@gt-fe/eap-runtime 0.0.13 → 0.0.15

package/dist/registry/graph-loader.js

@@ -8,11 +8,12 @@ import { readFile } from 'node:fs/promises';
 import path from 'node:path';
 import { resolveAgentModelAllowlist, resolveAgentDefaultModel } from './agent-models.js';
 import { DEFAULT_HTTP_TOOL_TIMEOUT_SEC, DEFAULT_TOOL_TIMEOUT_SEC } from '../tool/timeout.js';
-import { resolveStdioMcpCwd } from './mcp-stdio-cwd.js';
+import { isMcpDiscoverStub } from '../tool/mcp-list.js';
 import { createDeploymentSnapshotFromEnv, isInlineToolDeclaration, overlayToolRefPolicies, } from './deployment-snapshot.js';
 import { preloadGraphNotFound, preloadManifestNotFound, preloadSkillNotFound, preloadToolNotFound, } from './preload-errors.js';
 import { PreinstallDefinitionCache, resolveDefinitionCacheEnabled, } from './definition-cache.js';
 import { inspectAgentSkillTree, listSkillDirs, readSkillAssembly, resolveSkillHotReload, skillAssemblyStamp, } from './skill-tree.js';
+import { emitRuntimeLog } from '../common/logger.js';
 function applyDefaultModelToGraph(graph, defaultModel) {
     if (!defaultModel)
         return graph;
@@ -57,6 +58,24 @@ function asGraphDsl(value) {
     }
     return undefined;
 }
+export async function resolveAgentSystemPromptFromLoader(loader, agentCode, version) {
+    if (!loader || typeof loader.resolveAgentSystemPrompt !== 'function')
+        return undefined;
+    return loader.resolveAgentSystemPrompt(agentCode, version);
+}
+/** manifest.prompts.system；空白视为缺省。 */
+export function extractAgentSystemPrompt(manifest) {
+    if (!manifest || typeof manifest !== 'object' || Array.isArray(manifest))
+        return undefined;
+    const prompts = manifest.prompts;
+    if (!prompts || typeof prompts !== 'object' || Array.isArray(prompts))
+        return undefined;
+    const system = prompts.system;
+    if (typeof system !== 'string')
+        return undefined;
+    const trimmed = system.trim();
+    return trimmed ? trimmed : undefined;
+}
 function extractSkillPrompt(manifest) {
     const prompt = manifest.prompt;
     if (typeof prompt === 'string' && prompt.trim())
@@ -96,35 +115,6 @@ async function resolveSkillPrompt(manifest, bundlePath) {
     }
     return raw;
 }
-function extractSkillToolsets(manifest) {
-    const refs = manifest.toolsets;
-    if (!Array.isArray(refs))
-        return undefined;
-    const out = [];
-    for (const item of refs) {
-        if (typeof item === 'string') {
-            out.push(item);
-            continue;
-        }
-        if (!item || typeof item !== 'object')
-            continue;
-        const o = item;
-        const code = o.code != null ? String(o.code) : undefined;
-        const enabled = Array.isArray(o.enabled)
-            ? o.enabled.filter((x) => typeof x === 'string')
-            : undefined;
-        out.push({
-            name: o.name != null ? String(o.name) : code,
-            version: o.version != null ? String(o.version) : undefined,
-            code,
-            enabled,
-            autoApprove: Array.isArray(o.autoApprove)
-                ? o.autoApprove.filter((x) => typeof x === 'string')
-                : undefined,
-        });
-    }
-    return out.length > 0 ? out : undefined;
-}
 function extractEmbeddedGraph(manifest) {
     const candidates = [
         manifest.graph,
@@ -224,6 +214,67 @@ function toLoadedToolDefinition(record) {
         inputSchema: m.input,
     };
 }
+function resolveMcpServer(manifest) {
+    if (manifest.server && typeof manifest.server === 'object' && !Array.isArray(manifest.server)) {
+        return manifest.server;
+    }
+    const execution = manifest.execution;
+    if (execution && typeof execution === 'object' && !Array.isArray(execution)) {
+        const exec = execution;
+        if (exec.server && typeof exec.server === 'object' && !Array.isArray(exec.server)) {
+            return exec.server;
+        }
+        const mcp = exec.mcp;
