wagent-framework 2.0.0a1__tar.gz → 2.0.0a2__tar.gz

{wagent_framework-2.0.0a1 → wagent_framework-2.0.0a2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: wagent-framework
-Version: 2.0.0a1
+Version: 2.0.0a2
 Summary: Open, composable Python framework for building agents
 Home-page: https://github.com/LuckyStar2456/W-Agent-FrameWork
 Author: LuckyStar2456
@@ -73,7 +73,7 @@ Dynamic: summary
 
 W-Agent 是一个面向本地开发者的 Python 开源 Agent 开发框架。它的目标不是提供托管平台或固定 Harness，而是提供稳定、可扩展的协议与可自由装配的模块，让开发者能够替换模型、路由、Agent Loop、Workflow、工具、状态、沙箱和界面实现。
 
-当前稳定发布版本是 `1.5.2`；仓库主线已进入 `2.0.0a1` 开发阶段。1.x 工程底座继续保留，微内核、模型、工具、单 Agent 与本地 Workflow 基础已经实现，其余下一代能力按路线图分阶段交付。文档使用明确状态标记，避免把规划能力描述为现有能力。
+当前稳定发布版本是 `1.5.2`，最新 Alpha 为 `2.0.0a2`。1.x 工程底座继续保留，微内核、模型、工具、单 Agent 与本地 Workflow 基础已经实现，其余下一代能力按路线图分阶段交付。文档使用明确状态标记，避免把规划能力描述为现有能力。
 
 ## 状态标记
 
@@ -109,7 +109,7 @@ W-Agent 遵循以下原则：
 - Skill 加载、签名校验、MCP JWT 认证、Redis 分布式锁。
 - LangChain 工具适配、FastAPI 示例集成和测试辅助设施。
 
-以下 Phase 1 能力在当前 `2.0.0a1` 源码中为 `Implemented`：
+以下 Phase 1 能力在当前 `2.0.0a2` 源码中为 `Implemented`：
 
 - `PluginSpec`、装饰器、YAML 引用与 Python entry point 发现。
 - 统一、版本感知、分作用域的能力注册表与不可变快照。
@@ -131,12 +131,13 @@ W-Agent 遵循以下原则：
 - 注册即安全探测的 `ModelRegistrationProbeService` 与外部路由健康桥接；直接调用 `ModelRegistry.register()` 仍保持无副作用。
 - 工具 Definition/Binding/Registry 分层、Python/HTTP/无 Shell 命令模板、MCP 绑定与 2026-07-28 stdio/Streamable HTTP 客户端、参数校验、权限/逐调用审批、超时/取消和 Prompt-free 审计。
 - 可替换 `AgentLoop` 协议与有界单 Agent `ReactAgentLoop`，覆盖模型→工具→结果→模型闭环、严格 Token 预算、JSONL RunEvent/尝试账本记录和审批断点恢复。
+- 应用提供的版本化价格表、普通/缓存输入与输出费用计量、失败关闭的 Run 费用预算，以及 Session/Checkpoint/评测费用可见性。
 - 统一 `WorkflowRegistry`、可替换 `WorkflowEngineProtocol` 与 `LocalWorkflowEngine`，支持静态 DAG、状态图、Python 入口、节点事件、取消，以及内存/JSONL 节点边界暂停恢复。
 - 统一 `SandboxProvider`/`SandboxRegistry`、Docker/OCI 生命周期后端、显式运行时授权的 `UnsafeLocalSandboxProvider`，以及受工具策略保护的 `sandbox_command_tool()`。
 - Agent/Workflow 双向适配器，以及可完全覆盖的客服/RAG 与编码 Agent 模板。
 - `CompositionManifest` 的确定性编码、安全预览，以及冲突安全的本地版本和别名管理。
 - 本地 Session 创建/列表/归档、JSON 持久化、跨 Run 文本上下文与审批恢复协调。
-- 严格本地 JSON 装配的文本 Agent CLI/TUI 运行入口，使用环境变量凭据引用、逐次调用确认和可见 Token 预算/计量。
+- 严格本地 JSON 装配的文本 Agent CLI/TUI 运行入口，使用环境变量凭据引用、逐次调用确认、可见 Token/费用预算与计量。
 - 确定性脚本化 Model Provider、显式授权的 JSONL 录制/顺序回放，以及本地评测运行器与脱敏 JSON 报告。
 
