fix(reports): make skipped-task JUnit names platform-independent

_skipped_suite built the testcase name with str(Path(path).with_suffix("")),
which yields "tasks\opt" on Windows and "tasks/opt" on Linux. That made the
JUnit testcase identity depend on the OS that generated the report, so the same
logical run would split into two identities in CI history/flake tracking (and it
broke the Windows smoke job).

Normalize separators and parse with PurePosixPath so the emitted name is always
"/"-separated. Adds a regression test feeding a Windows-style path.

Also switch the action's model in the README/tutorial examples from Haiku to
Sonnet (docs only; the dogfood job stays on Haiku for per-PR cost).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: uipreliga

README.md

@@ -105,7 +105,7 @@ the job summary, and fails the step on any task/gate failure:
 - uses: UiPath/coder_eval@v0    # becomes @v1 once 1.0.0 ships; @vX.Y.Z pins exactly
   with:
     tasks: tests/tasks/**/*.yaml
-    model: claude-haiku-4-5-20251001
+    model: claude-sonnet-5
     anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}
 ```
 

docs/tutorials/02-ci-pipeline.md

@@ -172,7 +172,7 @@ job-summary + fail-on-failure into one step:
       - uses: UiPath/coder_eval@v0       # …then run the gate (pin @vX.Y.Z in production)
         with:
           tasks: tests/tasks/**/*.yaml
-          model: claude-haiku-4-5-20251001
+          model: claude-sonnet-5
           anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}
 ```
 

src/coder_eval/reports_junit.py

@@ -24,7 +24,7 @@
 import math
 import re
 import xml.etree.ElementTree as ET
-from pathlib import Path
+from pathlib import Path, PurePosixPath
 from typing import Any, Literal
 
 from .evaluation.judge_context import truncate
@@ -235,20 +235,32 @@ def _task_case(row: dict[str, Any], run_dir: Path) -> ET.Element:
     return case
 
 
+def _skipped_name(path: str) -> str:
+    """Stable testcase name for a skipped task: suffix-stripped, ``/``-separated.
+
+    Uses the whole path rather than just the stem, because two skipped tasks
+    sharing a basename (``suiteA/task.yaml``, ``suiteB/task.yaml``) would
+    otherwise collapse into one identity that some JUnit ingesters merge.
+
+    Separators are normalized to ``/`` and the path is parsed with
+    ``PurePosixPath`` so the emitted name does not depend on the OS that
+    generated the report — the same logical run must produce the same testcase
+    identity on Windows and Linux, or CI history/flake tracking splits in two.
+    """
+    return str(PurePosixPath(path.replace("\\", "/")).with_suffix(""))
+
+
 def _skipped_suite(summary: RunSummary) -> ET.Element | None:
     """Build the synthetic ``skipped`` testsuite from ``RunSummary.skipped_tasks``."""
     if not summary.skipped_tasks:
         return None
     suite = ET.Element("testsuite", {"name": "skipped"})
     cases: list[ET.Element] = []
     for entry in summary.skipped_tasks:
-        # Use the suffix-stripped PATH, not just the stem: two skipped tasks
-        # sharing a basename (suiteA/task.yaml, suiteB/task.yaml) would otherwise
-        # collapse into one identity that some JUnit ingesters merge.
         case = ET.SubElement(
             suite,
             "testcase",
-            {"name": _xml_safe(str(Path(entry.path).with_suffix(""))), "classname": "skipped"},
+            {"name": _xml_safe(_skipped_name(entry.path)), "classname": "skipped"},
         )
         ET.SubElement(case, "skipped", {"message": _xml_safe(entry.reason)})
         cases.append(case)

tests/test_reports_junit.py

@@ -302,9 +302,12 @@ def test_skipped_tasks_suite(write_run_json: Callable[..., Path], tmp_path: Path
     assert ts is not None
     skipped_cases = ts.findall("testcase")
     assert len(skipped_cases) == 2
-    # Suffix-stripped full path (not just the stem) so same-basename tasks stay distinct.
+    # Suffix-stripped full path (not just the stem) so same-basename tasks stay
+    # distinct. Always '/'-separated: the same logical run must yield the same
+    # testcase identity on Windows and Linux, or CI history splits in two.
     names = {c.get("name") for c in skipped_cases}
     assert names == {"tasks/broken", "tasks/opt"}
+    assert not any("\\" in n for n in names)
     assert all(c.find("skipped") is not None for c in skipped_cases)
 
 
@@ -580,3 +583,13 @@ def test_skipped_names_unique_for_same_stem(write_run_json: Callable[..., Path],
     ts = _find_testsuite(root, "skipped")
     names = {c.get("name") for c in ts.findall("testcase")}
     assert len(names) == 2, f"skipped testcase names collided: {names}"
+
+
+def test_skipped_name_is_platform_independent(write_run_json: Callable[..., Path], tmp_path: Path) -> None:
+    """A Windows-style path recorded in run.json must still emit a '/'-separated
+    name, so a report generated on Windows matches one generated on Linux."""
+    run_dir = tmp_path / "run"
+    write_run_json(run_dir, [], skipped=[(r"tasks\win\task.yaml", "skip: true")])
+    root = fromstring(generate_junit_xml(run_dir))
+    name = _find_testsuite(root, "skipped").find("testcase").get("name")
+    assert name == "tasks/win/task"