Add standalone runtime test repro skills

.github/skills/standalone-test-repro-cmd/SKILL.md

@@ -0,0 +1,66 @@
+---
+name: standalone-test-repro-cmd
+description: >
+  Resolve a CoreCLR runtime test from a generated .cmd or .sh wrapper path for
+  the standalone-test-repro workflow. Use when the requested test is identified
+  by a CI or Helix wrapper path, with or without a named test scenario.
+  Continue with the parent standalone-test-repro skill after resolution.
+---
+
+# Resolve a Runtime Test Wrapper
+
+This is the wrapper-path resolution sub-skill for
+[`../standalone-test-repro/SKILL.md`](../standalone-test-repro/SKILL.md).
+Resolve the exact source test, owning project, and effective invocation, then
+continue with the parent's shared extraction steps.
+
+## Input
+
+Accept a generated `.cmd` or `.sh` path copied from a CI failure, for example:
+
+```text
+JIT\Directed\ConvertToInt\checked\ConvertToInt.cmd
+```
+
+The path may be absolute, relative to a Helix work item or runtime test output
+root, and may use either directory separator. The surrounding request may also
+name a scenario such as `jitstress1`; preserve it separately from the path for
+the parent workflow.
+
+## Resolution
+
+1. Normalize the identifier:
+   - Trim quotes, Markdown formatting, leading test-result labels, and trailing
+     punctuation.
+   - Remove command-line arguments while retaining the path through `.cmd` or
+     `.sh`.
+   - Convert separators only for host filesystem operations.
+   - Do not treat a separately supplied scenario as part of the wrapper path.
+2. If the wrapper exists locally, read it and record:
+   - The managed assembly or executable
+   - `corerun` arguments, including `-p` and `-e`
+   - Test arguments and expected exit code
+   - Environment variables
+   - Working-directory assumptions
+   - Copied data or native dependencies
+3. Map an artifact or Helix path back to `src/tests/`:
+   - Strip prefixes through the runtime test output root.
+   - Remove configuration-only artifact path segments.
+   - Search the remaining relative directory for the wrapper stem, likely
+     `.csproj`, source file, and matching `AssemblyName`.
+4. If the wrapper is unavailable, use the path and stem to search under
+   `src/tests/`; derive invocation settings from the project.
+5. Inspect the owning project, applicable imports, and its explicit or evaluated
+   `Compile` items.
+6. Confirm the assembly contains only the requested runnable test. If it
+   contains multiple independently runnable tests, stop and ask for a fully
+   qualified method name.
+
+Do not assume a wrapper only runs `corerun Test.dll`. Its host options,
+environment, arguments, setup, and working directory may be required to
+reproduce the failure. When available, the generated wrapper is the ground
+truth for the effective invocation.
+
+Resolution is complete only when the exact source method, owning project, and
+effective invocation are known. Return any requested scenario unchanged for
+canonicalization by the `test-scenario-env` skill.

.github/skills/standalone-test-repro-method/SKILL.md

@@ -0,0 +1,65 @@
+---
+name: standalone-test-repro-method
+description: >
+  Resolve a CoreCLR runtime test from a fully qualified method name for the
+  standalone-test-repro workflow. Use for xUnit display names and methods in
+  merged or multi-test assemblies, with or without a named test scenario.
+  Continue with the parent standalone-test-repro skill after resolution.
+---
+
+# Resolve a Runtime Test Method
+
+This is the method-name resolution sub-skill for
+[`../standalone-test-repro/SKILL.md`](../standalone-test-repro/SKILL.md).
+Resolve the exact source test, owning project, and test invocation, then
+continue with the parent's shared extraction steps.
+
+## Input
+
+Accept a fully qualified method name copied from a CI failure, for example:
+
+```text
+Namespace.TypeName.MethodName
+```
+
+It may include an assembly prefix, xUnit display-name arguments, quotes,
+parentheses, nested-type notation, or surrounding CI log text. The surrounding
+request may also name a scenario such as `jitstress1`; preserve it separately
+from the method identifier for the parent workflow.
+
+## Resolution
+
+1. Normalize the identifier:
+   - Trim quotes, Markdown formatting, leading result labels, and trailing
+     punctuation.
+   - Preserve xUnit argument text for theory-case resolution.
+   - Remove an assembly prefix only after verifying it is an assembly name.
+   - Never reduce the identifier to only the final method segment.
+   - Do not treat a separately supplied scenario as part of the method name.
+2. Parse the longest plausible namespace/type prefix and method name. Account
+   for nested types (`+` in reflection names and `.` in C# source) and generic
+   arity suffixes.
+3. Search C# sources under `src/tests/` for the exact method declaration and
+   confirm its containing type and namespace. Prefer symbol information when
+   available; otherwise combine targeted searches for method, type, and
+   namespace.
+4. Find the owning `.csproj` by its explicit or evaluated `Compile` items.
+   Account for wildcard includes and merged projects such as JIT regression
+   assemblies.
+5. Inspect the project and source for:
+   - `[Fact]`, `[Theory]`, custom attributes, or `TestEntryPoint`
+   - Static or instance invocation and constructor requirements
+   - Fixtures, setup, cleanup, and disposal
+   - Async and generic return or argument handling
+   - Conditional compilation and platform guards
+   - Test data and the specific failing theory row
+6. If display-name arguments identify a theory row, reproduce exactly that
+   row. If not, generate invocations for all locally resolvable rows and label
+   them. Use the `ask_user` tool when runtime-discovered data or ambiguity
+   prevents a faithful choice.
+7. If multiple source methods still match, present the concrete candidates with
+   `ask_user`; do not guess.
+
+Resolution is complete only when the exact source method, owning project, and
+required invocation semantics are known. Return any requested scenario
+unchanged for canonicalization by the `test-scenario-env` skill.