 当前 `BaseAgent` 仍是简单的 1.x 抽象；`LegacyAgentAdapter` 已能把它严格桥接为文本 Workflow 节点，新的 ReAct Runtime 独立提供。专用 OpenAI Responses 与 vLLM 差异适配、模型跨流恢复、多模态/工具事件通用回放、并行或嵌套 Workflow、装配依赖安装/加载确认和完整交互 TUI 仍为 `Planned`，不能当作现成功能使用。Docker 后端已有模拟 CLI 生命周期测试，但不代表当前机器已安装或启动 Docker；厂商模板经过模拟传输测试，也不代表所有远程型号已经在线验证。
@@ -159,6 +160,12 @@ W-Agent 遵循以下原则：
 pip install wagent-framework
 ```
 
+安装最新 Alpha：
+
+```bash
+pip install --pre wagent-framework==2.0.0a2
+```
+
 可选依赖：
 
 ```bash
@@ -168,7 +175,7 @@ pip install "wagent-framework[wasm]"
 pip install "wagent-framework[tui]"
 ```
 
-PyPI 1.x 支持 Python 3.9+；当前 `2.0.0a1` 源码要求 Python 3.11+。
+PyPI 1.x 支持 Python 3.9+；`2.0.0a2` 与后续 2.x Alpha 要求 Python 3.11+。
 
 ## 1.x 最小示例
 

{wagent_framework-2.0.0a1 → wagent_framework-2.0.0a2}/README.md

@@ -4,7 +4,7 @@
 
 W-Agent 是一个面向本地开发者的 Python 开源 Agent 开发框架。它的目标不是提供托管平台或固定 Harness，而是提供稳定、可扩展的协议与可自由装配的模块，让开发者能够替换模型、路由、Agent Loop、Workflow、工具、状态、沙箱和界面实现。
 
-当前稳定发布版本是 `1.5.2`；仓库主线已进入 `2.0.0a1` 开发阶段。1.x 工程底座继续保留，微内核、模型、工具、单 Agent 与本地 Workflow 基础已经实现，其余下一代能力按路线图分阶段交付。文档使用明确状态标记，避免把规划能力描述为现有能力。
+当前稳定发布版本是 `1.5.2`，最新 Alpha 为 `2.0.0a2`。1.x 工程底座继续保留，微内核、模型、工具、单 Agent 与本地 Workflow 基础已经实现，其余下一代能力按路线图分阶段交付。文档使用明确状态标记，避免把规划能力描述为现有能力。
 
 ## 状态标记
 
@@ -40,7 +40,7 @@ W-Agent 遵循以下原则：
 - Skill 加载、签名校验、MCP JWT 认证、Redis 分布式锁。
 - LangChain 工具适配、FastAPI 示例集成和测试辅助设施。
 
-以下 Phase 1 能力在当前 `2.0.0a1` 源码中为 `Implemented`：
+以下 Phase 1 能力在当前 `2.0.0a2` 源码中为 `Implemented`：
 
 - `PluginSpec`、装饰器、YAML 引用与 Python entry point 发现。
