tests-shelllisted

# Backend tests: shell (bats / plain-bash harness) Detect the shape — shell projects use one of two, and they are not interchangeable: - **`*.bats` files, or `bats` in CI** → [bats-core]. `@test "name" { … }`, `run <cmd>` then assert on `$status` and `$output`, `setup`/`teardown` per test, `setup_file`/`teardown_file` per file. - **A plain-bash harness** (a `tests/` directory of `*.test.sh` plus a `lib.sh` of assertions, run by a `run.sh`) → follow the harness that is there. Read its assertion helpers first and use them; do not introduce bats beside it. ## What a shell test asserts A script's observable behavior is its **exit code**, its **stdout/stderr**, and the **files it touched**. Assert on the ones that matter and say which: - Exit code for a script whose job is to pass or fail (a guard, a check, a validator). - Output for a script whose job is to produce something — match a substring, not the whole stream, so unrelated wording changes don't fail the test. - Filesystem state for a script that writes: build the fixture, run, then assert on what is there. **Test both directions.** A guard tested only on what it blocks will pass while blocking everything; one tested only on what it allows will pass while blocking nothing. Every rule needs an allow case and a block case, and a new rule lands with both. ## Fixtures Build a throwaway tree per case rather than sharing mutable state between tests. Use the harness's own helper if it has one; otherwise `mktemp -d