agentevals-cli 0.9.6__tar.gz → 0.9.7__tar.gz

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentevals-cli
-Version: 0.9.6
+Version: 0.9.7
 Summary: Standalone framework to evaluate agent correctness based on portable OpenTelemetry traces
 License-File: LICENSE
 Requires-Python: >=3.11

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/examples/custom_sink/agentevals_example_custom_sink/sink.py

@@ -38,7 +38,7 @@ class DemoNdjsonSink(ResultSink):
     async def _write(self, payload: dict) -> None:
         async with self._lock:
             self._path.parent.mkdir(parents=True, exist_ok=True)
-            with self._path.open("a") as f:  # noqa: ASYNC230
+            with self._path.open("a", encoding="utf-8") as f:  # noqa: ASYNC230
                 f.write(json.dumps(payload) + "\n")
 
     async def emit_partial(self, run_id: UUID, results: list[Result], attempt: int) -> None:

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/api/app.py

@@ -141,6 +141,7 @@ def create_app(
     *,
     trace_manager: StreamingTraceManager | None = None,
     enable_streaming: bool = False,
+    static_dir: Path | None = None,
 ) -> FastAPI:
     """Create the main agentevals API app."""
     app = FastAPI(
@@ -216,7 +217,8 @@ def create_app(
                 },
             )
 
-    static_dir = Path(__file__).parent.parent / "_static"
+    if static_dir is None:
+        static_dir = Path(__file__).parent.parent / "_static"
     has_ui = static_dir.is_dir() and (static_dir / "index.html").exists()
 
     if has_ui and not os.getenv("AGENTEVALS_HEADLESS"):
@@ -229,10 +231,12 @@ def create_app(
         async def root():
             return FileResponse(static_dir / "index.html")
 
+        static_root = static_dir.resolve()
+
         @app.get("/{path:path}")
         async def spa_fallback(path: str):
-            file_path = static_dir / path
-            if file_path.is_file():
+            file_path = (static_dir / path).resolve()
+            if file_path.is_relative_to(static_root) and file_path.is_file():
                 return FileResponse(file_path)
             return FileResponse(static_dir / "index.html")
 

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/api/debug_routes.py

@@ -114,7 +114,7 @@ def _collect_temp_files(session_ids: set[str] | None = None) -> dict[str, str]:
                 if sid not in session_ids:
                     continue
             try:
-                with open(path) as f:
+                with open(path, encoding="utf-8") as f:
                     files[basename] = f.read()
             except OSError:
                 logger.debug("Could not read temp file %s", path)

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/api/models.py

@@ -119,6 +119,24 @@ class TraceConversionMetadata(CamelModel):
     model: str | None = None
     response_model: str | None = None
     provider: str | None = None
+    schema_version: str | None = Field(
+        default=None,
+        description=(
+            "Version segment parsed from the emitting scope's schema URL "
+            "(the path component after the schema family prefix). Per the "
+            "OpenTelemetry schema URL spec, this version is only meaningful "
+            "within its schema family — the host/path prefix before the "
+            "version. It is comparable across traces only if they share a "
+            "family, and equals the OpenTelemetry Semantic Conventions "
+            "version only for the OTel family "
+            "(https://opentelemetry.io/schemas/<version>). For traces from "
+            "a custom or mirrored schema family, this is that family's own "
+            "version number, not an OTel semconv version. The full schema "
+            "URL (otel.schema_url) is preserved separately wherever the "
+            "family needs to be disambiguated; this field is informational "
+            "only and does not drive extraction or alias-resolution logic."
+        ),
+    )
     start_time: int | None = None
     user_input_preview: str | None = None
     final_output_preview: str | None = None

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/api/otlp_processing.py

@@ -19,6 +19,7 @@ from ..trace_attrs import (
     OTEL_GENAI_CONVERSATION_ID,
     OTEL_GENAI_INPUT_MESSAGES,
     OTEL_GENAI_OUTPUT_MESSAGES,
+    OTEL_SCHEMA_URL,
     OTEL_SCOPE,
     OTEL_SCOPE_VERSION,
 )
@@ -38,6 +39,7 @@ async def process_traces(body: dict, manager: StreamingTraceManager) -> None:
     for resource_span in body.get("resourceSpans", []):
         resource_attrs = resource_span.get("resource", {}).get("attributes", [])
         metadata = _extract_agentevals_metadata(resource_attrs)
+        resource_schema_url = resource_span.get("schemaUrl", "")
 