 - 统一、版本感知、分作用域的能力注册表与不可变快照。
@@ -62,12 +62,13 @@ W-Agent 遵循以下原则：
 - 注册即安全探测的 `ModelRegistrationProbeService` 与外部路由健康桥接；直接调用 `ModelRegistry.register()` 仍保持无副作用。
 - 工具 Definition/Binding/Registry 分层、Python/HTTP/无 Shell 命令模板、MCP 绑定与 2026-07-28 stdio/Streamable HTTP 客户端、参数校验、权限/逐调用审批、超时/取消和 Prompt-free 审计。
 - 可替换 `AgentLoop` 协议与有界单 Agent `ReactAgentLoop`，覆盖模型→工具→结果→模型闭环、严格 Token 预算、JSONL RunEvent/尝试账本记录和审批断点恢复。
+- 应用提供的版本化价格表、普通/缓存输入与输出费用计量、失败关闭的 Run 费用预算，以及 Session/Checkpoint/评测费用可见性。
 - 统一 `WorkflowRegistry`、可替换 `WorkflowEngineProtocol` 与 `LocalWorkflowEngine`，支持静态 DAG、状态图、Python 入口、节点事件、取消，以及内存/JSONL 节点边界暂停恢复。
 - 统一 `SandboxProvider`/`SandboxRegistry`、Docker/OCI 生命周期后端、显式运行时授权的 `UnsafeLocalSandboxProvider`，以及受工具策略保护的 `sandbox_command_tool()`。
 - Agent/Workflow 双向适配器，以及可完全覆盖的客服/RAG 与编码 Agent 模板。
 - `CompositionManifest` 的确定性编码、安全预览，以及冲突安全的本地版本和别名管理。
 - 本地 Session 创建/列表/归档、JSON 持久化、跨 Run 文本上下文与审批恢复协调。
-- 严格本地 JSON 装配的文本 Agent CLI/TUI 运行入口，使用环境变量凭据引用、逐次调用确认和可见 Token 预算/计量。
+- 严格本地 JSON 装配的文本 Agent CLI/TUI 运行入口，使用环境变量凭据引用、逐次调用确认、可见 Token/费用预算与计量。
 - 确定性脚本化 Model Provider、显式授权的 JSONL 录制/顺序回放，以及本地评测运行器与脱敏 JSON 报告。
 
 当前 `BaseAgent` 仍是简单的 1.x 抽象；`LegacyAgentAdapter` 已能把它严格桥接为文本 Workflow 节点，新的 ReAct Runtime 独立提供。专用 OpenAI Responses 与 vLLM 差异适配、模型跨流恢复、多模态/工具事件通用回放、并行或嵌套 Workflow、装配依赖安装/加载确认和完整交互 TUI 仍为 `Planned`，不能当作现成功能使用。Docker 后端已有模拟 CLI 生命周期测试，但不代表当前机器已安装或启动 Docker；厂商模板经过模拟传输测试，也不代表所有远程型号已经在线验证。
@@ -90,6 +91,12 @@ W-Agent 遵循以下原则：
 pip install wagent-framework
 ```
 
