unique_toolkit 1.7.0__tar.gz → 1.8.1__tar.gz

{unique_toolkit-1.7.0 → unique_toolkit-1.8.1}/CHANGELOG.md

@@ -5,6 +5,13 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), 
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.8.1] - 2026-10-03
+- Fix bug where sub agent evaluation config variable `include_evaluation` did not include aliases for previous names.
+
+## [1.8.0] - 2026-10-03
+- Sub Agents now block when executing the same sub-agent multiple times with `reuse_chat` set to `True`.
+- Sub Agents tool, evaluation and post-processing refactored and tests added.
+
 ## [1.7.0] - 2025-10-01
 - Add functionality to remove text in `get_user_visible_chat_history`
 

{unique_toolkit-1.7.0 → unique_toolkit-1.8.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: unique_toolkit
-Version: 1.7.0
+Version: 1.8.1
 Summary: 
 License: Proprietary
 Author: Cedric Klinkert
@@ -118,6 +118,13 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), 
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.8.1] - 2026-10-03
+- Fix bug where sub agent evaluation config variable `include_evaluation` did not include aliases for previous names.
+
+## [1.8.0] - 2026-10-03
+- Sub Agents now block when executing the same sub-agent multiple times with `reuse_chat` set to `True`.
+- Sub Agents tool, evaluation and post-processing refactored and tests added.
+
 ## [1.7.0] - 2025-10-01
 - Add functionality to remove text in `get_user_visible_chat_history`
 

{unique_toolkit-1.7.0 → unique_toolkit-1.8.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "unique_toolkit"
-version = "1.7.0"
+version = "1.8.1"
 description = ""
 authors = [
     "Cedric Klinkert <cedric.klinkert@unique.ch>",

unique_toolkit-1.8.1/unique_toolkit/agentic/tools/a2a/__init__.py

@@ -0,0 +1,20 @@
+from unique_toolkit.agentic.tools.a2a.config import ExtendedSubAgentToolConfig
+from unique_toolkit.agentic.tools.a2a.evaluation import (
+    SubAgentEvaluationService,
+    SubAgentEvaluationServiceConfig,
+)
+from unique_toolkit.agentic.tools.a2a.manager import A2AManager
+from unique_toolkit.agentic.tools.a2a.postprocessing import (
+    SubAgentResponsesPostprocessor,
+)
+from unique_toolkit.agentic.tools.a2a.tool import SubAgentTool, SubAgentToolConfig
+
+__all__ = [
+    "SubAgentToolConfig",
+    "SubAgentTool",
+    "SubAgentResponsesPostprocessor",
+    "A2AManager",
+    "ExtendedSubAgentToolConfig",
+    "SubAgentEvaluationServiceConfig",
+    "SubAgentEvaluationService",
+]

unique_toolkit-1.8.1/unique_toolkit/agentic/tools/a2a/config.py

@@ -0,0 +1,17 @@
+from pydantic import Field
+
+from unique_toolkit.agentic.tools.a2a.evaluation import SubAgentEvaluationConfig
+from unique_toolkit.agentic.tools.a2a.postprocessing import SubAgentDisplayConfig
+from unique_toolkit.agentic.tools.a2a.tool import SubAgentToolConfig
+
+
+# SubAgentToolConfig with display and evaluation configs
+class ExtendedSubAgentToolConfig(SubAgentToolConfig):
+    response_display_config: SubAgentDisplayConfig = Field(
+        default_factory=SubAgentDisplayConfig,
+        description="Configuration for how to display the sub-agent response.",
+    )
+    evaluation_config: SubAgentEvaluationConfig = Field(
+        default_factory=SubAgentEvaluationConfig,
+        description="Configuration for handling assessments of the sub-agent response.",
+    )

