You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
v0.6.2: Fix: on Windows, a zero time budget could still run one question; the README shows a real --score run
A time budget of 0 skips every question, on Windows too. When --scaffold runs your code beside each question, it stops once its time
budget is spent. The check was "more time than the budget has passed", and
Windows' clock ticks about every 15.6 ms, so with a budget of 0 a fast start
measured no time at all and still ran one question. The check is now
"at least the budget".
Changed
Skill 0.3.9. The README and the Skill say a suite that goes green
"can mean nothing", not that it does: a green suite may be hollow, and
scoring it is how you find out.
The README shows a real --score run on the bundled pricing example,
starts with the one free command to try, and lists the four problems qikly
surfaced in other libraries, with their upstream status. The picture is
drawn from a real run by tools/render_score_screenshot.py.
For contributors
The suite runs in a few minutes. Two test files scored the same project
before every test; each now scores it once. With pytest-xdist the suite
runs in parallel (pytest -q -n auto --dist loadgroup), as CI now does; the
tests that work in the project folder stay on one worker. pytest 9.2 with
pytest-xdist 3.8.0 ignores xdist_group marks, so tests/conftest.py
tags them itself until xdist releases its fix.