+安装最新 Alpha：
+
+```bash
+pip install --pre wagent-framework==2.0.0a2
+```
+
 可选依赖：
 
 ```bash
@@ -99,7 +106,7 @@ pip install "wagent-framework[wasm]"
 pip install "wagent-framework[tui]"
 ```
 
-PyPI 1.x 支持 Python 3.9+；当前 `2.0.0a1` 源码要求 Python 3.11+。
+PyPI 1.x 支持 Python 3.9+；`2.0.0a2` 与后续 2.x Alpha 要求 Python 3.11+。
 
 ## 1.x 最小示例
 

{wagent_framework-2.0.0a1 → wagent_framework-2.0.0a2}/setup.py

@@ -8,7 +8,7 @@ with open(os.path.join(here, "README.md"), "r", encoding="utf-8") as f:
 
 setup(
     name="wagent-framework",
-    version="2.0.0a1",
+    version="2.0.0a2",
     description="Open, composable Python framework for building agents",
     long_description=long_description,
     long_description_content_type="text/markdown",

{wagent_framework-2.0.0a1 → wagent_framework-2.0.0a2}/tests/test_cli.py

@@ -10,6 +10,7 @@ from w_agent import (
     ModelCapability,
     ModelDescriptor,
     ModelMessage,
+    ModelCost,
     RunResult,
     StopReason,
     TokenUsage,
@@ -24,7 +25,7 @@ def test_cli_version_and_profile_json_are_machine_readable():
     profiles = runner.invoke(app, ["profile", "list", "--json"])
 
     assert version.exit_code == 0
-    assert "2.0.0a1" in version.stdout
+    assert "2.0.0a2" in version.stdout
     assert profiles.exit_code == 0
     assert [item["key"] for item in json.loads(profiles.stdout)] == [
         "customer-support",
@@ -332,6 +333,8 @@ def test_cli_evaluate_runs_dataset_and_writes_private_report(tmp_path, monkeypat
                 usage=TokenUsage(3, 2),
                 model_calls=1,
                 reported_usage_calls=1,
+                cost=ModelCost("USD", "prices-v1", "0.01", "0.02"),
+                priced_usage_calls=1,
             )
             return SimpleNamespace(result=result)
 
@@ -364,6 +367,8 @@ def test_cli_evaluate_runs_dataset_and_writes_private_report(tmp_path, monkeypat
     assert payload["usage"]["input_tokens"] == 3
     assert payload["usage"]["output_tokens"] == 2
     assert payload["usage_complete"] is True
+    assert payload["cost"]["total"] == "0.03"
+    assert payload["cost_complete"] is True
     assert "output" not in payload["cases"][0]
     assert "secret prompt" not in persisted
     assert "VISIBLE_OUTPUT" not in persisted

{wagent_framework-2.0.0a1 → wagent_framework-2.0.0a2}/tests/test_evaluation.py

@@ -10,6 +10,7 @@ from w_agent import (
     JsonEvaluationReporter,
     LocalEvaluationRunner,
     MessageRole,
+    ModelCost,
     ModelMessage,
     RunEvent,
     RunEventType,
@@ -21,7 +22,7 @@ from w_agent import (
 )
 
 
-def _result(name, output, *, reason=StopReason.COMPLETED, events=()):
+def _result(name, output, *, reason=StopReason.COMPLETED, events=(), cost=None):
     return RunResult(
         f"run-{name}",
         reason,
@@ -33,6 +34,8 @@ def _result(name, output, *, reason=StopReason.COMPLETED, events=()):
         usage=TokenUsage(4, 2),
         model_calls=1,
         reported_usage_calls=1,
+        cost=cost,
+        priced_usage_calls=1 if cost is not None else 0,
     )
 
 
@@ -64,6 +67,29 @@ async def test_local_evaluation_scores_cases_and_aggregates_usage():
     assert report.cases[1].passed is False
 
 
+@pytest.mark.asyncio
+async def test_local_evaluation_aggregates_versioned_cost_metrics():
+    async def target(case):
+        return _result(
+            case.name,
+            "ok",
+            cost=ModelCost("USD", "prices-v1", "0.01", "0.02", "0.003"),
+        )
+
+    report = await LocalEvaluationRunner().run(
+        (EvaluationCase("one", "a"), EvaluationCase("two", "b")),
+        target,
+    )
+    payload = evaluation_report_to_dict(report)
+
+    assert report.cost is not None
+    assert str(report.cost.total) == "0.066"
+    assert report.cost_complete is True
+    assert payload["cost"]["total"] == "0.066"
+    assert payload["cost_complete"] is True
+    assert payload["cases"][0]["cost"]["price_table_version"] == "prices-v1"
+
+
 @pytest.mark.asyncio
 async def test_evaluation_report_omits_prompts_outputs_and_exception_messages(tmp_path):
     cases = (EvaluationCase("failure", "secret prompt", "unused"),)

{wagent_framework-2.0.0a1 → wagent_framework-2.0.0a2}/tests/test_local_runtime.py

@@ -125,6 +125,81 @@ async def test_local_runtime_assembles_full_text_run_and_persists_session(tmp_pa
     assert "not-persisted" not in persisted
 