unique_toolkit-1.8.1/unique_toolkit/agentic/tools/a2a/evaluation/__init__.py

@@ -0,0 +1,13 @@
+from unique_toolkit.agentic.tools.a2a.evaluation.config import (
+    SubAgentEvaluationConfig,
+    SubAgentEvaluationServiceConfig,
+)
+from unique_toolkit.agentic.tools.a2a.evaluation.evaluator import (
+    SubAgentEvaluationService,
+)
+
+__all__ = [
+    "SubAgentEvaluationService",
+    "SubAgentEvaluationServiceConfig",
+    "SubAgentEvaluationConfig",
+]

unique_toolkit-1.8.1/unique_toolkit/agentic/tools/a2a/evaluation/_utils.py

@@ -0,0 +1,66 @@
+import logging
+
+import unique_sdk
+
+from unique_toolkit.chat.schemas import (
+    ChatMessageAssessmentLabel,
+    ChatMessageAssessmentStatus,
+)
+
+logger = logging.getLogger(__name__)
+
+_ASSESSMENT_LABEL_COMPARISON_DICT: dict[str, int] = {
+    ChatMessageAssessmentLabel.RED: 0,
+    ChatMessageAssessmentLabel.YELLOW: 1,
+    ChatMessageAssessmentLabel.GREEN: 2,
+}
+
+
+def _sort_assessments(
+    assessments: list[unique_sdk.Space.Assessment],
+) -> list[unique_sdk.Space.Assessment]:
+    return sorted(
+        assessments,
+        key=lambda x: _ASSESSMENT_LABEL_COMPARISON_DICT[x["label"]],  # type: ignore (should be checked before sorting)
+    )
+
+
+def _worst_label(
+    *labels: str,
+) -> str:
+    return min(
+        labels,
+        key=lambda x: _ASSESSMENT_LABEL_COMPARISON_DICT[x],
+    )
+
+
+def _get_valid_assessments(
+    assessments: list[unique_sdk.Space.Assessment],
+    display_name: str,
+    sequence_number: int,
+) -> list[unique_sdk.Space.Assessment]:
+    valid_assessments = []
+    for assessment in assessments:
+        if (
+            assessment["label"] is None
+            or assessment["label"] not in ChatMessageAssessmentLabel
+        ):
+            logger.warning(
+                "Unkown assistant label %s for assistant %s (sequence number: %s) will be ignored",
+                assessment["label"],
+                display_name,
+                sequence_number,
+            )
+            continue
+        if assessment["status"] != ChatMessageAssessmentStatus.DONE:
+            logger.warning(
+                "Assessment %s for assistant %s (sequence number: %s) is not done (status: %s) will be ignored",
+                assessment["label"],
+                display_name,
+                sequence_number,
+                assessment["status"],
+            )
+            continue
+        valid_assessments.append(assessment)
+
+    return valid_assessments

{unique_toolkit-1.7.0 → unique_toolkit-1.8.1}/unique_toolkit/agentic/tools/a2a/evaluation/config.py

@@ -1,8 +1,9 @@
 from pathlib import Path
 
-from pydantic import BaseModel, Field
+from pydantic import AliasChoices, BaseModel, Field
 
 from unique_toolkit._common.default_language_model import DEFAULT_GPT_4o
+from unique_toolkit._common.pydantic_helpers import get_configuration_dict
 from unique_toolkit._common.validators import LMI, get_LMI_default_field
 from unique_toolkit.chat.schemas import (
     ChatMessageAssessmentType,
@@ -22,12 +23,13 @@ with open(Path(__file__).parent / "summarization_user_message.j2", "r") as file:
     DEFAULT_SUMMARIZATION_USER_MESSAGE_TEMPLATE = file.read().strip()
 
