refactor(negotiation)!: remove deprecated state-machine negotiation APIs

README.md

@@ -170,7 +170,7 @@ Confirm the following limitations before use:
 
 - The built-in LLM call chain is uniformly exposed as an OpenAI adaptation layer.
 - The LLM prompt resources are built in and do not support custom extension (`prompts` and `errors` are always read from the built-in resources).
-- Negotiation session state travels in the message metadata (the SDK is stateless) and no negotiation state store is provided; the legacy state-machine negotiation APIs (`start_negotiation` / `receive_negotiation` / `continue_negotiation`) are deprecated since 1.1.0.
+- Negotiation session state travels in the message metadata (the SDK is stateless) and no negotiation state store is provided; the legacy state-machine negotiation APIs (`start_negotiation` / `receive_negotiation` / `continue_negotiation`) were deprecated since v1.1.0 and removed in v1.2.0.
 - Bundled resources and language coverage are limited, and remote resource loading such as `registry-center` (the registry center) is not included.
 - This document mainly introduces the SDK itself and does not cover CLI tools, hosted services, deployment flows, or ready-to-use application solutions.
 

README_zh.md

@@ -168,7 +168,7 @@ uv run mypy src      # 类型检查
 
 - 内置 LLM 调用链对外统一为 OpenAI 适配层。
 - 大模型提示词资源已内置，不支持自定义扩展（`prompts` 与 `errors` 始终从内置资源加载）。
-- 协商会话状态随消息 metadata 往返（SDK 无状态），不提供协商状态存储；旧的状态机协商 API（`start_negotiation` / `receive_negotiation` / `continue_negotiation`）自 1.1.0 起废弃。
+- 协商会话状态随消息 metadata 往返（SDK 无状态），不提供协商状态存储；旧的状态机协商 API（`start_negotiation` / `receive_negotiation` / `continue_negotiation`）自 v1.1.0 起废弃，已在 v1.2.0 版本删除。
 - 随包资源与语言覆盖有限，不包含 `registry-center`（注册中心）等远程资源加载能力。
 - 本文档主要介绍 SDK 本身，不涉及 CLI、托管服务、部署流程或可直接使用的应用方案。
 

a2a-t-sample/env.example

@@ -59,6 +59,3 @@ A2AT_LLM_SESSION_MAX_PER_PROVIDER=100
 # the dedicated logger a2a_t.llm.call is enabled for DEBUG; where the lines are printed is
 # decided by the embedding application's logging framework.
 A2AT_LLM_DETAIL_LOG_ENABLED=false
-
-# Negotiation
-A2AT_NEGOTIATION_STATE_STORE_TYPE=in_memory

docs/en/API_Reference.md

@@ -72,7 +72,7 @@ The A2A-T SDK exposes only two entry points: the client entry `A2ATClient` (prom
 
 - Some APIs involve LLM calls. When using them, control the call frequency and concurrency based on the concurrency the connected model service can provide and your business latency requirements.
 - APIs involving LLM calls apply length protection to `text` inputs: when the input exceeds `A2AT_INPUT_TEXT_MAX_CHARS` configured in `.env` (the `package_data/.env` or the file pointed to by `env_path`), the error code `input.text_too_long` is reported; the default of this configuration item is 16384 (16×1024). Structured data that does not involve LLM calls is not limited.
-- The SDK is stateless: negotiation session state travels with the A2A-T metadata (`negotiationContext`) of each message and is not stored inside the SDK; the legacy `start_negotiation` / `receive_negotiation` / `continue_negotiation` state-machine negotiation APIs are deprecated since 1.1.0, emit a `DeprecationWarning` when called, and will be removed in the next release.
+- The SDK is stateless: negotiation session state travels with the A2A-T metadata (`negotiationContext`) of each message and is not stored inside the SDK; the legacy `start_negotiation` / `receive_negotiation` / `continue_negotiation` state-machine negotiation APIs were deprecated since v1.1.0, emitted a `DeprecationWarning` when called, and were removed in v1.2.0.
 
 ## 1.3 API Descriptions
 

docs/en/developer_guide.md

@@ -87,7 +87,7 @@ flowchart TD
 1. Python 3.12+ is required.
 2. Complete multi-agent protocol interaction development also requires the `a2a-sdk`, version 1.1.0+.
 3. The SDK is stateless: negotiation session state travels with the A2A-T metadata (`negotiationContext`) of each message and is not stored inside the SDK.
-4. The legacy state-machine negotiation demo (`start_negotiation` / `receive_negotiation` / `continue_negotiation` and the `negotiation/{types,store,runtime,handling}` packages) is deprecated since 1.1.0 (calls emit a `DeprecationWarning`) and will be removed in the next release.
+4. The legacy state-machine negotiation demo (`start_negotiation` / `receive_negotiation` / `continue_negotiation` and the `negotiation/{types,store,runtime,handling}` packages): the state-machine negotiation APIs were deprecated since v1.1.0, emitted a `DeprecationWarning` when called, and were removed in v1.2.0.
 5. The A2A-T SDK does not provide an agent HTTP service framework, a registry-center client, or authentication and key management capabilities; these must be integrated by the business system.
 