         if not metadata.get("conversation_id"):
             metadata["conversation_id"] = _prescan_conversation_id(resource_span)
@@ -46,9 +48,10 @@ async def process_traces(body: dict, manager: StreamingTraceManager) -> None:
             scope = scope_span.get("scope", {})
             scope_name = scope.get("name", "")
             scope_version = scope.get("version", "")
+            schema_url = scope_span.get("schemaUrl", "") or resource_schema_url
 
             for span_data in scope_span.get("spans", []):
-                span = _normalize_span(span_data, scope_name, scope_version)
+                span = _normalize_span(span_data, scope_name, scope_version, schema_url)
                 trace_id = span.get("traceId", "")
 
                 if not trace_id:
@@ -200,12 +203,13 @@ def fix_protobuf_id_fields(data) -> None:
 _GENAI_EVENT_KEYS = {OTEL_GENAI_INPUT_MESSAGES, OTEL_GENAI_OUTPUT_MESSAGES}
 
 
-def _normalize_span(span_data: dict, scope_name: str, scope_version: str) -> dict:
+def _normalize_span(span_data: dict, scope_name: str, scope_version: str, schema_url: str = "") -> dict:
     """Normalize an OTLP span for the downstream pipeline.
 
     Performs two transformations:
-    1. Injects otel.scope.name/version from the scopeSpans level into span
-       attributes (the pipeline expects them there).
+    1. Injects otel.scope.name/version (and otel.schema_url) from the
+       scopeSpans level into span attributes (the pipeline expects them
+       there).
     2. Promotes gen_ai.input.messages and gen_ai.output.messages from span
        events to span attributes. Some SDKs (e.g. Strands with
        OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental) store
@@ -222,6 +226,9 @@ def _normalize_span(span_data: dict, scope_name: str, scope_version: str) -> dic
     if scope_version and OTEL_SCOPE_VERSION not in existing_keys:
         attrs.append({"key": OTEL_SCOPE_VERSION, "value": {"stringValue": scope_version}})
         existing_keys.add(OTEL_SCOPE_VERSION)
+    if schema_url and OTEL_SCHEMA_URL not in existing_keys:
+        attrs.append({"key": OTEL_SCHEMA_URL, "value": {"stringValue": schema_url}})
+        existing_keys.add(OTEL_SCHEMA_URL)
 
     for event in span.get("events", []):
         for attr in event.get("attributes", []):

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/api/routes.py

@@ -119,6 +119,15 @@ def _camel_keys(obj: Any) -> Any:
     return obj
 
 
+def _safe_upload_path(temp_dir: str, filename: str, idx: int = 0) -> str:
+    """Build a save path inside ``temp_dir`` from a client-controlled filename.
+
+    ``basename`` drops any directory components, and the ``idx`` prefix keeps
+    same-named uploads from clobbering each other.
+    """
+    return os.path.join(temp_dir, f"{idx}_{os.path.basename(filename)}")
+
+
 def _load_eval_set_dict(path: str | None) -> dict | None:
     """Read the uploaded eval set file back into a dict for persistence.
 
@@ -129,7 +138,7 @@ def _load_eval_set_dict(path: str | None) -> dict | None:
     if not path:
         return None
     try:
-        with open(path) as f:
+        with open(path, encoding="utf-8") as f:
             return json.load(f)
     except (OSError, json.JSONDecodeError):
         logger.warning("could not re-read eval_set file at %s for persistence", path)
@@ -361,7 +370,7 @@ async def validate_eval_set(
 ):
     temp_dir = tempfile.mkdtemp()
     try:
-        eval_set_path = os.path.join(temp_dir, eval_set_file.filename or "eval_set.json")
+        eval_set_path = os.path.join(temp_dir, f"evalset_{os.path.basename(eval_set_file.filename or 'eval_set.json')}")
         with open(eval_set_path, "wb") as f:  # noqa: ASYNC230
             content = await eval_set_file.read()
             f.write(content)
@@ -433,8 +442,7 @@ async def convert_trace_files(
                     detail=f"Invalid file extension for {original}. Only .json and .jsonl files are allowed.",
                 )
 
-            safe_name = f"{idx}_{os.path.basename(original)}"
-            trace_path = os.path.join(temp_dir, safe_name)
+            trace_path = _safe_upload_path(temp_dir, original, idx)
             with open(trace_path, "wb") as f:  # noqa: ASYNC230
                 content = await trace_file.read()
 