+        if (mcp && typeof mcp === 'object' && !Array.isArray(mcp)) {
+            const nested = mcp.server;
+            if (nested && typeof nested === 'object' && !Array.isArray(nested)) {
+                return nested;
+            }
+        }
+    }
+    return undefined;
+}
+function mcpDiscoverStub(pre, options) {
+    const m = pre.manifest;
+    const server = resolveMcpServer(m);
+    const toolRuntime = (m.toolRuntime ?? m.runtime);
+    const timeout = (typeof server?.timeout === 'number' ? server.timeout : undefined)
+        ?? DEFAULT_TOOL_TIMEOUT_SEC;
+    const transport = (typeof server?.transport === 'string' && server.transport)
+        || (server?.url ? 'streamable_http' : 'stdio');
+    const execution = {
+        protocol: 'mcp',
+        transport,
+        server,
+        timeout,
+        discover: true,
+        bundlePath: pre.bundlePath,
+    };
+    if (options?.enabledTools && options.enabledTools.length > 0) {
+        execution.enabledTools = options.enabledTools;
+    }
+    if (options?.autoApprove && options.autoApprove.length > 0) {
+        execution.autoApprove = options.autoApprove;
+    }
+    return {
+        toolId: pre.toolId,
+        name: pre.name,
+        manifest: {
+            id: pre.toolId,
+            name: pre.name,
+            version: pre.version,
+            type: 'mcp',
+            status: 'active',
+            description: m.description,
+            toolRuntime,
+            dataScope: normalizeDataScope(m.permissions && typeof m.permissions === 'object' && !Array.isArray(m.permissions)
+                ? m.permissions.dataScope
+                : m.dataScope),
+            requireConfirmation: false,
+            execution,
+        },
+    };
+}
 function fromPreinstallTool(pre) {
     const m = pre.manifest;
     const permissions = m.permissions;
@@ -255,14 +306,16 @@ function fromPreinstallTool(pre) {
     };
 }
 /**
- * 展开预装工具集（toolset.manifest.json，含 tools[]）为多个 LoadedToolDefinition。
- * 以工具集为管理单元：单个 MCP 服务 / 单个代码包 = 一个工具集，可含多个子工具；
- * 子工具继承集级 server/package 配置，合成各自的 execution。
- * 无 tools[] 时回退为单工具 manifest（tool.manifest.json 结构）。
+ * 展开预装工具集为 LoadedToolDefinition。
+ * - type=mcp：忽略清单 tools[]，产出一条 discover stub；子工具在执行期 tools/list。
+ * - type=code/http：展开 tools[]；无 tools[] 时回退单工具 manifest。
  * enabled 为可选白名单（agent.manifest toolsets[].enabled）；缺省启用全部子工具。
  */
 function expandPreinstallToolset(pre, options) {
     const m = pre.manifest;
+    if (m.type === 'mcp') {
+        return [mcpDiscoverStub(pre, options)];
+    }
     const tools = m.tools;
     const enabled = options?.enabledTools;
     const enabledSet = enabled && enabled.length > 0 ? new Set(enabled) : null;
@@ -281,7 +334,6 @@ function expandPreinstallToolset(pre, options) {
     }
     const setType = m.type;
     const toolRuntime = (m.toolRuntime ?? m.runtime);
-    const server = m.server;
     const pkg = m.package;
     const out = [];
     for (const raw of tools) {
@@ -302,28 +354,10 @@ function expandPreinstallToolset(pre, options) {
                 baseUrl: st.baseUrl ?? gateway.baseUrl,
                 pathTemplate: st.pathTemplate,
                 allowedHosts: gateway.allowedHosts,
+                header: gateway.header,
                 timeout: st.timeout ?? gateway.timeout ?? DEFAULT_HTTP_TOOL_TIMEOUT_SEC,
             };
         }
-        else if (setType === 'mcp') {
-            type = 'mcp';
-            const transport = server?.transport ?? (server?.url ? 'streamable_http' : 'stdio');
-            const resolvedServer = server
-                ? {
-                    ...server,
-                    ...(transport === 'stdio'
-                        ? { cwd: resolveStdioMcpCwd(pre.bundlePath, server.cwd) }
-                        : {}),
-                }
-                : server;
-            execution = {
-                protocol: 'mcp',
-                transport,
