# Verification
Repository retains only co-located benchmark/profiling workloads and support they use. No unit, integration, doctest or correctness-test suites. Do not add tests or test CI.
## Local checks
```sh
cargo fmt --check
cargo check --locked
cargo clippy --locked --all-targets --all-features -- -D warnings
git diff --check
```
These checks establish formatting and compilation, not runtime correctness. Review changed behavior and perform relevant manual checks with isolated settings and synthetic data. Do not require live credentials for ordinary checks.
## Release checks
No CI workflows. Release-profile compilation, dependency policy, build/proc-macro inventory and public Rust API compatibility run locally during [release verification](../release.md#2-verify-once). No automated test jobs or test-health artifacts.
The `magi_code` library exposes only `run_from_env()` and `run_from_env_with_exit_code()` for launching the CLI. Settings, context, sessions and provider data are internal implementation details, not supported library interfaces. New public exports require intended external use. Preserve runtime interfaces while removing test-only adapters.
## Performance measurements
[Profiling benchmarks](profiling-benchmarks.md) lists retained workloads and exact manual commands. Workloads use Rust's test runner but remain ignored, opt-in performance measurements. Their assertions validate measurement inputs; they do not provide correctness coverage.