@CONTRIBUTING.md
Before producing a commit, run both review agents in parallel (single message, two Agent tool calls):
rust-code-reviewer— correctness, safety, ownership/lifetimes, performancerust-api-ergonomics-reviewer— downstream-consumer perspective: happy-path friction, lints that fire in user crates, runtime footguns, doc drift, generated-code readability
The two are complementary (different lenses on the same diff) and produce largely non-overlapping findings. Address all Critical, High, and Medium findings from both. For Low / advisory findings, flag these to the user to decide.
For changes that touch only internal/test/bench code with no public-API surface, the ergonomics reviewer may be skipped.