-                server: resolvedServer,
-                toolName: name,
-                timeout: st.timeout ?? server?.timeout ?? DEFAULT_TOOL_TIMEOUT_SEC,
-            };
-        }
         else {
             type = 'code';
             execution = {
@@ -522,28 +556,24 @@ export class GraphLoader {
         catch (err) {
             const message = err instanceof Error ? err.message : String(err);
             skillWarnings.push(message);
-            console.warn(JSON.stringify({
+            emitRuntimeLog('warn', 'graph-loader', message, {
                 event: 'runtime.warmAgent.skills_failed',
-                level: 'warn',
                 agentCode,
                 agentVersion: version ?? null,
-                message,
-            }));
+            });
         }
         // 预期跳过（无 userId 的 users 层 Skill）用 debug，避免单测/生产噪音
         if (skippedSkills.length > 0) {
-            console.debug(JSON.stringify({
+            emitRuntimeLog('debug', 'graph-loader', 'warmAgent skipped user skills', {
                 event: 'runtime.warmAgent.skills_skipped',
-                level: 'debug',
                 agentCode,
                 agentVersion: version ?? null,
                 userId: options?.userId ?? null,
                 skippedSkillIds: skippedSkills,
-            }));
+            });
         }
-        console.info(JSON.stringify({
+        emitRuntimeLog('info', 'graph-loader', 'warmAgent ok', {
             event: 'runtime.warmAgent.ok',
-            level: 'info',
             agentCode,
             agentVersion: version ?? null,
             graph: true,
@@ -551,7 +581,7 @@ export class GraphLoader {
             skills: skillsOk,
             skillWarningCount: skillWarnings.length,
             skillWarnings,
-        }));
+        });
         return {
             graph: true,
             tools: true,
@@ -632,7 +662,6 @@ export class GraphLoader {
                     type: 'llm',
                     config: {
                         model,
-                        systemPrompt: manifest.prompts?.system,
                     },
                 },
             ],
@@ -642,6 +671,24 @@ export class GraphLoader {
             },
         };
     }
+    /**
+     * Agent 基线系统提示词（manifest.prompts.system）。
+     * 每轮 execute/stream 注入 GraphState.agentSystemPrompt；不写进节点 config。
+     */
+    async resolveAgentSystemPrompt(agentCode, version) {
+        if (this.deploymentSnapshot) {
+            const snap = await this.deploymentSnapshot.loadAgentSnapshot(agentCode, version);
+            if (snap) {
+                return extractAgentSystemPrompt(snap.manifest);
+            }
+            if (this.blockRegistryDefinitions)
+                return undefined;
+        }
+        if (this.blockRegistryDefinitions || !this.registryClient)
+            return undefined;
+        const record = await this.registryClient.getAgent(agentCode);
+        return extractAgentSystemPrompt(record?.manifest);
+    }
     /**
      * 解析 agent tools[] + graph tool 节点 union；
      * server+预装：以预装 manifest 为准；desktop 合并 ILocalToolInstallStore。
@@ -753,6 +800,9 @@ export class GraphLoader {
                     continue;
                 const pre = await this.deploymentSnapshot.loadToolManifest({ name: toolName }, agentDir);
                 if (!pre) {
+                    // MCP 子工具名在 tools/list 后才出现，图内 type:tool 节点不必对应预装目录
+                    if (definitions.some(isMcpDiscoverStub))
+                        continue;
                     missing.push(toolName);
                     continue;
                 }
@@ -861,7 +911,6 @@ export class GraphLoader {
                     prompt,
                     bundlePath: pre.bundlePath,
                     entryFile,
-                    toolsets: extractSkillToolsets(pre.manifest),
                     inputDef: pre.manifest.inputDef,
                     outputDef: pre.manifest.outputDef,
                 });