 ## 1.3 Environment Preparation
@@ -1521,4 +1521,3 @@ The sample configuration file provided by the SDK is `env.example`; copy it to `
 | `A2AT_LLM_SESSION_MAX_PER_PROVIDER`   | Maximum number of tracked sessions per provider, default `100`                           |
 | `A2AT_LLM_MAX_ATTEMPTS`               | Maximum number of attempts for retryable LLM steps, range 1-10 (out-of-range values are clipped), default `3` |
 | `A2AT_LLM_DETAIL_LOG_ENABLED`         | Whether to print the complete LLM request/response messages (not truncated), default `false`; enabling it may print sensitive information or consume log space, use only in the DEBUG phase and keep it off in production; time/token/elapsed-time summary logs are emitted at DEBUG level by the dedicated logger `a2a_t.llm.call`, see 1.6 |
-| `A2AT_NEGOTIATION_STATE_STORE_TYPE`   | Negotiation state store of the deprecated state-machine negotiation demo (`in_memory`); the 1.1.0 negotiation content API is stateless and does not use this key |

docs/zh/API参考.md

@@ -72,7 +72,7 @@ A2A-T SDK 对外仅提供两个入口：客户端入口 `A2ATClient`（提示词
 
 - 部分接口涉及到LLM调用，使用时需根据对所接模型服务可提供的并发能力，结合业务时限要求，控制使用API时的调用频率和并发。
 - 涉及到LLM调用的API针对文本`text`输入会做长度防护，在输入超过`.env`（`package_data/.env` 或 `env_path` 指向的文件）中配置的 `A2AT_INPUT_TEXT_MAX_CHARS` 字符时，会报错误码 `input.text_too_long` ，该配置项默认配置为 16384（16×1024）；不涉及LLM调用的结构化数据不受此限制。
-- SDK 无状态：协商会话状态随每条消息的 A2A-T metadata（`negotiationContext`）往返，不保存在 SDK 内部；旧的 `start_negotiation` / `receive_negotiation` / `continue_negotiation` 状态机协商 API 自 1.1.0 起废弃，调用时发出 `DeprecationWarning`，将在下一版本删除。
+- SDK 无状态：协商会话状态随每条消息的 A2A-T metadata（`negotiationContext`）往返，不保存在 SDK 内部；旧的 `start_negotiation` / `receive_negotiation` / `continue_negotiation` 状态机协商 API 自 v1.1.0 起废弃，调用时发出 `DeprecationWarning`，已在 v1.2.0 版本删除。
 
 ## 1.3 API说明
 

docs/zh/开发指南.md

@@ -87,6 +87,7 @@ flowchart TD
 1. Python 版本要求为 3.12+。
 2. 完整的多智能体协议交互开发需要同时引入`a2a-sdk`，要求版本为 1.1.0+。
 3. SDK 无状态：协商会话状态随每条消息的 A2A-T metadata（`negotiationContext`）往返，不保存在 SDK 内部。
+4. 旧的状态机协商 demo（`start_negotiation` / `receive_negotiation` / `continue_negotiation` 以及 `negotiation/{types,store,runtime,handling}` 包）：状态机协商 API 自 v1.1.0 起废弃，调用时发出 `DeprecationWarning`，已在 v1.2.0 版本删除。
 5. A2A-T SDK 不提供 Agent HTTP 服务框架、注册中心客户端、鉴权和密钥管理能力，需由业务系统集成。
 
 ## 1.3 环境准备
@@ -1512,4 +1513,3 @@ SDK提供的样例配置文件为`env.example`，复制为 `package_data/.env`
 | `A2AT_LLM_SESSION_MAX_PER_PROVIDER`   | 单提供方会话跟踪数上限，默认 `100`                           |
 | `A2AT_LLM_MAX_ATTEMPTS`               | 可重试 LLM 步骤的最大尝试次数，取值范围 1-10（越界自动裁剪），默认 `3` |
 | `A2AT_LLM_DETAIL_LOG_ENABLED`         | 是否打印 LLM 请求/响应的完整报文（不截断），默认 `false`；开启后可能打印敏感信息或占用日志空间，仅限 DEBUG 阶段使用、正式运行必须关闭；时间/Token/耗时等摘要日志由专用 logger `a2a_t.llm.call` 以 DEBUG 级输出，详见 1.6 |
-| `A2AT_NEGOTIATION_STATE_STORE_TYPE`   | 已废弃状态机协商 demo 的协商状态存储（`in_memory`）；1.1.0 协商内容 API 无状态，不使用该键 |

env.example

@@ -59,10 +59,6 @@ A2AT_LLM_MAX_ATTEMPTS=3
 # decided by the embedding application's logging framework.
 A2AT_LLM_DETAIL_LOG_ENABLED=false
 
