Expand description
Code-family evaluators (T077–T079 — code family).
Public surface:
CodeExtractor+CodeExtractorStrategy— strategy object that lifts code from an assistant response (markdown fence / regex / LLM). Shared by every code evaluator so extraction logic lives in exactly one place.CargoCheckEvaluator/ClippyEvaluator— deterministic evaluators that shell out tocargo check/cargo clippyin a tempdir.llm_judge::CodeLlmJudgeEvaluator— judge-backed evaluator using thecode_llm_judge_v0template.
SandboxedExecutionEvaluator (T080–T083, behind evaluator-sandbox)
lives in the sandbox submodule. The module compiles unconditionally
but its implementation forks per-platform: Unix uses POSIX rlimits
(module-scoped #![allow(unsafe_code)] per FR-049) and Windows returns
crate::EvaluatorError::UnsupportedPlatform at evaluation time.
Re-exports§
pub use cargo_check::CargoCheckEvaluator;pub use clippy::ClippyEvaluator;pub use extractor::CodeExtractor;pub use extractor::CodeExtractorStrategy;pub use sandbox::SandboxLimits;pub use sandbox::SandboxOutcome;pub use sandbox::SandboxRunner;pub use sandbox::SandboxedExecutionEvaluator;pub use sandbox::ShellRunner;pub use sandbox::run_sandboxed;
Modules§
- cargo_
check - Deterministic
cargo checkevaluator (T077 — cargo-check portion). - clippy
- Deterministic
cargo clippyevaluator (T077 — clippy portion). - extractor
- Code extraction strategies (T078).
- llm_
judge - Judge-backed code quality evaluator (T079).
- sandbox
- Sandboxed execution evaluator (T080–T083, behind
evaluator-sandbox).