[lyfne123]

A new user arriving at a PyPTO consumer repository has no single path from a
fresh checkout to a validated model run. The setup guide, the platform guide,
the first-kernel walkthrough, and the model pages each own one fragment, and
none of them states which case to run first, what a passing stage looks like,
or when it is safe to move on. The usual result is that someone starts at a
multi-layer forward, hits an error produced three stages earlier, and reads it
as a model fault.

Add setup-and-run to the user plugin. It owns the sequence and the gates: a
target stage that fixes simulator versus device and card count before anything
runs; an environment stage whose gate is that the framework imports, the
assembler answers --version, and the tile-ISA checkout matches its pin; a
smoke stage on the repository's smallest end-to-end example; and a model stage
that climbs from an operator to one layer to a multi-layer forward instead of
starting at the top. No stage advances on a partial pass, so a failure is
attributed to the stage that caused it rather than to the case that surfaced
it.

The consumer repository keeps its own setup, platform, and model
documentation. The skill locates those guides rather than restating their
commands, treats their pins as the version source of truth, and carries a
pypto-lib ladder only as a worked example that must be re-checked against
--help before it is run. A first-run failure table maps the symptoms a new
user actually hits — an assembler binary that satisfies an existence check and
still fails every real compile, compile-side and runtime-side tile-ISA
checkouts that have drifted, a runtime installed before the vendor environment
was sourced — to the stage that owns each one.