@@ -317,7 +317,7 @@ their eval kind, and states latency for each:
   `agentic_general_question`, `agentic_guardrail` and `agentic_search`.
 - **Multi-turn** — kinds that drive a simulated user or fixture turns: `agentic_visualization` (and
   `vis_agentic`), `agentic_metric_skill`, `agentic_alert_skill`, `agentic_dashboard_skill`,
-  `agentic_report_skill`, `agentic_kda_skill`, `agentic_anomaly_detection`, `agentic_what_if` and
+  `agentic_document_skill`, `agentic_kda_skill`, `agentic_anomaly_detection`, `agentic_what_if` and
   `agentic_conversation`.
 
 ```json
 
@@ -15,11 +15,11 @@
 from gooddata_eval.core.agentic.anomaly_detection import evaluate_agentic_anomaly_detection
 from gooddata_eval.core.agentic.conversation import ConversationFixture, evaluate_agentic_conversation
 from gooddata_eval.core.agentic.dashboard_skill import evaluate_agentic_dashboard_skill
+from gooddata_eval.core.agentic.document_skill import evaluate_agentic_document_skill
 from gooddata_eval.core.agentic.general_question import evaluate_agentic_general_question
 from gooddata_eval.core.agentic.guardrail import evaluate_agentic_guardrail
 from gooddata_eval.core.agentic.kda_skill import evaluate_agentic_kda_skill
 from gooddata_eval.core.agentic.metric_skill import evaluate_agentic_metric_skill
-from gooddata_eval.core.agentic.report_skill import evaluate_agentic_report_skill
 from gooddata_eval.core.agentic.search_tool import evaluate_agentic_search_tool
 from gooddata_eval.core.agentic.visualization import evaluate_agentic_visualization
 from gooddata_eval.core.agentic.what_if import evaluate_agentic_what_if
@@ -39,6 +39,7 @@ class _LfKw(TypedDict, total=False):
     model_version_override: str | None
     reasoning_effort: ReasoningEffort | None
     submit_trace_link: SubmitTraceLink
+    run_metadata_extra: dict[str, Any] | None
 
 
 AGENTIC_TEST_KINDS = frozenset(
@@ -48,7 +49,7 @@ class _LfKw(TypedDict, total=False):
         "agentic_metric_skill",
         "agentic_alert_skill",
         "agentic_dashboard_skill",
-        "agentic_report_skill",
+        "agentic_document_skill",
         "agentic_search",
         "agentic_general_question",
         "agentic_guardrail",
@@ -99,8 +100,8 @@ class _LfKw(TypedDict, total=False):
 #
 # agentic_dashboard_skill is absent by default rather than by evidence: gen-ai holds the draft and
 # any chart it authors in conversation state and writes neither until a user saves from the UI, so
-# it is a candidate for the allowlist once the dataset has runs behind it. agentic_report_skill is
-# absent for the same reason: the report draft stays in conversation state until a user saves it.
+# it is a candidate for the allowlist once the dataset has runs behind it. agentic_document_skill is
+# absent for the same reason: the document draft stays in conversation state until a user saves it.
 WORKSPACE_MUTATING_TEST_KINDS = frozenset(AGENTIC_TEST_KINDS) - PARALLEL_SAFE_TEST_KINDS
 
 
@@ -153,11 +154,13 @@ def _dispatch_agentic(
     agent_id: str | None = None,
     submit_trace_link: SubmitTraceLink = run_trace_link_inline,
     gate: EvalGate = DEFAULT_GATE,
+    run_metadata_extra: dict[str, Any] | None = None,
 ) -> AgenticEvalOutcome:
     """Call the appropriate evaluate_agentic_* function for the item's test_kind.
 
     `gate` reaches every kind except those in UNGATED_AGENTIC_TEST_KINDS, which have no K
     to gate over; the CLI refuses --gate power for a dataset containing one.
+    `run_metadata_extra` reaches every kind unchanged.
