[lslusarczyk]

The rule tells developers and agents to keep them dumb, specific, and tied to behavior.
It loads when a file under tests/ or conftest.py is in context.

[PatKamin]

In overall, please, make it more human-readable

[lslusarczyk]

In overall, please, make it more human-readable

Done in the second commit. The first applies your comments. The second makes everything more human readable.

[PatKamin]

I would describe additionally:

• xfail rules - when it is allowed to use skip/xfail, always require an explanation message for skipping
• create a file naming convention (I would give up current numbering in functional tests, i.e.)
• enforce usage of id= in parametrized tests
• perhaps expect a coverage goal to be met, this would require having rules/instructions on splitting test cases into smoke/functional/nightly or sth like that

[lslusarczyk]

I would describe additionally:

* xfail rules - when it is allowed to use skip/xfail, always require an explanation message for skipping
* create a file naming convention (I would give up current numbering in functional tests, i.e.)
* enforce usage of `id=` in parametrized tests
* perhaps expect a coverage goal to be met, this would require having rules/instructions on splitting test cases into smoke/functional/nightly or sth like that

Applied two first points.

skip and xfail always take reason. Skip only when this machine cannot run the test. xfail names a known defect.

A new file is test_<behavior>.py. No new sequence number. Renaming the existing 012_... files is a separate change.

ids on every parametrize should not be a rule. True and "256" already read clearly. Pass ids when the value's repr is not the case: a tensor, a module, a long tuple.

A coverage percentage are not in this rule. The repo does not measure coverage yet.