 
+@pytest.mark.asyncio
+async def test_local_runtime_loads_explicit_pricing_and_persists_cost(tmp_path):
+    config = local_runtime_config_from_mapping(
+        {
+            "provider": {
+                "template": "fake",
+                "name": "test-provider",
+                "model": "test-model",
+            },
+            "pricing": {
+                "version": "prices-v1",
+                "currency": "USD",
+                "max_cost": "10",
+                "prices": [
+                    {
+                        "provider": "test-provider",
+                        "model": "test-model",
+                        "input_per_million": "1000000",
+                        "output_per_million": "1000000",
+                    }
+                ],
+            },
+        }
+    )
+    runtime = assemble_local_runtime(
+        config,
+        tmp_path / ".wagent",
+        templates=_templates({}),
+    )
+
+    run = await runtime.run("priced")
+
+    assert run.result.cost is not None
+    assert run.result.cost.total == 5
+    assert run.result.cost_complete is True
+    assert run.session.runs[0].cost is not None
+    assert run.session.runs[0].cost.total == 5
+    persisted = next((tmp_path / ".wagent" / "sessions").glob("*.json")).read_text(
+        encoding="utf-8"
+    )
+    assert '"price_table_version":"prices-v1"' in persisted
+
+
+def test_local_runtime_pricing_requires_configured_provider_model_rate(tmp_path):
+    config = local_runtime_config_from_mapping(
+        {
+            "provider": {
+                "template": "fake",
+                "name": "test-provider",
+                "model": "test-model",
+            },
+            "pricing": {
+                "version": "prices-v1",
+                "currency": "USD",
+                "max_cost": "1",
+                "prices": [
+                    {
+                        "provider": "other",
+                        "model": "test-model",
+                        "input_per_million": "1",
+                        "output_per_million": "2",
+                    }
+                ],
+            },
+        }
+    )
+
+    with pytest.raises(LocalRuntimeConfigError, match="configured provider"):
+        assemble_local_runtime(
+            config,
+            tmp_path,
+            templates=_templates({}),
+        )
+
+
 def test_local_runtime_requires_env_reference_and_rejects_secret_fields(tmp_path):
     with pytest.raises(LocalRuntimeConfigError, match="environment variable"):
         assemble_local_runtime(

wagent_framework-2.0.0a2/tests/test_pricing.py

@@ -0,0 +1,71 @@
+from decimal import Decimal
+
+import pytest
+
+from w_agent import (
+    ModelPrice,
+    PriceNotFoundError,
+    PriceTable,
+    PricingCatalog,
+    PricingError,
+    TokenUsage,
+)
+
+
+def test_versioned_price_table_quotes_cached_and_uncached_tokens_exactly():
+    table = PriceTable(
+        "2026-09-18",
+        "usd",
+        (
+            ModelPrice(
+                "provider",
+                "model",
+                input_per_million="2",
+                output_per_million="4",
+                cached_input_per_million="1",
+            ),
+        ),
+    )
+
+    cost = table.quote("provider", "model", TokenUsage(1000, 200, 400))
+
+    assert cost.currency == "USD"
+    assert cost.price_table_version == "2026-09-18"
+    assert cost.input_cost == Decimal("0.0012")
+    assert cost.cached_input_cost == Decimal("0.0004")
+    assert cost.output_cost == Decimal("0.0008")
+    assert cost.total == Decimal("0.0024")
+
+
+def test_pricing_catalog_requires_explicit_unique_versions_and_model_rates():
+    table = PriceTable(
+        "v1",
+        "USD",
+        (ModelPrice("provider", "model", "1", "2"),),
+    )
+    catalog = PricingCatalog((table,))
+
+    assert catalog.quote("v1", "provider", "model", TokenUsage(1, 1)).total > 0
+    with pytest.raises(PriceNotFoundError):
+        catalog.quote("v1", "provider", "other", TokenUsage(1, 1))
+    with pytest.raises(PricingError, match="already exists"):
+        catalog.register(table)
+
+
+def test_price_table_rejects_invalid_cached_usage_and_duplicate_rates():
+    with pytest.raises(PricingError, match="duplicate"):
+        PriceTable(
+            "v1",
+            "USD",
+            (
+                ModelPrice("provider", "model", "1", "2"),
+                ModelPrice("provider", "model", "3", "4"),
+            ),
+        )
+    table = PriceTable(
+        "v1",
+        "USD",
+        (ModelPrice("provider", "model", "1", "2"),),
+    )
+    with pytest.raises(PricingError, match="cannot exceed"):
+        table.quote("provider", "model", TokenUsage(1, 0, 2))

