@jsonstudio/llms 0.6.631 → 0.6.743

package/dist/router/virtual-router/sticky-session-store.js

@@ -108,3 +108,39 @@ export function saveRoutingInstructionStateAsync(key, state) {
         // ignore sync write failures
     }
 }
+export function saveRoutingInstructionStateSync(key, state) {
+    if (!isPersistentKey(key)) {
+        return;
+    }
+    const dir = resolveSessionDir();
+    const filename = keyToFilename(key);
+    if (!dir || !filename) {
+        return;
+    }
+    const filepath = path.join(dir, filename);
+    if (!state) {
+        try {
+            fs.unlinkSync(filepath);
+        }
+        catch {
+            // ignore unlink failures
+        }
+        return;
+    }
+    const payload = {
+        version: 1,
+        state: serializeRoutingInstructionState(state)
+    };
+    try {
+        fs.mkdirSync(dir, { recursive: true });
+    }
+    catch {
+        // ignore mkdir errors
+    }
+    try {
+        fs.writeFileSync(filepath, JSON.stringify(payload), { encoding: 'utf8' });
+    }
+    catch {
+        // ignore sync write failures
+    }
+}

package/dist/router/virtual-router/types.d.ts

@@ -297,6 +297,19 @@ export interface RoutingDiagnostics {
     poolId?: string;
     confidence: number;
 }
+export interface StopMessageStateSnapshot {
+    stopMessageText: string;
+    stopMessageMaxRepeats: number;
+    /**
+     * stopMessage 来源：
+     * - 'explicit'：来自用户显式指令
+     * - 'auto'：系统基于空响应/错误自动推导
+     */
+    stopMessageSource?: string;
+    stopMessageUsed?: number;
+    stopMessageUpdatedAt?: number;
+    stopMessageLastUsedAt?: number;
+}
 export interface RoutingStatusSnapshot {
     routes: Record<string, {
         providers: string[];
@@ -373,3 +386,12 @@ export interface VirtualRouterHealthStore {
      */
     recordProviderError?(event: ProviderErrorEvent): void;
 }
+export interface ProviderQuotaViewEntry {
+    providerKey: string;
+    inPool: boolean;
+    reason?: string;
+    priorityTier?: number;
+    cooldownUntil?: number | null;
+    blacklistUntil?: number | null;
+}
+export type ProviderQuotaView = (providerKey: string) => ProviderQuotaViewEntry | null;

