Fix HTML report template guidance, numeric metric aggregation, and output path handling

openagent_eval/reports/html.py

@@ -6,7 +6,7 @@
 
 from __future__ import annotations
 
-from datetime import datetime, timezone
+from datetime import UTC, datetime
 from pathlib import Path
 from typing import Any
 
@@ -58,7 +58,8 @@ def _load_template(self) -> str:
 
         raise FileNotFoundError(
             f"HTML template not found at {builtin_path}. "
-            "Install jinja2 and ensure the template file exists."
+            "Ensure the package is installed correctly and includes "
+            "openagent_eval/reports/templates/report.html, or provide a valid custom template path."
         )
 
     def _render_template(self, context: dict[str, Any]) -> str:
@@ -92,12 +93,17 @@ def generate(self, report: Any) -> str:
         result = report.result
         summary = report.summary
         config = report.config
-        now = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S UTC")
+        now = datetime.now(UTC).strftime("%Y-%m-%d %H:%M:%S UTC")
 
         # Compute overall score
         metrics = summary.get("metrics_summary", {})
+        numeric_metric_values = [
+            value for value in metrics.values()
+            if isinstance(value, (int, float)) and not isinstance(value, bool)
+        ]
         overall_score = (
-            sum(metrics.values()) / len(metrics) if metrics else None
+            sum(numeric_metric_values) / len(numeric_metric_values)
+            if numeric_metric_values else None
         )
 
         # Format results for template
@@ -142,8 +148,10 @@ def generate_to_file(self, report: Any, output_path: Path | str) -> Path:
             Path to the written file.
         """
         path = Path(output_path)
-        if not str(path).endswith(".html"):
+        if path.suffix == "":
             path = path / "report.html"
+        elif path.suffix.lower() != ".html":
+            path = path.with_suffix(".html")
         path = self._ensure_output_dir(path)
         content = self.generate(report)
         path.write_text(content, encoding="utf-8")

tests/unit/test_reports/test_html.py

@@ -111,6 +111,17 @@ def test_generate_to_file_adds_extension(
         assert result_path.exists()
         assert str(result_path).endswith(".html")
 
+    def test_generate_to_file_replaces_non_html_extension(
+        self, evaluation_report: Any, tmp_path: Path
+    ) -> None:
+        """generate_to_file() replaces non-.html suffixes with .html."""
+        report = HTMLReport()
+        output_path = tmp_path / "report.txt"
+        result_path = report.generate_to_file(evaluation_report, output_path)
+
+        assert result_path == tmp_path / "report.html"
+        assert result_path.exists()
+
     def test_generate_to_file_creates_directory(
         self, evaluation_report: Any, tmp_path: Path
     ) -> None:
@@ -142,3 +153,40 @@ def test_css_styles_present(self, evaluation_report: Any) -> None:
         result = report.generate(evaluation_report)
         assert "<style>" in result
         assert "var(--primary)" in result
+
+    def test_generate_ignores_non_numeric_metrics_for_overall_score(
+        self, evaluation_report: Any
+    ) -> None:
+        """generate() computes overall score from numeric metric values only."""
+        report = HTMLReport()
+        evaluation_report.summary["metrics_summary"] = {
+            "precision": 0.8,
+            "count": 2,
+            "ignored_bool": True,
+            "ignored_text": "0.9",
+            "ignored_dict": {"value": 1.0},
+        }
+        captured_context: dict[str, Any] = {}
+
+        def _capture_context(context: dict[str, Any]) -> str:
+            captured_context.update(context)
+            return "ok"
+
+        report._render_template = _capture_context
+        report.generate(evaluation_report)
+
+        assert captured_context["overall_score"] == pytest.approx(1.4)
+
+    def test_load_template_missing_builtin_has_installation_guidance(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """_load_template() raises a packaging-focused guidance message."""
+        report = HTMLReport()
+        monkeypatch.setattr(Path, "exists", lambda _: False)
+
+        with pytest.raises(FileNotFoundError) as exc_info:
+            report._load_template()
+
+        message = str(exc_info.value)
+        assert "openagent_eval/reports/templates/report.html" in message
+        assert "Install jinja2" not in message