{wagent_framework-2.0.0a1 → wagent_framework-2.0.0a2}/tests/test_react_agent.py

@@ -18,6 +18,10 @@ from w_agent import (
     ModelRegistry,
     ModelRouter,
     JsonlRunStore,
+    CostBudget,
+    ModelPrice,
+    PriceTable,
+    PricingCatalog,
     ReactAgentLoop,
     RunResumeConflictError,
     RunContext,
@@ -79,7 +83,7 @@ class AgentProvider:
         yield FinishEvent(reason)
 
 
-def _loop(responses, binding, *, store=None, policy=None):
+def _loop(responses, binding, *, store=None, policy=None, pricing=None):
     provider = AgentProvider(responses)
     models = ModelRegistry()
     models.register("fake", provider)
@@ -96,6 +100,7 @@ def _loop(responses, binding, *, store=None, policy=None):
         tools,
         tool_executor,
         store=store,
+        pricing=pricing,
     ), provider
 
 
@@ -387,6 +392,90 @@ async def test_react_loop_can_fail_closed_when_provider_omits_usage():
     assert result.usage_complete is False
 
 
+@pytest.mark.asyncio
+async def test_react_loop_enforces_versioned_cost_budget_before_tools():
+    calls = 0
+
+    def ping() -> str:
+        nonlocal calls
+        calls += 1
+        return "pong"
+
+    pricing = PricingCatalog(
+        (
+            PriceTable(
+                "prices-v1",
+                "USD",
+                (ModelPrice("fake", "chat", "1000000", "1000000"),),
+            ),
+        )
+    )
+    loop, _ = _loop(
+        [
+            (
+                (ToolCallContent("ping-1", "ping", "{}"),),
+                FinishReason.TOOL_CALLS,
+                TokenUsage(8, 4),
+            )
+        ],
+        python_tool(ping),
+        pricing=pricing,
+    )
+
+    result = await loop.run(
+        AgentDefinition(
+            "cost-bounded",
+            cost_budget=CostBudget("10", "prices-v1", "USD"),
+        ),
+        _context(),
+    )
+
+    assert result.stop_reason is StopReason.COST_BUDGET
+    assert result.cost is not None
+    assert result.cost.total == 12
+    assert result.cost_complete is True
+    assert result.priced_usage_calls == 1
+    assert calls == 0
+    cost_event = next(
+        event for event in result.events if event.type is RunEventType.COST_USAGE
+    )
+    assert cost_event.data["cost"]["total"] == "12"
+
+
+@pytest.mark.asyncio
+async def test_react_cost_budget_fails_closed_without_usage_or_pricing():
+    pricing = PricingCatalog(
+        (
+            PriceTable(
+                "prices-v1",
+                "USD",
+                (ModelPrice("fake", "chat", "1", "2"),),
+            ),
+        )
+    )
+    unmetered, _ = _loop(
+        [((TextContent("unmetered"),), FinishReason.STOP)],
+        python_tool(lambda: "unused", name="unused"),
+        pricing=pricing,
+    )
+    missing_resolver, provider = _loop(
+        [((TextContent("must not run"),), FinishReason.STOP, TokenUsage(1, 1))],
+        python_tool(lambda: "unused", name="unused"),
+    )
+    definition = AgentDefinition(
+        "strict-cost",
+        cost_budget=CostBudget("1", "prices-v1", "USD"),
+    )
+
+    unmetered_result = await unmetered.run(definition, _context())
+    missing_result = await missing_resolver.run(definition, _context())
+
+    assert unmetered_result.stop_reason is StopReason.COST_UNAVAILABLE
+    assert unmetered_result.cost_complete is False
+    assert missing_result.stop_reason is StopReason.COST_UNAVAILABLE
+    assert provider.requests == []
+
+
 @pytest.mark.asyncio
 async def test_react_attempt_ledger_marks_retry_usage_incomplete():
     loop, _ = _loop(
@@ -471,6 +560,70 @@ async def test_attempt_checkpoint_persists_ledger_without_failure_message(tmp_pa
     assert "secret provider response" not in persisted
 