-# Negotiation
-# Selects the negotiation state store implementation (legacy state-machine demo, deprecated).
-A2AT_NEGOTIATION_STATE_STORE_TYPE=in_memory
-
 # Live-LLM corpus tests (dev only; fully decoupled from the production A2AT_LLM_* keys)
 # Base URL of the live endpoint used by the live corpus family; the family skips when unset.
 A2AT_TEST_LLM_BASE_URL=

package_data/env.example

@@ -57,7 +57,3 @@ A2AT_LLM_MAX_ATTEMPTS=3
 # the dedicated logger a2a_t.llm.call is enabled for DEBUG; where the lines are printed is
 # decided by the embedding application's logging framework.
 A2AT_LLM_DETAIL_LOG_ENABLED=false
-
-# Negotiation
-# Selects the negotiation state store implementation (legacy state-machine demo, deprecated).
-A2AT_NEGOTIATION_STATE_STORE_TYPE=in_memory

src/a2a_t/client/a2at_client.py

@@ -6,16 +6,12 @@
 twelve negotiation content methods
 of the shared :class:`~a2a_t.negotiation.generation.content_service.NegotiationContentService`
 (eight generation plus four validation methods) and the extension-agnostic template queries of the
-:class:`~a2a_t.common.prompt_resources.catalog.TemplateQueryService`. The three legacy
-``start/receive/continue_negotiation`` methods of the deprecated state-machine negotiation demo
-are kept for one release with a :class:`DeprecationWarning` and will be removed in the next
-release (port decision D1).
+:class:`~a2a_t.common.prompt_resources.catalog.TemplateQueryService`.
 """
 
 from __future__ import annotations
 
 import logging
-import warnings
 from collections.abc import Mapping
 from pathlib import Path
 from typing import Any
@@ -27,7 +23,6 @@
 from a2a_t.core.validation_pipeline import FilledParamData
 from a2a_t.llm.config_loader import LLMConfigLoader
 from a2a_t.llm.factory import LLMClientFactory
-from a2a_t.negotiation.common.models import ContinueNegotiationInput, StartNegotiationInput
 from a2a_t.negotiation.content.models import (
     NegotiationAbortData,
     NegotiationEndingData,
@@ -37,7 +32,6 @@
 from a2a_t.observability.llm_decorator import A2ATLLMClientDecorator
 from a2a_t.observability.negotiation_metrics import report_terminal_negotiation_rounds
 
-from .negotiation.negotiation_orchestrator_builder import ClientNegotiationOrchestratorBuilder
 from .prompt_generation.models import PromptGenerationResult
 from .prompt_generation.prompt_generation_orchestrator_builder import PromptGenerationOrchestratorBuilder
 
@@ -60,17 +54,6 @@ def _default_env_path() -> Path:
     return Path(__file__).resolve().parents[3] / "package_data" / ".env"
 
 
-def _deprecation_message(method: str) -> str:
-    """Return the D1 deprecation message of one legacy negotiation facade method."""
-    return (
-        f"A2ATClient.{method} is deprecated since 1.1.0 and will be removed in the next release; "
-        "the state-machine negotiation demo it belongs to is retired. Use the negotiation content "
-        "API instead: generate_negotiation_{propose,accept,reject,abort}_prompt_from_data / "
-        "..._from_text and validate_{propose,accept,reject,abort}_prompt_and_data_filling, backed "
-        "by a2a_t.negotiation.generation.NegotiationContentService."
-    )
-
-
 def _parse_template_uri(template_uri: str | TemplateUri | None) -> TemplateUri:
     """Parse the raw template URI string boundary fail-fast (D16, Java ``parseTemplateUri``).
 
@@ -111,10 +94,6 @@ class A2ATClient:
     hand-written URI strings. Business failures are raised as catalog-coded exceptions of the
     :mod:`a2a_t.core.errors.exceptions` tree (catch :class:`~a2a_t.core.errors.exceptions.A2ATError`
     and branch on ``code_str``); programming errors stay ``TypeError`` / ``ValueError``.
-
-    The three legacy ``start/receive/continue_negotiation`` methods of the retired state-machine
-    negotiation demo are kept for one release with a :class:`DeprecationWarning` and will be
-    removed in the next release.
     """
 
     def __init__(
@@ -125,18 +104,18 @@ def __init__(
     ) -> None:
         """Create one client facade from a ``.env`` file.
 