 
-class SubAgentEvaluationConfig(BaseModel):
+class SubAgentEvaluationServiceConfig(BaseModel):
+    model_config = get_configuration_dict()
+
     assessment_type: ChatMessageAssessmentType = Field(
         default=ChatMessageAssessmentType.COMPLIANCE,
         description="The type of assessment to use in the display.",
     )
-
     summarization_model: LMI = get_LMI_default_field(DEFAULT_GPT_4o)
     summarization_system_message: str = Field(
         default=DEFAULT_EVALUATION_SYSTEM_MESSAGE_TEMPLATE,
@@ -37,3 +39,17 @@ class SubAgentEvaluationConfig(BaseModel):
         default=DEFAULT_SUMMARIZATION_USER_MESSAGE_TEMPLATE,
         description="The user message template for the summarization model.",
     )
+
+
+class SubAgentEvaluationConfig(BaseModel):
+    model_config = get_configuration_dict()
+
+    include_evaluation: bool = Field(
+        default=True,
+        description="Whether to include the evaluation in the response.",
+        validation_alias=AliasChoices(
+            "includeEvaluation",
+            "displayEvalution",  # typo in old config name
+            "display_evalution",
+        ),
+    )

{unique_toolkit-1.7.0 → unique_toolkit-1.8.1}/unique_toolkit/agentic/tools/a2a/evaluation/evaluator.py

@@ -1,5 +1,4 @@
 import logging
-from collections import defaultdict
 from typing import override
 
 import unique_sdk
@@ -12,8 +11,16 @@ from unique_toolkit.agentic.evaluation.schemas import (
     EvaluationMetricName,
     EvaluationMetricResult,
 )
-from unique_toolkit.agentic.tools.a2a.evaluation.config import SubAgentEvaluationConfig
-from unique_toolkit.agentic.tools.a2a.service import SubAgentTool
+from unique_toolkit.agentic.tools.a2a.evaluation._utils import (
+    _get_valid_assessments,
+    _sort_assessments,
+    _worst_label,
+)
+from unique_toolkit.agentic.tools.a2a.evaluation.config import (
+    SubAgentEvaluationConfig,
+    SubAgentEvaluationServiceConfig,
+)
+from unique_toolkit.agentic.tools.a2a.tool import SubAgentTool
 from unique_toolkit.chat.schemas import (
     ChatMessageAssessmentLabel,
     ChatMessageAssessmentStatus,
@@ -27,20 +34,19 @@ logger = logging.getLogger(__name__)
 
 
 class _SubAgentToolInfo(TypedDict):
-    assessments: list[list[unique_sdk.Space.Assessment]]
+    assessments: dict[int, list[unique_sdk.Space.Assessment]]
     display_name: str
 
 
 NO_ASSESSMENTS_FOUND = "NO_ASSESSMENTS_FOUND"
 
 
-class SubAgentsEvaluation(Evaluation):
+class SubAgentEvaluationService(Evaluation):
     DISPLAY_NAME = "Sub Agents"
 
     def __init__(
         self,
-        config: SubAgentEvaluationConfig,
-        sub_agent_tools: list[SubAgentTool],
+        config: SubAgentEvaluationServiceConfig,
         language_model_service: LanguageModelService,
     ):
         super().__init__(EvaluationMetricName.SUB_AGENT)
@@ -49,14 +55,6 @@ class SubAgentsEvaluation(Evaluation):
         self._assistant_id_to_tool_info: dict[str, _SubAgentToolInfo] = {}
         self._language_model_service = language_model_service
 
-        for sub_agent_tool in sub_agent_tools:
-            if sub_agent_tool.config.evaluation_config.display_evalution:
-                sub_agent_tool.subscribe(self)
-                self._assistant_id_to_tool_info[sub_agent_tool.config.assistant_id] = {
-                    "assessments": [],
-                    "display_name": sub_agent_tool.display_name(),
-                }
-
     @override
     def get_assessment_type(self) -> ChatMessageAssessmentType:
         return self._config.assessment_type
@@ -71,102 +69,46 @@ class SubAgentsEvaluation(Evaluation):
 
