From 638605e4fc1c42e1117da4efadeebe9040801b5a Mon Sep 17 00:00:00 2001
From: Dan Morosanu <dan.morosanu@uipath.com>
Date: Tue, 14 Jul 2026 18:14:51 +0300
Subject: [PATCH] fix: detect Windows skill paths in telemetry

---
 src/coder_eval/criteria/skill_triggered.py | 18 +++++++++---------
 tests/test_early_stop.py                   |  8 ++++++++
 tests/test_skill_triggered.py              | 16 ++++++++++++++++
 3 files changed, 33 insertions(+), 9 deletions(-)

diff --git a/src/coder_eval/criteria/skill_triggered.py b/src/coder_eval/criteria/skill_triggered.py
index 06b90d0e..3d1aaf65 100644
--- a/src/coder_eval/criteria/skill_triggered.py
+++ b/src/coder_eval/criteria/skill_triggered.py
@@ -34,8 +34,11 @@
 _NO = "no"
 
 # Extracts the skill name between ``skills/<name>/`` path segments (the Codex
-# file-read signal). The leading class keeps a bare ``skills//`` from matching.
-_SKILL_PATH_RE = re.compile(r"skills/([A-Za-z0-9][A-Za-z0-9_-]*)/")
+# file-read signal). Accept both POSIX and Windows separators because telemetry
+# records the command exactly as the agent emitted it. One-or-more separators
+# also handles JSON-escaped commands that retain doubled backslashes. The
+# lookahead permits overlapping matches such as ``.../skills/skills/<name>/...``.
+_SKILL_PATH_RE = re.compile(r"(?=skills[\\/]+([A-Za-z0-9][A-Za-z0-9_-]*)[\\/]+)")
 
 
 def _engaged_skill_names(cmd: CommandTelemetry) -> set[str]:
@@ -46,9 +49,9 @@ def _engaged_skill_names(cmd: CommandTelemetry) -> set[str]:
 
     - the Claude ``Skill`` tool call's ``skill`` parameter, namespace-stripped
       via ``.split(":")[-1]`` (the SDK reports ``plugin:skill-name``); and
-    - every ``skills/<name>/`` path segment appearing in any string parameter
-      (the Codex / file-read signal — repo layout or the ``.agents/skills``
-      sandbox symlink).
+    - every ``skills/<name>/`` or ``skills\\<name>\\`` path segment appearing
+      in any string parameter (the Codex / file-read signal — repo layout or
+      the ``.agents/skills`` sandbox symlink).
 
     Returns the (possibly empty) set of engaged skill names for this command.
     """
@@ -77,10 +80,7 @@ def _engaged_skill(cmd: CommandTelemetry, skill_name: str) -> bool:
     (Bash ``parameters['command']``) or a file-path parameter. The trailing
     slash prevents prefix collisions (``uipath-agents`` vs ``uipath-agents-foo``).
     """
-    if cmd.tool_name == "Skill" and cmd.parameters.get("skill", "").split(":")[-1] == skill_name:
-        return True
-    needle = f"skills/{skill_name}/"
-    return any(isinstance(v, str) and needle in v for v in cmd.parameters.values())
+    return skill_name in _engaged_skill_names(cmd)
 
 
 @register_criterion
diff --git a/tests/test_early_stop.py b/tests/test_early_stop.py
index f4af3a85..33f6a694 100644
--- a/tests/test_early_stop.py
+++ b/tests/test_early_stop.py
@@ -242,6 +242,14 @@ def test_file_read_path_segment(self) -> None:
         got = _engaged_skill_names(_cmd("Read", {"file_path": "/repo/skills/date-teller/SKILL.md"}))
         assert got == {"date-teller"}
 
+    def test_windows_file_read_path_segment(self) -> None:
+        got = _engaged_skill_names(_cmd("Read", {"file_path": r"C:\repo\skills\date-teller\SKILL.md"}))
+        assert got == {"date-teller"}
+
+    def test_json_escaped_windows_path_segment(self) -> None:
+        got = _engaged_skill_names(_cmd("Bash", {"command": r"type C:\\repo\\skills\\date-teller\\SKILL.md"}))
+        assert got == {"date-teller"}
+
     def test_bash_command_path_segment(self) -> None:
         assert _engaged_skill_names(_cmd("Bash", {"command": "cat skills/foo/refs.md"})) == {"foo"}
 
diff --git a/tests/test_skill_triggered.py b/tests/test_skill_triggered.py
index 5dbe1ad7..ea1a8168 100644
--- a/tests/test_skill_triggered.py
+++ b/tests/test_skill_triggered.py
@@ -133,6 +133,22 @@ def test_codex_reads_skill_md_agents_symlink_tp(self) -> None:
         )
         assert result.score == 1.0 and result.observed_label == "yes"
 
+    def test_codex_reads_skill_md_windows_path_tp(self) -> None:
+        result = _check(
+            expected_skill="uipath-agents",
+            skill_name="uipath-agents",
+            commands=[self._sed(r"C:\sandbox\.agents\skills\uipath-agents\SKILL.md")],
+        )
+        assert result.score == 1.0 and result.observed_label == "yes"
+
+    def test_codex_reads_skill_md_json_escaped_windows_command_tp(self) -> None:
+        result = _check(
+            expected_skill="uipath-agents",
+            skill_name="uipath-agents",
+            commands=[self._sed(r"C:\\sandbox\\.agents\\skills\\uipath-agents\\SKILL.md")],
+        )
+        assert result.score == 1.0 and result.observed_label == "yes"
+
     def test_codex_reads_wrong_skill_not_counted(self) -> None:
         # Read uipath-rpa's SKILL.md, but criterion filters on uipath-agents -> observed="no".
         result = _check(