package/dist/runtime/agent-runtime.d.ts

@@ -11,7 +11,7 @@ import type { ICheckpointSaver } from '../checkpoint/contracts/checkpoint-saver.
 import type { ITraceCollector } from '../contracts/trace-collector.js';
 import type { IStateGraphEngine } from '../graph/contracts/state-graph.contract.js';
 import { HitlHandler } from '../graph/hitl/hitl-handler.js';
-import type { GraphLoader } from '../registry/graph-loader.js';
+import { type GraphLoader } from '../registry/graph-loader.js';
 import type { SessionContextService } from '../session/session-context.service.js';
 import type { MessageService } from '../session/message.service.js';
 import type { SessionService, SessionRecord, MessageRecord } from '../session/session.service.js';
@@ -86,8 +86,8 @@ export declare class AgentRuntime {
      * “Session 历史 + 当前输入”途径。
      *
      * 行为：
-     * - `role` 仅接受 `user` / `assistant`；`system` 应写入 graph.json LLM 节点的
-     *   `config.systemPrompt`（由 SystemPromptBuilder 拼平台规则 + 开发者提示词）；
+     * - `role` 仅接受 `user` / `assistant`；`system` 由 SystemPromptBuilder 拼
+     *   （平台规则 + `prompts.system` 基线 + 节点 `systemPrompt`）；
      *   `tool` 必须与前置 assistant.toolCalls 严格配对，不得单独作为种子。
      * - 幂等语义（exactly-once）：当 session 已存在任何历史消息时，直接返回
      *   `seeded: false`，不重复写入。

package/dist/runtime/agent-runtime.js

@@ -8,6 +8,7 @@
  */
 import { randomUUID } from 'node:crypto';
 import { HITL_TIMEOUT_ERROR_CODE, HITL_TIMEOUT_MESSAGE, HitlAlreadyTimedOutError, HitlNotWaitingError, buildHitlWaitedPayload, isHitlDeadlinePassed, readPersistedHitlWaiting, unwrapHitlPaused, } from '../graph/hitl/hitl-handler.js';
+import { resolveAgentSystemPromptFromLoader } from '../registry/graph-loader.js';
 import { toLoadedSkillMap } from '../tool/system-prompt-builder.js';
 import { assembleLoadedTools } from '../tool/harness-primitives.js';
 import { createFileLlmTurnLogFromEnv, } from '../debug/llm-turn-log.js';
@@ -71,8 +72,8 @@ export class AgentRuntime {
      * “Session 历史 + 当前输入”途径。
      *
      * 行为：
-     * - `role` 仅接受 `user` / `assistant`；`system` 应写入 graph.json LLM 节点的
-     *   `config.systemPrompt`（由 SystemPromptBuilder 拼平台规则 + 开发者提示词）；
+     * - `role` 仅接受 `user` / `assistant`；`system` 由 SystemPromptBuilder 拼
+     *   （平台规则 + `prompts.system` 基线 + 节点 `systemPrompt`）；
      *   `tool` 必须与前置 assistant.toolCalls 严格配对，不得单独作为种子。
      * - 幂等语义（exactly-once）：当 session 已存在任何历史消息时，直接返回
      *   `seeded: false`，不重复写入。
@@ -100,7 +101,7 @@ export class AgentRuntime {
                 throw new AppError(ErrorCodes.PARAMS_INVALID, `seedSessionMessages messages[${index}] must be an object`, { exposeMessage: true });
             }
             if (message.role !== 'user' && message.role !== 'assistant') {
-                throw new AppError(ErrorCodes.PARAMS_INVALID, `seedSessionMessages messages[${index}].role "${String(message.role)}" is not allowed; only 'user' and 'assistant' are accepted (system belongs to config.systemPrompt; tool must pair with prior assistant.toolCalls)`, { exposeMessage: true });
+                throw new AppError(ErrorCodes.PARAMS_INVALID, `seedSessionMessages messages[${index}].role "${String(message.role)}" is not allowed; only 'user' and 'assistant' are accepted (system belongs to SystemPromptBuilder: platform rules + prompts.system + node systemPrompt; tool must pair with prior assistant.toolCalls)`, { exposeMessage: true });
             }
             if (typeof message.content !== 'string' || message.content.length === 0) {
                 throw new AppError(ErrorCodes.PARAMS_INVALID, `seedSessionMessages messages[${index}].content must be a non-empty string`, { exposeMessage: true });
@@ -271,6 +272,7 @@ export class AgentRuntime {
         const agentDir = this.deps.graphLoader
             ? (await this.deps.graphLoader.resolveAgentDir(agentCode, params.agentVersion)) ?? undefined
             : undefined;
+        const agentSystemPrompt = await resolveAgentSystemPromptFromLoader(this.deps.graphLoader, agentCode, params.agentVersion);
         const history = messageRecordsToGraphMessages(sessionContext.messages);
         const messages = await buildTurnGraphMessages({
             history,
@@ -301,6 +303,7 @@ export class AgentRuntime {
                 skillDefinitions,
                 promptRoots,
                 agentDir,
+                agentSystemPrompt,
                 skillInput: resolveSkillInputFromInvokeInput(params.input),
                 traceId: resolveTraceId(params),
                 traceCollector: this.deps.traceCollector,
@@ -490,6 +493,9 @@ export class AgentRuntime {
                 (this.deps.graphLoader
                     ? (await this.deps.graphLoader.resolveAgentDir(agentCode, agentVersion)) ?? undefined
                     : undefined),
+            agentSystemPrompt: (await resolveAgentSystemPromptFromLoader(this.deps.graphLoader, agentCode, agentVersion)) ?? (typeof resumedState.agentSystemPrompt === 'string'
+                ? resumedState.agentSystemPrompt
+                : undefined),
             activatedSkill: resumedState.activatedSkill,
             skillRoute: resumedState.skillRoute ??
                 (typeof resumedState.activatedSkill === 'string' && resumedState.activatedSkill
@@ -652,11 +658,15 @@ export class AgentRuntime {
             (this.deps.graphLoader
                 ? (await this.deps.graphLoader.resolveAgentDir(agentCode, agentVersion)) ?? undefined
                 : undefined);
+        const agentSystemPrompt = (await resolveAgentSystemPromptFromLoader(this.deps.graphLoader, agentCode, agentVersion)) ?? (typeof forkedCheckpoint.state.agentSystemPrompt === 'string'
+            ? forkedCheckpoint.state.agentSystemPrompt
+            : undefined);
         try {
             const result = await this.deps.graphEngine.invoke(graph, {
                 ...forkedCheckpoint.state,
                 agentVersion,
                 agentDir,
+                agentSystemPrompt,
                 userId,
                 delegationToken: overrides.executionContext?.delegationToken,
                 executionContext: overrides.executionContext,