@@ -487,6 +495,7 @@ async def convert_trace_files(
                 meta = TraceConversionMetadata(
                     agent_name=meta_dict.get("agent_name"),
                     model=meta_dict.get("model"),
+                    schema_version=meta_dict.get("schema_version"),
                     start_time=meta_dict.get("start_time"),
                     user_input_preview=meta_dict.get("user_input_preview"),
                     final_output_preview=meta_dict.get("final_output_preview"),
@@ -547,7 +556,7 @@ async def evaluate_traces(
             raise HTTPException(status_code=400, detail=f"Invalid credentialRefs: {exc}") from exc
 
         trace_paths = []
-        for trace_file in trace_files:
+        for idx, trace_file in enumerate(trace_files):
             if not trace_file.filename:
                 continue
 
@@ -557,7 +566,7 @@ async def evaluate_traces(
                     detail=f"Invalid file extension for {trace_file.filename}. Only .json and .jsonl files are allowed.",
                 )
 
-            trace_path = os.path.join(temp_dir, trace_file.filename)
+            trace_path = _safe_upload_path(temp_dir, trace_file.filename, idx)
             with open(trace_path, "wb") as f:  # noqa: ASYNC230
                 content = await trace_file.read()
 
@@ -586,7 +595,7 @@ async def evaluate_traces(
                     detail="Invalid file extension for eval set. Only .json files are allowed.",
                 )
 
-            eval_set_path = os.path.join(temp_dir, eval_set_file.filename)
+            eval_set_path = os.path.join(temp_dir, f"evalset_{os.path.basename(eval_set_file.filename)}")
             with open(eval_set_path, "wb") as f:  # noqa: ASYNC230
                 content = await eval_set_file.read()
                 if len(content) > 10 * 1024 * 1024:
@@ -668,7 +677,7 @@ async def evaluate_traces_stream(
                 return
 
             trace_paths = []
-            for trace_file in trace_files:
+            for idx, trace_file in enumerate(trace_files):
                 if not trace_file.filename:
                     continue
 
@@ -676,7 +685,7 @@ async def evaluate_traces_stream(
                     yield f"data: {SSEErrorEvent(error=f'Invalid file extension for {trace_file.filename}').model_dump_json(by_alias=True)}\n\n"
                     return
 
-                trace_path = os.path.join(temp_dir, trace_file.filename)
+                trace_path = _safe_upload_path(temp_dir, trace_file.filename, idx)
                 with open(trace_path, "wb") as f:  # noqa: ASYNC230
                     content = await trace_file.read()
 
@@ -699,7 +708,7 @@ async def evaluate_traces_stream(
                     yield f"data: {SSEErrorEvent(error='Invalid file extension for eval set').model_dump_json(by_alias=True)}\n\n"
                     return
 
-                eval_set_path = os.path.join(temp_dir, eval_set_file.filename)
+                eval_set_path = os.path.join(temp_dir, f"evalset_{os.path.basename(eval_set_file.filename)}")
                 with open(eval_set_path, "wb") as f:  # noqa: ASYNC230
                     content = await eval_set_file.read()
                     if len(content) > 10 * 1024 * 1024:

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/api/streaming_routes.py

@@ -5,6 +5,7 @@ from __future__ import annotations
 import asyncio
 import json
 import logging
+import os
 from typing import TYPE_CHECKING
 
 from fastapi import APIRouter, Depends, HTTPException, Request
@@ -15,6 +16,7 @@ from ..config import BuiltinMetricDef, EvalParams, EvalRunConfig, EvaluatorDef
 from ..converter import convert_traces
 from ..loader.otlp import OtlpJsonLoader
 from ..runner import RunResult, run_evaluation
+from ..streaming.exports import EXPORT_DIR, export_name
 from ..trace_attrs import OTEL_GENAI_INPUT_MESSAGES, OTEL_GENAI_REQUEST_MODEL
 from .dependencies import require_trace_manager
 from .models import (
@@ -214,6 +216,7 @@ async def evaluate_sessions(
     if not golden_session:
         raise HTTPException(status_code=404, detail="Golden session not found")
 