 
+@pytest.mark.asyncio
+async def test_cost_ledger_survives_approval_checkpoint_resume(tmp_path):
+    saved = []
+    binding = python_tool(
+        lambda text: saved.append(text) or "saved",
+        name="save",
+        side_effect=ToolSideEffect.WRITE,
+    )
+    pricing = PricingCatalog(
+        (
+            PriceTable(
+                "prices-v1",
+                "USD",
+                (ModelPrice("fake", "chat", "1000000", "1000000"),),
+            ),
+        )
+    )
+    definition = AgentDefinition(
+        "priced-writer",
+        cost_budget=CostBudget("100", "prices-v1", "USD"),
+    )
+    first_loop, _ = _loop(
+        [
+            (
+                (ToolCallContent("write-1", "save", '{"text":"value"}'),),
+                FinishReason.TOOL_CALLS,
+                TokenUsage(6, 2),
+            )
+        ],
+        binding,
+        store=JsonlRunStore(tmp_path),
+        pricing=pricing,
+    )
+
+    pending = await first_loop.run(definition, _context())
+    checkpoint = await JsonlRunStore(tmp_path).load_checkpoint("run-1")
+
+    assert pending.stop_reason is StopReason.NEEDS_APPROVAL
+    assert checkpoint is not None
+    assert checkpoint.cost is not None
+    assert checkpoint.cost.total == 8
+    assert checkpoint.priced_usage_calls == 1
+
+    second_loop, _ = _loop(
+        [((TextContent("saved"),), FinishReason.STOP, TokenUsage(7, 3))],
+        binding,
+        store=JsonlRunStore(tmp_path),
+        pricing=pricing,
+    )
+    resumed = await second_loop.resume(
+        "run-1",
+        tool_context=ToolExecutionContext(
+            approved_call_ids=frozenset({"write-1"})
+        ),
+    )
+
+    assert resumed.stop_reason is StopReason.COMPLETED
+    assert resumed.cost is not None
+    assert resumed.cost.total == 18
+    assert resumed.cost_complete is True
+    assert resumed.priced_usage_calls == 2
+    assert saved == ["value"]
+
+
 @pytest.mark.asyncio
 async def test_react_approval_resumes_from_jsonl_without_repeating_model(tmp_path):
     saved = []

{wagent_framework-2.0.0a1 → wagent_framework-2.0.0a2}/tests/test_tui.py

@@ -11,6 +11,7 @@ from w_agent import (
     ModelCapability,
     ModelDescriptor,
     ModelMessage,
+    ModelCost,
     RunResult,
     StopReason,
     TokenUsage,
@@ -227,6 +228,8 @@ async def test_tui_runs_private_evaluation_and_closes_runtime(tmp_path, monkeypa
                     usage=TokenUsage(4, 2),
                     model_calls=1,
                     reported_usage_calls=1,
+                    cost=ModelCost("USD", "prices-v1", "0.01", "0.02"),
+                    priced_usage_calls=1,
                 )
             )
 
@@ -252,6 +255,7 @@ async def test_tui_runs_private_evaluation_and_closes_runtime(tmp_path, monkeypa
         rendered = str(app.query_one("#evaluation-result").content)
         assert "Passed: 1/1" in rendered
         assert "in=4, out=2" in rendered
+        assert "Cost: 0.03 USD" in rendered
         assert "SECRET_PROMPT" not in rendered
         assert "VISIBLE_OUTPUT" not in rendered
         assert app.query_one("#evaluation-confirm").value == ""