 
     Every evaluate_agentic_* function returns an AgenticEvalOutcome (reasoning_steps,
     conversation_id, response_id, detail) on success and attaches the same four attributes
@@ -173,6 +176,7 @@ def _dispatch_agentic(
         "model_version_override": model_version_override,
         "reasoning_effort": reasoning_effort,
         "submit_trace_link": submit_trace_link,
+        "run_metadata_extra": run_metadata_extra,
     }
 
     if kind in ("vis_agentic", "agentic_visualization"):
@@ -215,8 +219,8 @@ def _dispatch_agentic(
             user_context=item.user_context,
             **lf_kw,
         )
-    elif kind == "agentic_report_skill":
-        return evaluate_agentic_report_skill(
+    elif kind == "agentic_document_skill":
+        return evaluate_agentic_document_skill(
             host=host,
             token=token,
             workspace_id=workspace_id,
@@ -413,6 +417,7 @@ def run_agentic_items(
     agent_id: str | None = None,
     concurrency: int = 1,
     gate: EvalGate = DEFAULT_GATE,
+    tags: tuple[str, ...] = (),
 ) -> EvalReport:
     """Run agentic items through evaluate_agentic_* and return an EvalReport.
 
@@ -421,9 +426,14 @@ def run_agentic_items(
     from the parallel ones -- a metric being created and dropped mid-run would otherwise be
     visible to a catalog-reading item running alongside it.
 
+    ``tags`` are added to every Langfuse trace gd-eval writes for these items, after the
+    automatic ones. They reach Langfuse whenever trace linking runs, with or without
+    ``use_langfuse``.
+
     Results are collected in dataset order regardless of completion order.
     """
     langfuse = make_langfuse_client() if use_langfuse else None
+    run_metadata_extra: dict[str, Any] | None = {"tags": list(tags)} if tags else None
 
     report = EvalReport(model=model_version, gate=normalize_gate(gate))
     total = len(items)
@@ -469,6 +479,7 @@ def _process_item(index: int, item: DatasetItem) -> ItemReport:
                 agent_id,
                 submit_trace_link=linker.submit,
                 gate=gate,
+                run_metadata_extra=run_metadata_extra,
             )
             if isinstance(outcome, AgenticEvalOutcome):
                 reasoning_steps = outcome.reasoning_steps
 
@@ -183,6 +183,17 @@ def _build_parser() -> argparse.ArgumentParser:
         action="store_true",
         help="Log scores and traces to Langfuse (requires --langfuse-dataset and LANGFUSE_* env vars).",
     )
+    run.add_argument(
+        "--tag",
+        dest="tags",
+        action="append",
+        default=[],
+        type=_parse_tag,
+        metavar="TAG",
+        help="Custom tag for every Langfuse trace gd-eval writes, next to the automatic ones. "
+        "Repeat for several tags. Applied wherever traces are written: single-turn kinds only "
+        "with --langfuse, agentic kinds whenever LANGFUSE_* env vars are set.",
+    )
     run.add_argument(
         "--agent-id",
         dest="agent_id",
@@ -377,6 +388,14 @@ def _parse_model_arg(val: str) -> tuple[str | None, str]:
     return None, val
 
 
+def _parse_tag(val: str) -> str:
+    """Strip one --tag value and reject a blank one, which Langfuse would store as an empty tag."""
+    tag = val.strip()
+    if not tag:
+        raise argparse.ArgumentTypeError("tag must not be blank")
+    return tag
+
+
 def _make_progress_callbacks(console: Console):
     """Build (on_item_start, on_run_done, on_item_done) callbacks that stream progress.
 
@@ -549,6 +568,7 @@ def _run(config: RunConfig) -> int:
                     model_id=resolved.model_id,
                     provider_type=resolved.provider_type,
                     reasoning_effort=config.reasoning_effort,
+                    tags=config.tags,
                 )
 
                 def on_langfuse_item_done(
@@ -574,6 +594,7 @@ def on_langfuse_item_done(
                     model_version=resolved.model_id,
                     reasoning_effort=config.reasoning_effort,
                     use_langfuse=config.log_to_langfuse,
+                    tags=config.tags,
                     run_ts=run_ts,
                     on_item_start=on_item_start,
                     on_item_done=on_item_done,
@@ -728,6 +749,7 @@ def main(argv: list[str] | None = None) -> int:
             agent_id=args.agent_id or os.environ.get("GD_EVAL_AGENT_ID"),
             turn_timeout_s=args.turn_timeout,
             item_timeout_s=args.item_timeout,
+            tags=tuple(dict.fromkeys(args.tags)),
         )
         return _run(config)
     except (
 
@@ -24,6 +24,14 @@
     evaluate_agentic_dashboard_skill,
     run_agentic_dashboard_skill,
 )
+from gooddata_eval.core.agentic.document_skill import (
+    AgenticDocumentSummary,
+    DocumentEvaluation,
+    DocumentRunResult,
+    DocumentSkillAssertionError,
+    evaluate_agentic_document_skill,
+    run_agentic_document_skill,
+)
 from gooddata_eval.core.agentic.general_question import (
     AgenticGeneralQuestionSummary,
     GeneralQuestionAssertionError,
@@ -53,14 +61,6 @@
     evaluate_agentic_metric_skill,
     run_agentic_metric_skill,
 )
-from gooddata_eval.core.agentic.report_skill import (
-    AgenticReportSummary,
-    ReportEvaluation,
-    ReportRunResult,
-    ReportSkillAssertionError,
-    evaluate_agentic_report_skill,
-    run_agentic_report_skill,
-)
 from gooddata_eval.core.agentic.search_tool import (
     AgenticSearchSummary,
     SearchResult,
@@ -76,13 +76,22 @@
     run_agentic_visualization,
 )
 
+# Deprecated names of the document skill, kept for existing callers.