         value = ChatMessageAssessmentLabel.GREEN
 
-        # Use a dict in order to compare labels (RED being the worst)
-        label_comparison_dict = defaultdict(
-            lambda: 3
-        )  # Unkown labels are highest in the sorting
-        label_comparison_dict[ChatMessageAssessmentLabel.GREEN] = 2
-        label_comparison_dict[ChatMessageAssessmentLabel.YELLOW] = 1
-        label_comparison_dict[ChatMessageAssessmentLabel.RED] = 0
-
-        for assistant_id, tool_info in self._assistant_id_to_tool_info.items():
+        for tool_info in self._assistant_id_to_tool_info.values():
             sub_agent_assessments = tool_info["assessments"] or []
-            for i, assessments in enumerate(sub_agent_assessments, start=1):
-                valid_assessments = []
-                for assessment in assessments:
-                    if (
-                        assessment["label"] is None
-                        or assessment["label"] not in ChatMessageAssessmentLabel
-                    ):
-                        logger.warning(
-                            "Unkown assistant label %s for assistant %s will be ignored",
-                            assessment["label"],
-                            assistant_id,
-                        )
-                        continue
-                    if assessment["status"] != ChatMessageAssessmentStatus.DONE:
-                        logger.warning(
-                            "Assessment %s for assistant %s is not done (status: %s) will be ignored",
-                            assessment["label"],
-                            assistant_id,
-                        )
-                        continue
-                    valid_assessments.append(assessment)
+            display_name = tool_info["display_name"]
 
+            for sequence_number in sorted(sub_agent_assessments):
+                assessments = sub_agent_assessments[sequence_number]
+
+                valid_assessments = _get_valid_assessments(
+                    assessments, display_name, sequence_number
+                )
                 if len(valid_assessments) == 0:
                     logger.info(
-                        "No valid assessment found for assistant %s", assistant_id
+                        "No valid assessment found for assistant %s (sequence number: %s)",
+                        display_name,
+                        sequence_number,
                     )
                     continue
 
-                assessments = sorted(
-                    valid_assessments, key=lambda x: label_comparison_dict[x["label"]]
-                )
+                assessments = _sort_assessments(valid_assessments)
+                value = _worst_label(value, assessments[0]["label"])  # type: ignore
 
-                for assessment in assessments:
-                    value = min(
-                        value,
-                        assessment["label"],
-                        key=lambda x: label_comparison_dict[x],
-                    )
                 data = {
                     "name": tool_info["display_name"],
                     "assessments": assessments,
                 }
                 if len(sub_agent_assessments) > 1:
-                    data["name"] += f" {i}"
+                    data["name"] += f" {sequence_number}"
 
                 sub_agents_display_data.append(data)
 
         if len(sub_agents_display_data) == 0:
             logger.warning("No valid sub agent assessments found")
+
             return EvaluationMetricResult(
                 name=self.get_name(),
                 value=NO_ASSESSMENTS_FOUND,
                 reason="No sub agents assessments found",
             )
 
-        should_summarize = False
-        reason = ""
-
-        if len(sub_agents_display_data) > 1:
-            should_summarize = True
-        elif len(sub_agents_display_data) == 1:
-            if len(sub_agents_display_data[0]["assessments"]) > 1:
-                should_summarize = True
-            else:
-                reason = (
-                    sub_agents_display_data[0]["assessments"][0]["explanation"] or ""
-                )
-
-        if should_summarize:
-            messages = (
-                MessagesBuilder()
-                .system_message_append(self._config.summarization_system_message)
-                .user_message_append(
-                    Template(self._config.summarization_user_message_template).render(
-                        sub_agents=sub_agents_display_data,
-                    )
-                )
-                .build()
-            )
-
-            reason = await self._language_model_service.complete_async(
-                messages=messages,
-                model_name=self._config.summarization_model.name,
-                temperature=0.0,
-            )
-            reason = str(reason.choices[0].message.content)
+        reason = await self._get_reason(sub_agents_display_data)
 
         return EvaluationMetricResult(
             name=self.get_name(),
@@ -196,8 +138,30 @@ class SubAgentsEvaluation(Evaluation):
             type=self.get_assessment_type(),
         )
 
