Miri: dynamic undefined-behavior detection for Rust

Resolve bundled resources from the activated skill root; run Cargo commands from the consumer's workspace. Follow the upstream Miri documentation matching the installed nightly, not assumed stable flag behavior.

Step 1: Define the checked execution and trust boundary

1. Identify the package, public API/unsafe contract, tests or binary, Cargo features, target, inputs and concurrency behavior. Read repository instructions, toolchain files, Cargo manifests/lockfile/configuration and existing Miri CI before changing them.
2. Distinguish UB detection, experimental aliasing/provenance policy, additional leak checking, and unsupported-operation diagnosis. Read references/research-scope.md when interpreting the supplied POPL26 page, soundness claims, research artifact, or tool limits.
3. Treat a clean run as evidence about selected executions under the recorded model/configuration, not proof of a sound library, exhaustive schedules, arbitrary native FFI, or future-compiler compatibility.
4. Review Cargo build scripts, proc macros and dependencies before executing them. Miri's default host isolation is deterministic API emulation, not a security sandbox. Keep real secrets and cryptographic key generation out of interpreted tests.

Gate: the intended behavior, actual execution coverage, tool/model assumptions and host-code trust are explicit.

Step 2: Establish a reproducible nightly