+    eval_set_file = None
     try:
         eval_set_response = await _do_create_eval_set(
             CreateEvalSetRequest(
@@ -225,7 +228,9 @@ async def evaluate_sessions(
 
         import tempfile
 
-        eval_set_file = tempfile.NamedTemporaryFile(mode="w", suffix=".json", delete=False)
+        eval_set_file = tempfile.NamedTemporaryFile(
+            mode="w", suffix=".json", delete=False, dir=str(EXPORT_DIR), encoding="utf-8"
+        )
         json.dump(eval_set_response.data.eval_set, eval_set_file)
         eval_set_file.close()
 
@@ -307,6 +312,14 @@ async def evaluate_sessions(
     except Exception as exc:
         logger.exception("Failed to evaluate sessions")
         raise HTTPException(status_code=500, detail=str(exc)) from exc
+    finally:
+        # This eval set is consumed only by run_evaluation above and is never
+        # served for download, so remove it once evaluation completes.
+        if eval_set_file is not None:
+            try:
+                os.unlink(eval_set_file.name)
+            except OSError:
+                pass
 
 
 @streaming_router.post("/prepare-evaluation", response_model=StandardResponse[PrepareEvaluationData])
@@ -328,13 +341,8 @@ async def prepare_evaluation(
             manager,
         )
 
-        import os
-        import tempfile
-
-        temp_dir = tempfile.gettempdir()
-
-        eval_set_file = os.path.join(temp_dir, f"eval_set_{request.golden_session_id}.json")
-        with open(eval_set_file, "w") as f:  # noqa: ASYNC230
+        eval_set_file = EXPORT_DIR / export_name(request.golden_session_id, prefix="eval_set_", suffix=".json")
+        with open(eval_set_file, "w", encoding="utf-8") as f:  # noqa: ASYNC230
             json.dump(eval_set_response.data.eval_set, f)
 
         trace_files = []
@@ -353,7 +361,7 @@ async def prepare_evaluation(
 
         return StandardResponse(
             data=PrepareEvaluationData(
-                eval_set_url=f"/api/streaming/download/{os.path.basename(eval_set_file)}",
+                eval_set_url=f"/api/streaming/download/{eval_set_file.name}",
                 trace_urls=[f"/api/streaming/download/{os.path.basename(tf['file_path'])}" for tf in trace_files],
                 num_traces=len(trace_files),
             )
@@ -368,20 +376,25 @@ async def prepare_evaluation(
 
 @streaming_router.get("/download/{filename}")
 async def download_file(filename: str):
-    """Download a prepared trace or eval set file."""
-    import os
-    import tempfile
+    """Download a prepared trace or eval set file from the export directory.
+
+    Only bare filenames are accepted, and the resolved path must stay inside
+    the export directory. Anything that resolves outside it returns the same
+    404 as a missing file.
+    """
+    if filename in {"", ".", ".."} or filename != os.path.basename(filename):
+        raise HTTPException(status_code=400, detail="Invalid filename")
 
-    temp_dir = tempfile.gettempdir()
-    file_path = os.path.join(temp_dir, filename)
+    export_root = EXPORT_DIR.resolve()
+    candidate = (export_root / filename).resolve()
 
-    if not os.path.exists(file_path):  # noqa: ASYNC240
+    if not candidate.is_relative_to(export_root):
         raise HTTPException(status_code=404, detail="File not found")
 
-    if not file_path.startswith(temp_dir):
-        raise HTTPException(status_code=400, detail="Invalid file path")
+    if not candidate.is_file():  # noqa: ASYNC240
+        raise HTTPException(status_code=404, detail="File not found")
 
-    return FileResponse(file_path, media_type="application/json", filename=filename)
+    return FileResponse(candidate, media_type="application/json", filename=filename)
 
 
 @streaming_router.post("/get-trace", response_model=StandardResponse[GetTraceData])
@@ -415,7 +428,7 @@ async def get_trace(
         # evaluate-sessions path. Serializing spans here independently would
         # drop the resource attribute and lose agent identity on the run.
         trace_file = await manager._save_spans_to_temp_file(session)
-        with open(trace_file) as f:  # noqa: ASYNC230
+        with open(trace_file, encoding="utf-8") as f:  # noqa: ASYNC230
             trace_content = f.read()
         num_spans = sum(1 for line in trace_content.splitlines() if line.strip())
 