-        The configuration, the LLM client and both orchestrators are resolved eagerly so that a
-        misconfiguration surfaces at construction instead of on the first call. The negotiation
-        content service and the template query service are assembled lazily on first use: both
-        capture resource snapshots at that point, so constructing the facade stays side-effect free
-        for the prompt-only flows.
+        The configuration and the LLM client are resolved eagerly so that a misconfiguration
+        surfaces at construction instead of on the first call. The negotiation content service and
+        the template query service are assembled lazily on first use: both capture resource
+        snapshots at that point, so constructing the facade stays side-effect free for the
+        prompt-only flows.
 
         Args:
             env_path: path of the ``.env`` file carrying the SDK configuration; ``None`` uses the
                 packaged ``package_data/.env`` default. The resolved file must exist — copy
                 ``env.example`` to ``.env`` first (every key is optional, an empty file works).
-            logger: optional logger injected into the LLM client and the orchestrators; ``None``
-                uses each component's module-level logger.
+            logger: optional logger injected into the LLM client and the prompt generation
+                orchestrator; ``None`` uses each component's module-level logger.
 
         Raises:
             ConfigFileNotFoundError: when the resolved ``.env`` path does not exist.
@@ -155,10 +134,6 @@ def __init__(
             llm_client=self._llm_client,
             logger=logger,
         )
-        self._negotiation_orchestrator = ClientNegotiationOrchestratorBuilder().build(
-            env_path=resolved_env_path,
-            logger=logger,
-        )
         # Lazily assembled negotiation content service and template query service: both capture
         # resource snapshots at first use, so constructing the facade stays side-effect free for
         # the prompt-only flows. Tests may inject either seam before the first call.
@@ -381,57 +356,6 @@ def generate_notification_prompt_from_data_with_schema(
             data, schema, _parse_template_uri(template_uri)
         )
 
-    def start_negotiation(self, input: StartNegotiationInput) -> dict[str, object]:
-        """Start a client-side negotiation round.
-
-        Deprecated (D1): the state-machine negotiation demo is retired; this method emits a
-        :class:`DeprecationWarning`, forwards to the legacy orchestrator unchanged, and will be
-        removed in the next release. Use the negotiation content API instead —
-        :meth:`generate_negotiation_propose_prompt_from_data` and its siblings.
-
-        Args:
-            input: legacy negotiation start input.
-
-        Returns:
-            the legacy orchestrator's round result as a plain mapping.
-        """
-        warnings.warn(_deprecation_message("start_negotiation"), DeprecationWarning, stacklevel=2)
-        return self._negotiation_orchestrator.start_negotiation(input)
-
-    def receive_negotiation(self, message: str, context: dict[str, object]) -> dict[str, object]:
-        """Process a negotiation message received from the remote peer.
-
-        Deprecated (D1): the state-machine negotiation demo is retired; this method forwards to the
-        legacy orchestrator unchanged and will be removed in the next release. Use the negotiation
-        content API instead — :meth:`validate_propose_prompt_and_data_filling` and its siblings.
-
-        Args:
-            message: received negotiation message text.
-            context: legacy negotiation context mapping.
-
-        Returns:
-            the legacy orchestrator's round result as a plain mapping.
-        """
-        warnings.warn(_deprecation_message("receive_negotiation"), DeprecationWarning, stacklevel=2)
-        return self._negotiation_orchestrator.receive_negotiation(message, context)
-
-    def continue_negotiation(self, input: ContinueNegotiationInput) -> dict[str, object]:
-        """Continue an existing negotiation with a local response.
-
-        Deprecated (D1): the state-machine negotiation demo is retired; this method forwards to the
-        legacy orchestrator unchanged and will be removed in the next release. Use the negotiation
-        content API instead — :meth:`generate_negotiation_accept_prompt_from_data` and its
-        siblings.
-
-        Args:
-            input: legacy negotiation continuation input.
-
-        Returns:
-            the legacy orchestrator's round result as a plain mapping.
-        """
-        warnings.warn(_deprecation_message("continue_negotiation"), DeprecationWarning, stacklevel=2)
-        return self._negotiation_orchestrator.continue_negotiation(input)
-
     # ------------------------------------------------------------------
     # Negotiation content generation (from data: deterministic, zero LLM)
     # ------------------------------------------------------------------