package/dist/runtime/dev-runtime.d.ts

@@ -49,6 +49,8 @@ export interface DevRuntime {
     dispose: () => void;
     /** 装配期诊断摘要（CLI 启动时打印，便于排查环境问题） */
     diagnostics: {
+        /** 当前加载的 Runtime 包 version */
+        packageVersion: string;
         deploymentContext: string;
         toolExecutionMode: string;
         llmProviderKind: 'routing' | 'openai-http' | 'echo';

package/dist/runtime/dev-runtime.js

@@ -35,6 +35,7 @@ import { InvokeOrchestrator, StreamOrchestrator, SessionService, MessageService,
 import { createAgentRuntime } from './agent-runtime.js';
 import { assembleRuntimeStorage } from './storage-profile.js';
 import { resolveSkillHotReload, watchPreinstallSkillRoot, watchSkillRoots } from '../registry/skill-tree.js';
+import { RUNTIME_PACKAGE_VERSION } from '../common/runtime-package-version.js';
 /**
  * 装配本地 dev 模式所需的全部 Runtime 依赖
  *
@@ -185,6 +186,7 @@ export function createDevRuntime() {
         graphLoader,
         dispose: stopSkillWatch,
         diagnostics: {
+            packageVersion: RUNTIME_PACKAGE_VERSION,
             deploymentContext,
             toolExecutionMode,
             llmProviderKind,

package/dist/runtime/types.d.ts

@@ -91,8 +91,8 @@ export interface ForkResult {
  * 开发者提供的对话上下文（开场白 / few-shot）。
  *
  * 约束（AgentRuntime.seedSessionMessages 强制校验）：
- * - `role` 白名单为 `user` / `assistant`，`system` 走 graph.json 的
- *   `config.systemPrompt`（由 SystemPromptBuilder 拼平台规则），`tool` 必须与
+ * - `role` 白名单为 `user` / `assistant`，`system` 由 SystemPromptBuilder 拼
+ *   （平台规则 + `prompts.system` + 节点 `systemPrompt`），`tool` 必须与
  *   前置 assistant.toolCalls 严格配对，两者禁止作为种子；
  * - 该 session 已有历史时，视作已种入，跳过写入（exactly-once 幂等）。
  */