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/cli.py

@@ -840,8 +840,8 @@ def migrate_create(name: str, output_dir: str | None) -> None:
         f"-- Migration {next_version:06d}: {name}\n"
         "-- Once tagged in a release this file is immutable. Fix bugs by adding a NEW migration.\n\n"
     )
-    up_path.write_text(header)
-    down_path.write_text(header)
+    up_path.write_text(header, encoding="utf-8")
+    down_path.write_text(header, encoding="utf-8")
     click.echo(f"Created {up_path}")
     click.echo(f"Created {down_path}")
 

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/eval_config_loader.py

@@ -58,7 +58,7 @@ def load_eval_config(path: str | Path) -> EvalRunConfig:
     if not path.exists():
         raise FileNotFoundError(f"Eval config file not found: {path}")
 
-    with open(path) as f:
+    with open(path, encoding="utf-8") as f:
         data = yaml.safe_load(f)
 
     if not isinstance(data, dict):

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/evaluator/sources.py

@@ -66,7 +66,7 @@ def _read_cache(key: str, ttl: int = _CACHE_TTL_SECONDS) -> list[EvaluatorInfo]
     if not cache_file.exists():
         return None
     try:
-        data = json.loads(cache_file.read_text())
+        data = json.loads(cache_file.read_text(encoding="utf-8"))
         if time.time() - data.get("ts", 0) > ttl:
             return None
         return [EvaluatorInfo(**item) for item in data["evaluators"]]
@@ -83,7 +83,8 @@ def _write_cache(key: str, evaluators: list[EvaluatorInfo]) -> None:
                     "ts": time.time(),
                     "evaluators": [asdict(g) for g in evaluators],
                 }
-            )
+            ),
+            encoding="utf-8",
         )
     except Exception:
         pass

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/evaluator/venv.py

@@ -42,7 +42,11 @@ def _venv_key(evaluator_path: Path) -> str:
 
 def _is_venv_valid(venv_dir: Path, req_hash: str) -> bool:
     hash_file = venv_dir / _HASH_FILE
-    return _venv_python(venv_dir).exists() and hash_file.exists() and hash_file.read_text().strip() == req_hash
+    return (
+        _venv_python(venv_dir).exists()
+        and hash_file.exists()
+        and hash_file.read_text(encoding="utf-8").strip() == req_hash
+    )
 
 
 def _create_venv(venv_dir: Path, uv: str | None) -> None:
@@ -104,7 +108,7 @@ def ensure_venv(evaluator_path: Path) -> Path | None:
         stderr = exc.stderr.decode() if isinstance(exc.stderr, bytes) else (exc.stderr or "")
         raise RuntimeError(f"Failed to set up environment for evaluator '{evaluator_path.stem}': {stderr}") from exc
 
-    (venv_dir / _HASH_FILE).write_text(req_hash)
+    (venv_dir / _HASH_FILE).write_text(req_hash, encoding="utf-8")
     logger.info("Environment ready for '%s'", evaluator_path.stem)
     return _venv_python(venv_dir)
 

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/extraction.py

@@ -13,6 +13,7 @@ from __future__ import annotations
 
 import json
 import logging
+import re
 from typing import Any, Protocol, TypedDict, TypeVar
 
 from .loader.base import Span, Trace
@@ -22,6 +23,7 @@ from .trace_attrs import (
     ADK_SCOPE_VALUE,
     ADK_TOOL_CALL_ARGS,
     ADK_TOOL_RESPONSE,
+    GENAI_ATTRIBUTE_ALIASES,
     OTEL_ERROR_TYPE,
     OTEL_GENAI_INPUT_MESSAGES,
     OTEL_GENAI_OP,
@@ -44,6 +46,7 @@ from .trace_attrs import (
     OTEL_GENAI_USAGE_CACHE_READ_TOKENS,
     OTEL_GENAI_USAGE_INPUT_TOKENS,
     OTEL_GENAI_USAGE_OUTPUT_TOKENS,
+    OTEL_SCHEMA_URL,
     OTEL_SCOPE,
 )
 from .utils.genai_messages import (
@@ -58,6 +61,61 @@ logger = logging.getLogger(__name__)
 
