Cyberjury

AI-assisted security review for code diffs and repositories.

• Diff review checks the changes in a Git range, with the repository as context, and anchors each finding to the changed line it concerns.
• Repository review checks a repository in focused units and groups the findings into issues. Code outside a diff is not covered by a diff review, so a clean diff review is no substitute for it.

Structural code analysis covers C, C++, Go, Java, JavaScript, Python, Rust, Solidity, TypeScript, and TSX. Files in other languages are still reviewed from their raw source.

Install

pip install cyberjury

Configure

All settings come from a .env file in the working directory, and an exported shell value overrides the file. Copy .env.example to .env and fill in what you use.

Review

cyberjury review diff origin/main...HEAD /path/to/app
cyberjury review repository /path/to/repo

A repository review of a Git work tree covers the tracked files and the untracked files Git does not ignore, but not the contents of Git submodules. A plain directory is covered completely. Re-running a repository review resumes open work. Use --fresh to start a new review session.

Output

A review stores its results in reviews/<review-id> under CYBERJURY_STATE_DIR. The findings of each attempt are in attempts/<attempt>/findings.json, grouped into issues of findings that one code change fixes.

Each stage, review unit, issue link judgment, and attempt produces a cyberjury.record/v1 JSON record with a status (ok, incomplete, failed, interrupted, or running) and, when it is failed or incomplete, an error with a code (config, input, busy, model, or internal). The review prints progress to stderr as records are written, and prints the stage records and then the attempt record to stdout when it ends.