+AgenticReportSummary = AgenticDocumentSummary
+ReportEvaluation = DocumentEvaluation
+ReportRunResult = DocumentRunResult
+ReportSkillAssertionError = DocumentSkillAssertionError
+evaluate_agentic_report_skill = evaluate_agentic_document_skill
+run_agentic_report_skill = run_agentic_document_skill
+
 __all__ = [
     "AgenticAlertSummary",
     "AgenticDashboardSummary",
     "AgenticGeneralQuestionSummary",
     "AgenticGuardrailSummary",
     "AgenticKdaSummary",
     "AgenticMetricSummary",
+    "AgenticDocumentSummary",
     "AgenticReportSummary",
     "AgenticSearchSummary",
     "AgenticRunSummary",
@@ -95,6 +104,9 @@
     "DashboardEvaluation",
     "DashboardRunResult",
     "DashboardSkillAssertionError",
+    "DocumentEvaluation",
+    "DocumentRunResult",
+    "DocumentSkillAssertionError",
     "GeneralQuestionAssertionError",
     "GeneralQuestionResult",
     "GuardrailAssertionError",
@@ -116,6 +128,7 @@
     "evaluate_agentic_alert_skill",
     "evaluate_agentic_conversation",
     "evaluate_agentic_dashboard_skill",
+    "evaluate_agentic_document_skill",
     "evaluate_agentic_general_question",
     "evaluate_agentic_guardrail",
     "evaluate_agentic_kda_skill",
@@ -126,6 +139,7 @@
     "run_agentic_alert_skill",
     "run_agentic_conversation",
     "run_agentic_dashboard_skill",
+    "run_agentic_document_skill",
     "run_agentic_general_question",
     "run_agentic_guardrail",
     "run_agentic_kda_skill",
 
@@ -21,6 +21,7 @@
     ExperimentRun,
     ScoreTarget,
     build_experiment_root_span,
+    trace_tags,
 )
 
 # Part of this module's public surface: external callers import both names from here.
@@ -302,7 +303,10 @@ def _experiment_root_span(
     """gd-eval's own root span for one (dataset item, run) -- the whole experiment item."""
     start, end = _span_window(trace, window)
     session_id = conversation_id or getattr(trace, "session_id", None)
-    tags = tuple(tag for tag in ("gd-eval", run_metadata.get("testing_framework")) if tag)
+    custom_tags = run_metadata.get("tags") or ()
+    if isinstance(custom_tags, str):
+        custom_tags = (custom_tags,)
+    tags = trace_tags("gd-eval", run_metadata.get("testing_framework"), *custom_tags)
     # One predicate for both the name and the input, so a falsy question cannot name the span
     # after the item while the input still says "question".
     has_input = item_input is not None
@@ -514,7 +518,8 @@ def build_run_context(
         model_version_override: Explicit model-version tag; when set it wins over the
             workspace lookup.
         run_metadata_extra: Optional extra key/values added to the dataset-run metadata
-            (e.g. a testing-framework tag or a CI run id for scoping). Default None keeps
+            (e.g. a testing-framework tag or a CI run id for scoping). A `tags` list is
+            also added to the trace tags, after the automatic ones. Default None keeps
             behavior unchanged. The SDK-derived model_version is applied last and cannot
             be overwritten by this dict.
         reasoning_effort: Effort the run requested, stamped into both the run name and