package/dist/session/compression-middleware.d.ts

@@ -65,11 +65,12 @@ export declare class CompressionMiddleware {
         sessionId: string;
         taskId?: string;
         userId: string;
+        maxContextTokens?: number;
     }): Promise<MessageRecord[]>;
     /**
      * 检查是否需要压缩（不执行）
      */
-    checkCompression(messages: MessageRecord[]): {
+    checkCompression(messages: MessageRecord[], maxContextTokens?: number): {
         needsCompression: boolean;
         currentTokens: number;
         currentRounds: number;

package/dist/session/compression-middleware.js

@@ -65,13 +65,16 @@ export class CompressionMiddleware {
         // 转换为 GraphMessage 格式进行检查
         const graphMessages = this.toGraphMessages(messages);
         // 检查是否需要压缩
-        const checkResult = this.compression.checkCompression(graphMessages);
+        const windowOverride = context.maxContextTokens
+            ? { maxContextTokens: context.maxContextTokens }
+            : undefined;
+        const checkResult = this.compression.checkCompression(graphMessages, windowOverride);
         if (!checkResult.needsCompression) {
             return messages;
         }
         // 执行压缩
         const startTime = Date.now();
-        const compressedMessages = await this.compression.compress(graphMessages);
+        const compressedMessages = await this.compression.compress(graphMessages, windowOverride);
         const duration = Date.now() - startTime;
         // 转换回 MessageRecord 格式
         const compressedRecords = this.toMessageRecords(compressedMessages, messages);
@@ -106,9 +109,9 @@ export class CompressionMiddleware {
     /**
      * 检查是否需要压缩（不执行）
      */
-    checkCompression(messages) {
+    checkCompression(messages, maxContextTokens) {
         const graphMessages = this.toGraphMessages(messages);
-        return this.compression.checkCompression(graphMessages);
+        return this.compression.checkCompression(graphMessages, maxContextTokens ? { maxContextTokens } : undefined);
     }
     /**
      * 获取配置

package/dist/sse/events/stats-events.d.ts

@@ -41,4 +41,13 @@ export interface StatsNodeTokenUsagePayload {
     costEstimate: number;
     /** 该节点本次调用耗时（毫秒），可选 */
     durationMs?: number;
+    /** 本轮 prompt 占用的上下文窗 used/window；缺 window 则省略 */
+    used?: number;
+    window?: number;
+    /** 按字符比例把 used 拆到 system/messages/tools；合计 = used */
+    parts?: {
+        system: number;
+        messages: number;
+        tools: number;
+    };
 }

package/dist/token/compression-budget-integration.d.ts

@@ -18,7 +18,7 @@ import type { IMessageRepository } from '../session/session.service.js';
 export interface CompressionBudgetConfig {
     /** 模型 context window 大小 */
     modelContextWindow: number;
-    /** 历史消息占比阈值（超过此比例触发压缩，默认 0.8） */
+    /** 相对模型窗口的触发系数（默认 0.7，与滑动窗口同一公式） */
     historyThreshold?: number;
     /** 压缩后目标占比（默认 0.5） */
     targetRatio?: number;