+    def register_sub_agent_tool(
+        self, tool: SubAgentTool, evaluation_config: SubAgentEvaluationConfig
+    ) -> None:
+        if not evaluation_config.include_evaluation:
+            logger.warning(
+                "Sub agent tool %s has evaluation config `include_evaluation` set to False, responses will be ignored.",
+                tool.config.assistant_id,
+            )
+            return
+
+        if tool.config.assistant_id not in self._assistant_id_to_tool_info:
+            tool.subscribe(self)
+            self._assistant_id_to_tool_info[tool.config.assistant_id] = (
+                _SubAgentToolInfo(
+                    display_name=tool.display_name(),
+                    assessments={},
+                )
+            )
+
     def notify_sub_agent_response(
-        self, sub_agent_assistant_id: str, response: unique_sdk.Space.Message
+        self,
+        response: unique_sdk.Space.Message,
+        sub_agent_assistant_id: str,
+        sequence_number: int,
     ) -> None:
         if sub_agent_assistant_id not in self._assistant_id_to_tool_info:
             logger.warning(
@@ -206,10 +170,39 @@ class SubAgentsEvaluation(Evaluation):
             )
             return
 
-        self._assistant_id_to_tool_info[sub_agent_assistant_id]["assessments"].append(
+        sub_agent_assessments = self._assistant_id_to_tool_info[sub_agent_assistant_id][
+            "assessments"
+        ]
+        sub_agent_assessments[sequence_number] = (
             response[
                 "assessment"
             ].copy()  # Shallow copy as we don't modify individual assessments
             if response["assessment"] is not None
             else []
         )
+
+    async def _get_reason(self, sub_agents_display_data: list[dict]) -> str:
+        if (
+            len(sub_agents_display_data) == 1
+            and len(sub_agents_display_data[0]["assessments"]) == 1
+        ):
+            return sub_agents_display_data[0]["assessments"][0]["explanation"] or ""
+
+        messages = (
+            MessagesBuilder()
+            .system_message_append(self._config.summarization_system_message)
+            .user_message_append(
+                Template(self._config.summarization_user_message_template).render(
+                    sub_agents=sub_agents_display_data,
+                )
+            )
+            .build()
+        )
+
+        reason = await self._language_model_service.complete_async(
+            messages=messages,
+            model_name=self._config.summarization_model.name,
+            temperature=0.0,
+        )
+
+        return str(reason.choices[0].message.content)

{unique_toolkit-1.7.0 → unique_toolkit-1.8.1}/unique_toolkit/agentic/tools/a2a/manager.py

@@ -1,8 +1,8 @@
 from logging import Logger
 
-from unique_toolkit.agentic.tools.a2a.config import SubAgentToolConfig
-from unique_toolkit.agentic.tools.a2a.service import SubAgentTool, ToolProgressReporter
+from unique_toolkit.agentic.tools.a2a.tool import SubAgentTool, SubAgentToolConfig
 from unique_toolkit.agentic.tools.config import ToolBuildConfig
+from unique_toolkit.agentic.tools.tool_progress_reporter import ToolProgressReporter
 from unique_toolkit.app.schemas import ChatEvent
 
 

unique_toolkit-1.8.1/unique_toolkit/agentic/tools/a2a/postprocessing/__init__.py

@@ -0,0 +1,13 @@
+from unique_toolkit.agentic.tools.a2a.postprocessing.config import (
+    SubAgentDisplayConfig,
+    SubAgentResponseDisplayMode,
+)
+from unique_toolkit.agentic.tools.a2a.postprocessing.postprocessor import (
+    SubAgentResponsesPostprocessor,
+)
+
+__all__ = [
+    "SubAgentResponsesPostprocessor",
+    "SubAgentResponseDisplayMode",
+    "SubAgentDisplayConfig",
+]

unique_toolkit-1.7.0/unique_toolkit/agentic/tools/a2a/postprocessing/display.py → unique_toolkit-1.8.1/unique_toolkit/agentic/tools/a2a/postprocessing/_display.py