package/dist/servertool/engine.js

@@ -1,4 +1,8 @@
 import { runServerSideToolEngine } from './server-side-tools.js';
+import { ProviderProtocolError } from '../conversion/shared/errors.js';
+import { createHash } from 'node:crypto';
+import { loadRoutingInstructionStateSync, saveRoutingInstructionStateSync } from '../router/virtual-router/sticky-session-store.js';
+import { deserializeRoutingInstructionState, serializeRoutingInstructionState } from '../router/virtual-router/routing-instructions.js';
 export async function runServerToolOrchestration(options) {
     const engineOptions = {
         chatResponse: options.chat,
@@ -23,31 +27,214 @@ export async function runServerToolOrchestration(options) {
             flowId: engineResult.execution.flowId
         };
     }
+    const isStopMessageFlow = engineResult.execution.flowId === 'stop_message_flow';
+    const stopMessageSource = isStopMessageFlow ? getStopMessageSource(options.adapterContext) : undefined;
+    const isAutoStopMessage = isStopMessageFlow && stopMessageSource !== 'explicit';
+    const isErrorAutoFlow = engineResult.execution.flowId === 'iflow_model_error_retry';
+    const applyAutoLimit = isAutoStopMessage || isErrorAutoFlow;
     const routeHint = resolveRouteHint(options.adapterContext, engineResult.execution.flowId);
+    const loopState = buildServerToolLoopState(options.adapterContext, engineResult.execution.flowId, engineResult.execution.followup.payload);
+    if (applyAutoLimit && loopState && typeof loopState.repeatCount === 'number' && loopState.repeatCount >= 3) {
+        return {
+            chat: engineResult.finalChatResponse,
+            executed: true,
+            flowId: engineResult.execution.flowId
+        };
+    }
+    if (isAdapterClientDisconnected(options.adapterContext)) {
+        return {
+            chat: engineResult.finalChatResponse,
+            executed: true,
+            flowId: engineResult.execution.flowId
+        };
+    }
+    const followupEntryEndpoint = engineResult.execution.followup.entryEndpoint ||
+        options.entryEndpoint ||
+        '/v1/chat/completions';
     const metadata = {
         serverToolFollowup: true,
         stream: false,
+        ...(loopState ? { serverToolLoopState: loopState } : {}),
         ...(engineResult.execution.followup.metadata ?? {})
     };
     if (routeHint && typeof metadata.routeHint !== 'string') {
         metadata.routeHint = routeHint;
     }
-    const followup = await options.reenterPipeline({
-        entryEndpoint: '/v1/chat/completions',
-        requestId: `${options.requestId}${engineResult.execution.followup.requestIdSuffix}`,
-        body: engineResult.execution.followup.payload,
-        metadata
-    });
-    const followupBody = followup.body && typeof followup.body === 'object'
+    const maxAttempts = isStopMessageFlow ? 2 : 1;
+    const followupRequestId = buildFollowupRequestId(options.requestId, engineResult.execution.followup.requestIdSuffix);
+    let followup;
+    let lastError;
+    let reservation = null;
+    for (let attempt = 1; attempt <= maxAttempts; attempt += 1) {
+        try {
+            if (isStopMessageFlow) {
+                reservation = reserveStopMessageUsage(options.adapterContext);
+            }
+            followup = await options.reenterPipeline({
+                entryEndpoint: followupEntryEndpoint,
+                requestId: followupRequestId,
+                body: engineResult.execution.followup.payload,
+                metadata
+            });
+            lastError = undefined;
+            break;
+        }
+        catch (error) {
+            if (reservation) {
+                rollbackStopMessageUsage(reservation);
+                reservation = null;
+            }
+            lastError = error;
+            if (attempt >= maxAttempts) {
+                const wrapped = new ProviderProtocolError(`[servertool] Followup failed for flow ${engineResult.execution.flowId ?? 'unknown'} ` +
+                    `(attempt ${attempt}/${maxAttempts})`, {
+                    code: 'SERVERTOOL_FOLLOWUP_FAILED',
+                    details: {
+                        flowId: engineResult.execution.flowId,
+                        requestId: options.requestId,
+                        attempt,
+                        maxAttempts,
+                        error: error instanceof Error ? error.message : String(error ?? 'unknown')
+                    }
+                });
+                wrapped.cause = error;
+                throw wrapped;
+            }
+        }
+    }
+    const followupBody = followup && followup.body && typeof followup.body === 'object'
         ? followup.body
-        : engineResult.finalChatResponse;
-    const decorated = decorateFinalChatWithServerToolContext(followupBody, engineResult.execution);
+        : undefined;
+    if (isStopMessageFlow && !followupBody) {
+        const wrapped = new ProviderProtocolError(`[servertool] Followup returned empty response for flow ${engineResult.execution.flowId ?? 'unknown'}`, {
+            code: 'SERVERTOOL_FOLLOWUP_FAILED',
+            details: {
+                flowId: engineResult.execution.flowId,
+                requestId: options.requestId,
+                error: lastError instanceof Error ? lastError.message : undefined
+            }
+        });
+        wrapped.cause = lastError;
+        throw wrapped;
+    }
+    const decorated = decorateFinalChatWithServerToolContext(followupBody ?? engineResult.finalChatResponse, engineResult.execution);
     return {