package/dist/token/compression-budget-integration.js

@@ -9,7 +9,7 @@
  */
 import { createContextBudget } from './context-window-budget.js';
 import { CompressionMiddleware, createCompressionMiddlewareWithLLM, } from '../session/compression-middleware.js';
-import { estimateMessagesTokens } from '../memory/message-compression.service.js';
+import { estimateMessagesTokens, resolveCompressionMaxTokens, } from '../memory/message-compression.service.js';
 // ─────────────────────────────────────────────
 // CompressionBudgetIntegration
 // ─────────────────────────────────────────────
@@ -28,7 +28,7 @@ export class CompressionBudgetIntegration {
     constructor(config, repository, messageRepository) {
         this.config = {
             modelContextWindow: config.modelContextWindow,
-            historyThreshold: config.historyThreshold ?? 0.8,
+            historyThreshold: config.historyThreshold ?? 0.7,
             targetRatio: config.targetRatio ?? 0.5,
             autoCompress: config.autoCompress ?? true,
             llmConfig: config.llmConfig,
@@ -36,13 +36,17 @@ export class CompressionBudgetIntegration {
         // 创建预算管理器
         this.budget = createContextBudget(this.config.modelContextWindow);
         // 创建压缩中间件
+        const compressionConfig = {
+            maxContextTokens: this.config.modelContextWindow,
+        };
         if (this.config.llmConfig?.provider) {
             this.compression = createCompressionMiddlewareWithLLM(this.config.llmConfig.provider, {
                 llmConfig: this.config.llmConfig,
+                compressionConfig,
             }, repository, messageRepository);
         }
         else {
-            this.compression = new CompressionMiddleware({}, undefined, repository, messageRepository);
+            this.compression = new CompressionMiddleware({ compressionConfig }, undefined, repository, messageRepository);
         }
     }
     /**
@@ -55,12 +59,13 @@ export class CompressionBudgetIntegration {
         // 估算当前 Token 数
         const graphMessages = this.toGraphMessages(messages);
         const currentTokens = estimateMessagesTokens(graphMessages);
-        // 获取历史消息的 Token 上限
-        const maxTokens = this.budget.getMaxTokens('history');
-        // 计算使用率
-        const usageRatio = maxTokens > 0 ? currentTokens / maxTokens : 0;
-        // 检查是否超过阈值
-        const needsCompression = usageRatio > this.config.historyThreshold;
+        // 与滑动窗口同一阈值：window × 系数（默认 0.7）；historyThreshold 作为系数覆盖
+        const maxTokens = resolveCompressionMaxTokens(this.config.modelContextWindow, this.config.historyThreshold);
+        // 计算使用率（相对窗口，不是相对阈值）
+        const usageRatio = this.config.modelContextWindow > 0
+            ? currentTokens / this.config.modelContextWindow
+            : 0;
+        const needsCompression = currentTokens > maxTokens;
         // 计算压缩目标
         const targetTokens = needsCompression
             ? Math.floor(maxTokens * this.config.targetRatio)
@@ -96,7 +101,10 @@ export class CompressionBudgetIntegration {
         }
         // 执行压缩
         const startTime = Date.now();
-        const compressedMessages = await this.compression.processMessages(messages, context);
+        const compressedMessages = await this.compression.processMessages(messages, {
+            ...context,
+            maxContextTokens: this.config.modelContextWindow,
+        });
         const duration = Date.now() - startTime;
         // 计算压缩后的 Token 数
         const compressedGraphMessages = this.toGraphMessages(compressedMessages);

package/dist/token/context-window.d.ts

@@ -0,0 +1,42 @@
+/**
+ * 本轮 prompt 占用的上下文窗：used = 本轮 inputTokens，window = ModelRecord.capabilities.maxContextTokens。
+ * 不是账单合计，也不是 max_tokens。
+ * parts 按本轮 system / messages / tools 字符比例把 used 拆开（无独立 tokenizer）。
+ */
+export interface ContextUsageParts {
+    system: number;
+    messages: number;
+    tools: number;
+}
+export interface ContextWindowUsage {
+    used: number;
+    window?: number;
+    parts?: ContextUsageParts;
+}
+export declare function contextWindowUsage(used: number, window: number | undefined): ContextWindowUsage | undefined;
+export declare function splitContextUsage(used: number, sections: {
+    systemChars: number;
+    messageChars: number;
+    toolChars: number;
+}): ContextUsageParts | undefined;
+export declare function turnSectionChars(config: {
+    systemPrompt?: string;
+    tools?: unknown[];
+}, messages: Array<{
+    content?: string;
+    toolCalls?: unknown;
+}>): {
+    systemChars: number;
+    messageChars: number;
+    toolChars: number;
+};
+export declare function buildContextUsage(used: number, window: number | undefined, sections?: {
+    systemChars: number;
+    messageChars: number;
+    toolChars: number;
+}): ContextWindowUsage | undefined;
+export declare function pickContextFields(raw: {
+    used?: unknown;
+    window?: unknown;
+    parts?: unknown;
+}): Partial<ContextWindowUsage>;