 FORMAT_DETECTION_SPAN_LIMIT = 10
 
+# ---------------------------------------------------------------------------
+# Alias-aware attribute resolution
+# ---------------------------------------------------------------------------
+
+# Per the OTel schema URL spec (https://opentelemetry.io/docs/specs/otel/schemas/#schema-url),
+# a schema URL has the form `http[s]://server[:port]/path/<version>`.
+_SCHEMA_VERSION_RE = re.compile(r"^\d+\.\d+\.\d+(?:-[0-9A-Za-z.-]+)?(?:\+[0-9A-Za-z.-]+)?$")
+
+
+def resolve_attr(attrs: dict[str, Any], canonical_key: str) -> Any | None:
+    """Look up *canonical_key* in *attrs*, falling back to older aliased names.
+
+    Resolution order:
+    1. The canonical (current/newest) key, if present with a truthy value.
+    2. Each alias in ``GENAI_ATTRIBUTE_ALIASES[canonical_key]``, in order,
+       first truthy value wins.
+    3. ``None`` if neither the canonical key nor any alias is present.
+
+    The canonical value always takes precedence over aliases, even if both
+    are present on the same span. This resolves only against the flat attrs
+    dict at read time - it never mutates or rewrites stored/transmitted
+    attributes.
+    """
+    value = attrs.get(canonical_key)
+    if value:
+        return value
+
+    for alias in GENAI_ATTRIBUTE_ALIASES.get(canonical_key, []):
+        alias_value = attrs.get(alias)
+        if alias_value:
+            return alias_value
+
+    return None
+
+
+def resolve_schema_version(schema_url: str | None) -> str | None:
+    """Extract the version segment from a schema URL.
+
+    Per the OTel schema URL spec, the version is the last path segment
+    (e.g. "1.37.0" in "https://opentelemetry.io/schemas/1.37.0").
+
+    It's scoped to that URL's schema family and only equals the OTel semconv
+    version for the official OTel family. Degrades gracefully to None for
+    missing, empty, or malformed input - never raises.
+    """
+    if not schema_url or not isinstance(schema_url, str):
+        return None
+
+    last_segment = schema_url.rstrip("/").rsplit("/", 1)[-1]
+    if _SCHEMA_VERSION_RE.match(last_segment):
+        return last_segment
+
+    return None
+
+
 # ---------------------------------------------------------------------------
 # Pure extraction functions (operate on flat attribute dicts)
 # ---------------------------------------------------------------------------
@@ -189,18 +247,28 @@ class ExtendedModelInfo(TypedDict):
     cache_creation_tokens: int
     cache_read_tokens: int
     error_type: str | None
+    schema_version: str | None
 
 
 def extract_extended_model_info_from_attrs(attrs: dict[str, Any]) -> ExtendedModelInfo:
     """Extract extended model and provider metadata from span attributes.
 
-    Uses gen_ai.system as fallback for provider when gen_ai.provider.name is
-    absent (backward compat with pre-v1.37.0 instrumentors).
+    Uses the alias-resolving lookup for attributes with known historical
+    renames (e.g. provider falls back to gen_ai.system when
+    gen_ai.provider.name - the canonical, current name - is absent, for
+    backward compat with pre-v1.37.0 instrumentors).
+
+    ``schema_version`` is resolved from the scope's ``schema_url`` (captured
+    at ingest as the ``otel.schema_url`` attribute) and is ``None`` when
+    absent or malformed. This is purely informational metadata about which
+    schema version emitted the span - it is never used to detect whether a
+    span is GenAI (that remains the sole responsibility of the existing
+    gen_ai.* presence checks).
     """
     return {
         "request_model": attrs.get(OTEL_GENAI_REQUEST_MODEL),
         "response_model": attrs.get(OTEL_GENAI_RESPONSE_MODEL),
-        "provider": attrs.get(OTEL_GENAI_PROVIDER_NAME) or attrs.get(OTEL_GENAI_SYSTEM),
+        "provider": resolve_attr(attrs, OTEL_GENAI_PROVIDER_NAME),
         "finish_reasons": _parse_finish_reasons(attrs.get(OTEL_GENAI_RESPONSE_FINISH_REASONS)),
         "response_id": attrs.get(OTEL_GENAI_RESPONSE_ID),
         "temperature": _safe_cast(attrs.get(OTEL_GENAI_REQUEST_TEMPERATURE), float),
@@ -208,6 +276,7 @@ def extract_extended_model_info_from_attrs(attrs: dict[str, Any]) -> ExtendedMod
         "cache_creation_tokens": _safe_cast(attrs.get(OTEL_GENAI_USAGE_CACHE_CREATION_TOKENS), int, 0),
         "cache_read_tokens": _safe_cast(attrs.get(OTEL_GENAI_USAGE_CACHE_READ_TOKENS), int, 0),
         "error_type": attrs.get(OTEL_ERROR_TYPE),
+        "schema_version": resolve_schema_version(attrs.get(OTEL_SCHEMA_URL)),
     }
 
