatted, so its problems are rarely on the surface: a call that doesn't match its definition elsewhere in the repository, a convention the agent didn't know about, an authorization check it left out. A review that reads only the diff misses them.
Solemly reads the related files and the project structure around a change, and keeps a code index of each repository's functions, classes and imports (Python, JavaScript, TypeScript, Go and Java), so a pull request reviewed file by file is checked against the callers and definitions around each change. With cross-repo context on, it also sees the callers of a changed function in your organization's other repositories.
The same rules the agents follow
The CLAUDE.md, AGENTS.md, .cursorrules and copilot-instructions.md files that steer your coding agents are read as review rules, from the pull request's own commit. A rule the agent ignored becomes a finding. See AI code review that follows your CLAUDE.md and AGENTS.md.
Spot AI-written pull requests
Solemly spots pull requests written by AI and tracks how they fare, so your reports show how AI-written code is landing next to everything else: findings by severity, repository and author, trends over time, and a CSV export.
Hand the findings back to the agent
Under each finding there are Fix in Claude Code, Cursor and Codex links, and the review summary has Fix all. A link opens a Solemly page with the finding and a ready-made prompt, and your agent opens with it and fixes the problem on your machine. Solemly doesn't change any code itself.
With the MCP server connected, the agent can go further: list the findings on a pull request, fix them, push, request a new review and wait for the result. The Claude Code plugin's /solemly:review-loop repeats that until nothing HIGH or CRITICAL is left.
Hold the line
Turn on Block on high severity and make Solemly / Review a required check, and an agent's pull request can't merge while a CRITICAL or HIGH finding is open, however many it opens. See Block a merge on AI code review findings.
Questions
Does Solemly review pull requests opened by bots?
By default, yes: every pull request is reviewed. Review rules in Repository Settings or a .solemly.yaml can skip automatic reviews by author, branch, title or label, for example for dependency bots.
Which coding agents does it work with?
Findings carry Fix in Claude Code, Cursor and Codex links, and the MCP server connects the same three agents so they can read findings and request reviews. Instruction files for Claude Code, Codex, Cursor, Windsurf and GitHub Copilot are read as review rules.
Does Solemly change the code itself?
No. A fix link opens your agent with a ready-made prompt, and the agent works in your own environment. You review its changes like any other.
Can an agent fix the findings and get re-reviewed without me?
With the Claude Code plugin, /solemly:review-loop fixes the findings, asks before pushing, waits for the new review and repeats until nothing HIGH or CRITICAL is left.
In the docs
Related
Try it on your next pull request
Free for 14 days with every feature, no credit card. Install it on a repository and the next pull request gets reviewed.