Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@tests/test_cli.py`:
- Line 2401: The test function test_check_reads_source_once currently declares
an unused parameter tmp_path which causes lint noise; update the function
signature in the test (def test_check_reads_source_once(...)) to remove the
tmp_path parameter so it no longer accepts or references tmp_path, and run the
tests/linter to confirm the unused-argument warning is resolved.
- Around line 2425-2433: In test_verify_dev_stdin_subprocess, add an assertion
on result.stdout to confirm the verify subprocess completed the expected
verification path (not just exit code); e.g. check result.stdout (or
result.stdout.lower()) contains a verification marker such as "verified",
"verification completed", or the tool's summary line (use result.stdout in the
assertion message for diagnostics) so the test verifies output content as well
as returncode.
- Around line 2388-2399: Rename the misnamed test_run_stdin to reflect that it
reads from a regular file (e.g., test_run_file) and update its docstring
accordingly; locate the test function named test_run_stdin (which calls cmd_run
on stdin_path and uses self.SIMPLE_PROGRAM/tmp_path) and change the name and
docstring so the test name and description match that it exercises file input
rather than /dev/stdin (do not modify cmd_run behavior in this change).

In `@vera/cli.py`:
- Around line 51-53: Create a small helper function (e.g., load_and_parse or
read_and_parse) that takes a path argument, constructs a Path, calls
read_text(encoding="utf-8"), and calls parse(source, file=str(p)), returning the
parsed tree (and optionally the source if callers need it); then replace the
duplicated patterns Path(path).read_text(...) + parse(...) in the CLI command
handlers with calls to this helper (update call sites at the locations using
Path and parse such as the blocks currently invoking Path(path) / read_text /
parse) to centralise reading/parsing and reduce duplication.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@tests/test_cli.py`:
- Around line 2370-2430: The stdin tests in TestStdinInput only use an
import-free SIMPLE_PROGRAM and never exercise module resolution or default
output-path logic (e.g., for compile), so add two tests: one variant of the
existing subprocess tests that supplies a program containing "import lib;" to
run/verify/check against /dev/stdin (to exercise module resolution), and one
test that runs "vera.cli" with the "compile /dev/stdin" command (capturing
stdout/stderr and returncode) to ensure compile-time default output path
behavior is correct; update or add tests alongside test_run_dev_stdin_subprocess
and test_check_reads_source_once so they pass the program string via subprocess
input and assert correct return codes and expected outputs.

In `@vera/cli.py`:
- Around line 49-60: The helper _load_and_parse currently returns
Path("/dev/stdin") which causes ModuleResolver(_root=p.parent) and cmd_compile
output logic to treat stdin as a filesystem path; change _load_and_parse so
after reading the source it detects stdin (e.g. path == "-" or path ==
"/dev/stdin" or p.is_fifo()) and in that case return a separate logical/base
Path for resolution/output (e.g. Path.cwd()/Path("stdin") or
Path.cwd()/Path("stdin.rs")) instead of the actual /dev/stdin Path; update
callers that pass p to ModuleResolver(_root=...) and cmd_compile output naming
to use this returned logical/base Path rather than the raw p so imports and
output filenames resolve relative to the caller's working directory rather than
/dev.

---

Outside diff comments:
In `@TESTING.md`:
- Around line 9-13: TESTING.md currently has inconsistent example counts
(headline shows 30 examples while other sections still state 29); update all
occurrences of the example and conformance program totals in TESTING.md so they
match the refreshed headline metrics (e.g., change any "29" example references
to "30" and verify the "65 programs" count is consistent), then run
scripts/check_doc_counts.py to validate that TESTING.md, CLAUDE.md, and related
docs pass the automated checks and adjust any remaining mismatches reported by
that script.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@vera/cli.py`:
- Line 80: The unpacking of the tuple returned by _load_and_parse(path)
currently binds p and source which are unused; update the unpacking in
vera/cli.py to use underscore-prefixed names (e.g., _p, _source, tree or _, _,
tree) so unused variables are clearly marked and suppress Ruff RUF059; make the
same change in the cmd_ast code path where _load_and_parse is unpacked around
the referenced location (change the unused p and source there to
underscore-prefixed names as well).

---

Duplicate comments:
In `@tests/test_cli.py`:
- Around line 2370-2455: Add a new test in TestStdinInput that exercises module
resolution when reading /dev/stdin: create a helper module file (e.g.,
"lib.vera") in a tmp_path, write a main source that imports lib(helper) and call
subprocess.run([sys.executable, "-m", "vera.cli", "check", "/dev/stdin"],
input=main_src, capture_output=True, text=True, cwd=tmp_path) and assert
returncode == 0; this ensures the _load_and_parse normalization to
Path.cwd()/"stdin.vera" and ModuleResolver import lookup are covered.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@vera/cli.py`:
- Line 52: The return type annotation on _load_and_parse currently uses
unnecessary forward-reference quotes; update the signature of _load_and_parse to
use a normal annotation tuple[Path, str, Tree[object]] (no surrounding quotes),
relying on the module-level import of Tree so the annotation is a real type
rather than a string.