 

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/loader/auto.py

@@ -51,7 +51,7 @@ def detect_format(path: str) -> str | None:
     load can fall back to a default.
     """
     try:
-        with open(path) as f:
+        with open(path, encoding="utf-8") as f:
             if path.lower().endswith(".jsonl"):
                 return OTLP_JSON
             data = json.load(f)

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/loader/jaeger.py

@@ -43,7 +43,7 @@ class JaegerJsonLoader(TraceLoader):
         return "jaeger-json"
 
     def load(self, source: str) -> list[Trace]:
-        with open(source) as f:
+        with open(source, encoding="utf-8") as f:
             raw = json.load(f)
 
         if not isinstance(raw, dict) or "data" not in raw:

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/loader/otlp.py

@@ -8,6 +8,7 @@ import logging
 from ..trace_attrs import (
     OTEL_GENAI_INPUT_MESSAGES,
     OTEL_GENAI_OUTPUT_MESSAGES,
+    OTEL_SCHEMA_URL,
     OTEL_SCOPE,
     OTEL_SCOPE_VERSION,
 )
@@ -35,7 +36,7 @@ class OtlpJsonLoader(TraceLoader):
 
     def load(self, source: str) -> list[Trace]:
         """Load OTLP JSON file or JSONL (one span per line)."""
-        with open(source) as f:
+        with open(source, encoding="utf-8") as f:
             content = f.read().strip()
 
         if not content:
@@ -93,21 +94,23 @@ class OtlpJsonLoader(TraceLoader):
         all_spans = []
         for resource_span in resource_spans:
             resource_attrs = self._extract_attributes(resource_span.get("resource", {}).get("attributes", []))
+            resource_schema_url = resource_span.get("schemaUrl", "")
             scope_spans = resource_span.get("scopeSpans") or resource_span.get("instrumentationLibrarySpans", [])
             for scope_span in scope_spans:
                 scope = scope_span.get("scope") or scope_span.get("instrumentationLibrary") or {}
                 scope_name = scope.get("name", "")
                 scope_version = scope.get("version", "")
+                schema_url = scope_span.get("schemaUrl", "") or resource_schema_url
 
                 for span_data in scope_span.get("spans", []):
-                    span = self._parse_span(span_data, resource_attrs, scope_name, scope_version)
+                    span = self._parse_span(span_data, resource_attrs, scope_name, scope_version, schema_url)
                     all_spans.append(span)
 
         return self._build_traces(all_spans)
 
     def _parse_otlp_spans(self, spans_data: list[dict]) -> list[Trace]:
         """Parse flat list of OTLP spans (JSONL format for streaming)."""
-        all_spans = [self._parse_span(span_data, {}, "", "") for span_data in spans_data]
+        all_spans = [self._parse_span(span_data, {}, "", "", "") for span_data in spans_data]
         return self._build_traces(all_spans)
 
     _GENAI_EVENT_KEYS = {OTEL_GENAI_INPUT_MESSAGES, OTEL_GENAI_OUTPUT_MESSAGES}
@@ -118,6 +121,7 @@ class OtlpJsonLoader(TraceLoader):
         resource_attrs: dict,
         scope_name: str,
         scope_version: str,
+        schema_url: str = "",
     ) -> Span:
         """Convert OTLP span to normalized Span object."""
         attributes = self._extract_attributes(span_data.get("attributes", []))
@@ -126,6 +130,8 @@ class OtlpJsonLoader(TraceLoader):
             attributes[OTEL_SCOPE] = scope_name
         if scope_version:
             attributes[OTEL_SCOPE_VERSION] = scope_version
+        if schema_url:
+            attributes[OTEL_SCHEMA_URL] = schema_url
 
         self._promote_genai_event_attributes(span_data, attributes)
 

{agentevals_cli-0.9.6 → agentevals_cli-0.9.7}/src/agentevals/mcp_server.py

@@ -378,7 +378,7 @@ def create_server(server_url: str | None = None, **fastmcp_kwargs: Any) -> FastM
 
         raw = await _post("/api/streaming/get-trace", {"session_id": session_id})
 
-        with tempfile.NamedTemporaryFile(mode="w", suffix=".jsonl", delete=False) as f:
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".jsonl", delete=False, encoding="utf-8") as f:
             f.write(raw["traceContent"])
             tmp_path = f.name