@@ -2,7 +2,9 @@ import re
 from abc import ABC, abstractmethod
 from typing import Literal, override
 
-from unique_toolkit.agentic.tools.a2a.config import ResponseDisplayMode
+from unique_toolkit.agentic.tools.a2a.postprocessing.config import (
+    SubAgentResponseDisplayMode,
+)
 
 
 class _ResponseDisplayHandler(ABC):
@@ -84,13 +86,20 @@ class _DetailsResponseDisplayHandler(_ResponseDisplayHandler):
 
 
 _DISPLAY_HANDLERS = {
-    ResponseDisplayMode.DETAILS_OPEN: _DetailsResponseDisplayHandler(mode="open"),
-    ResponseDisplayMode.DETAILS_CLOSED: _DetailsResponseDisplayHandler(mode="closed"),
+    SubAgentResponseDisplayMode.DETAILS_OPEN: _DetailsResponseDisplayHandler(
+        mode="open"
+    ),
+    SubAgentResponseDisplayMode.DETAILS_CLOSED: _DetailsResponseDisplayHandler(
+        mode="closed"
+    ),
 }
 
 
-def build_sub_agent_answer_display(
-    display_name: str, display_mode: ResponseDisplayMode, answer: str, assistant_id: str
+def _build_sub_agent_answer_display(
+    display_name: str,
+    display_mode: SubAgentResponseDisplayMode,
+    answer: str,
+    assistant_id: str,
 ) -> str:
     if display_mode not in _DISPLAY_HANDLERS:
         return ""
@@ -102,8 +111,8 @@ def build_sub_agent_answer_display(
     )
 
 
-def remove_sub_agent_answer_from_text(
-    display_mode: ResponseDisplayMode, text: str, assistant_id: str
+def _remove_sub_agent_answer_from_text(
+    display_mode: SubAgentResponseDisplayMode, text: str, assistant_id: str
 ) -> str:
     if display_mode not in _DISPLAY_HANDLERS:
         return text

unique_toolkit-1.8.1/unique_toolkit/agentic/tools/a2a/postprocessing/_utils.py

@@ -0,0 +1,19 @@
+def _replace_references_in_text_non_overlapping(
+    text: str, ref_map: dict[int, int]
+) -> str:
+    for orig, repl in ref_map.items():
+        text = text.replace(f"<sup>{orig}</sup>", f"<sup>{repl}</sup>")
+    return text
+
+
+def _replace_references_in_text(text: str, ref_map: dict[int, int]) -> str:
+    # 2 phase replacement, since the map keys and values can overlap
+    max_ref = max(max(ref_map.keys(), default=0), max(ref_map.values(), default=0)) + 1
+    unique_refs = range(max_ref, max_ref + len(ref_map))
+
+    text = _replace_references_in_text_non_overlapping(
+        text, dict(zip(ref_map.keys(), unique_refs))
+    )
+    return _replace_references_in_text_non_overlapping(
+        text, dict(zip(unique_refs, ref_map.values()))
+    )

unique_toolkit-1.8.1/unique_toolkit/agentic/tools/a2a/postprocessing/config.py

@@ -0,0 +1,24 @@
+from enum import StrEnum
+
+from pydantic import BaseModel, Field
+
+from unique_toolkit._common.pydantic_helpers import get_configuration_dict
+
+
+class SubAgentResponseDisplayMode(StrEnum):
+    HIDDEN = "hidden"
+    DETAILS_OPEN = "details_open"
+    DETAILS_CLOSED = "details_closed"
+
+
+class SubAgentDisplayConfig(BaseModel):
+    model_config = get_configuration_dict()
+
+    mode: SubAgentResponseDisplayMode = Field(
+        default=SubAgentResponseDisplayMode.HIDDEN,
+        description="Controls how to display the sub agent response.",
+    )
+    remove_from_history: bool = Field(
+        default=True,
+        description="If set, sub agent responses will be removed from the history on subsequent calls to the assistant.",
+    )