mirror of
https://github.com/dotnet/skills.git
synced 2026-09-20 09:49:54 +08:00
Abhitej John
efe972568b
Vally's `experiment run` harness (0.7) treats a non-zero setup command as a fatal trial error and drops all recorded events, unlike the .NET skill-validator which never inspects setup exit codes. Two evals hit this (or a related fixture bug) and produced zero verdicts in the cross-family matrix: - dotnet-msbuild/binlog-failure-analysis: the intentionally-broken solution makes `dotnet build -bl` exit non-zero (the binlog is still written). Guard the command with `|| exit 0`, which is portable across cmd.exe and POSIX sh (`true` is not a cmd builtin, and vally shells out via `cmd /c` on Windows). Also preserve any staged skill directory in the source-strip step so the skilled variant keeps its SKILL.md instead of aborting with "contains no SKILL.md". - dotnet-test/run-tests: the multi-TFM SDK 9 scenario pointed at a removed fixture (fixtures/multi-tfm/); repoint to the existing multi-tfm-mtp-sdk9 fixture, matching the skill-validator eval.yaml which already uses it. Scope is limited to the .vally.yaml variants; the sibling skill-validator eval.yaml files are already correct and need no change. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>