CASOON Open Source
The refactoring compiles and Clippy is quiet, yet something is off. judge reads your Rust workspace, reports what the change left behind with the evidence for every finding, and tells CI whether it made things worse.
cargo install cargo-judge
You moved modules, split crates or reshaped the API. judge shows what the move left behind, beyond what the compiler and Clippy already report.
judge reads only the checked-out tree, no Git history or authorship. Code written by people and code written by agents get the same findings.
Save the state you accept as a baseline. Every later run reports what was resolved and what is new, and fails only on new findings that are backed by evidence.
Run judge once on a tree you accept and save its findings as a baseline. On every later tree, cargo judge compare sorts the findings into unchanged, resolved and introduced.
Only introduced findings that are backed by evidence and carry warning or failure severity fail the verdict and the exit code. Heuristic findings are listed, but never block a merge.
Fast tier or deep tier
The default Fast Tier needs no build step and reads the tree as written. Compiling in rust-analyzer itself, behind the deep Cargo feature, lets dead-code and explain --why-live work from real reference data instead of syntax-level guesses.
dead-code reports unused-pub-workspace — no reference from another workspace crate and no reachability from a recognised entry point, not proof.explain --why-live prints the shortest evidenced route from fn main, a test or an FFI boundary to the item, each edge classified as static, dynamic, macro, generated or unknown.ra_ap_* crates takes noticeably longer once; cargo judge inspect confirms deep: available afterward.cargo install cargo-judge --features deep
cargo judge inspect # "deep: available"
cargo judge dead-code
cargo judge explain my_crate::parser::parse --why-liveInstall it, run it in the workspace root, and turn the result into a baseline for CI. Crate boundaries in judge.toml, the output formats and the rust-analyzer deep tier are covered in the documentation.
cargo-judge from crates.io with the line above.cargo judge in the workspace root for the grouped triage report.cargo judge compare against it.cargo judge # triage report
cargo judge --save-baseline # accept this state
cargo judge compare .judge/baseline.jsonexplain-rule:** removed references to the internal, unpublished todo.md.network feature.judge.toml, the CI step and the deep tier are covered in the documentation.cargo judge compare reaches its verdict and exit code as a diagram.--all-features; links to private, feature-gated or removed items are now plain code spans.Structural and cross-file concerns that only become visible across a workspace — duplicated code, complexity concentration, public API surface and dependency hygiene. Anything the compiler or Clippy already tells you, judge does not repeat.
No. It analyzes only the current source tree, workspace structure and dependency graph, never commit history, authorship or telemetry, so human-written and agent-refactored code get the same findings.
Save a baseline from a tree you accept with --save-baseline, then run cargo judge compare against it in CI. Only new findings backed by evidence and warning or failure severity fail the verdict; heuristic findings are listed but never block a merge.
The default Fast Tier needs no build step. The Deep Tier, behind the deep Cargo feature, loads the workspace into rust-analyzer for dead-code and why-live analysis, at the cost of a noticeably longer build.
Not by default. Dependency checks against crates.io, such as phantom or yanked dependencies, are opt-in via --check-crates-io, and judge never runs cargo-audit itself.