         chat: decorated,
         executed: true,
         flowId: engineResult.execution.flowId
     };
 }
+function reserveStopMessageUsage(adapterContext) {
+    if (!adapterContext || typeof adapterContext !== 'object') {
+        return null;
+    }
+    const sessionId = typeof adapterContext.sessionId === 'string'
+        ? adapterContext.sessionId.trim()
+        : '';
+    const conversationId = typeof adapterContext.conversationId === 'string'
+        ? adapterContext.conversationId.trim()
+        : '';
+    const stickyKey = sessionId ? `session:${sessionId}` : conversationId ? `conversation:${conversationId}` : '';
+    if (!stickyKey) {
+        return null;
+    }
+    let state = loadRoutingInstructionStateSync(stickyKey);
+    if (!state || !state.stopMessageText || !state.stopMessageMaxRepeats) {
+        const fallback = resolveStopMessageSnapshot(adapterContext.stopMessageState);
+        if (!fallback) {
+            return null;
+        }
+        state = createStopMessageState(fallback);
+    }
+    const text = typeof state.stopMessageText === 'string' ? state.stopMessageText.trim() : '';
+    const maxRepeats = typeof state.stopMessageMaxRepeats === 'number' && Number.isFinite(state.stopMessageMaxRepeats)
+        ? Math.max(1, Math.floor(state.stopMessageMaxRepeats))
+        : 0;
+    if (!text || maxRepeats <= 0) {
+        return null;
+    }
+    const previousState = cloneRoutingInstructionState(state);
+    const used = typeof state.stopMessageUsed === 'number' && Number.isFinite(state.stopMessageUsed)
+        ? Math.max(0, Math.floor(state.stopMessageUsed))
+        : 0;
+    const nextUsed = used + 1;
+    state.stopMessageUsed = nextUsed;
+    state.stopMessageLastUsedAt = Date.now();
+    if (nextUsed >= maxRepeats) {
+        state.stopMessageText = undefined;
+        state.stopMessageMaxRepeats = undefined;
+        state.stopMessageUsed = undefined;
+        state.stopMessageUpdatedAt = undefined;
+        state.stopMessageLastUsedAt = undefined;
+    }
+    saveRoutingInstructionStateSync(stickyKey, state);
+    return { stickyKey, previousState };
+}
+function rollbackStopMessageUsage(reservation) {
+    saveRoutingInstructionStateSync(reservation.stickyKey, reservation.previousState);
+}
+function cloneRoutingInstructionState(state) {
+    if (!state) {
+        return null;
+    }
+    try {
+        const serialized = serializeRoutingInstructionState(state);
+        return deserializeRoutingInstructionState(serialized);
+    }
+    catch {
+        return null;
+    }
+}
+function resolveStopMessageSnapshot(raw) {
+    if (!raw || typeof raw !== 'object' || Array.isArray(raw)) {
+        return null;
+    }
+    const record = raw;
+    const text = typeof record.stopMessageText === 'string' ? record.stopMessageText.trim() : '';
+    const maxRepeats = typeof record.stopMessageMaxRepeats === 'number' && Number.isFinite(record.stopMessageMaxRepeats)
+        ? Math.max(1, Math.floor(record.stopMessageMaxRepeats))
+        : 0;
+    if (!text || maxRepeats <= 0) {
+        return null;
+    }
+    const used = typeof record.stopMessageUsed === 'number' && Number.isFinite(record.stopMessageUsed)
+        ? Math.max(0, Math.floor(record.stopMessageUsed))
+        : 0;
+    const updatedAt = typeof record.stopMessageUpdatedAt === 'number' && Number.isFinite(record.stopMessageUpdatedAt)
+        ? record.stopMessageUpdatedAt
+        : undefined;
+    const lastUsedAt = typeof record.stopMessageLastUsedAt === 'number' && Number.isFinite(record.stopMessageLastUsedAt)
+        ? record.stopMessageLastUsedAt
+        : undefined;
+    const source = typeof record.stopMessageSource === 'string' &&
+        record.stopMessageSource.trim()
+        ? record.stopMessageSource.trim()
+        : undefined;
+    return {
+        text,
+        maxRepeats,
+        used,
+        ...(source ? { source } : {}),
+        ...(updatedAt ? { updatedAt } : {}),
+        ...(lastUsedAt ? { lastUsedAt } : {})
+    };
+}
+function createStopMessageState(snapshot) {
+    return {
+        forcedTarget: undefined,
+        stickyTarget: undefined,
+        allowedProviders: new Set(),
+        disabledProviders: new Set(),
+        disabledKeys: new Map(),
+        disabledModels: new Map(),
+        stopMessageSource: snapshot.source && snapshot.source.trim() ? snapshot.source.trim() : 'explicit',
+        stopMessageText: snapshot.text,
+        stopMessageMaxRepeats: snapshot.maxRepeats,
+        stopMessageUsed: snapshot.used,
+        stopMessageUpdatedAt: snapshot.updatedAt,
+        stopMessageLastUsedAt: snapshot.lastUsedAt
+    };
+}
 function decorateFinalChatWithServerToolContext(chat, execution) {
     if (!execution || !execution.context) {
         return chat;
@@ -102,3 +289,142 @@ function resolveRouteHint(adapterContext, flowId) {
     }
     return routeId;
 }
+function buildServerToolLoopState(adapterContext, flowId, payload) {
+    if (!payload || typeof payload !== 'object' || Array.isArray(payload)) {
+        return null;
+    }
+    const trackPayload = typeof flowId === 'string' && flowId.trim() && flowId !== 'stop_message_flow';
+    const payloadHash = trackPayload ? hashPayload(payload) : '__servertool_auto__';
+    if (!payloadHash) {
+        return null;
+    }
+    const previous = readServerToolLoopState(adapterContext);
+    const sameFlow = previous && previous.flowId === flowId;
+    const samePayload = !trackPayload || (previous && previous.payloadHash === payloadHash);
+    const prevCount = previous && typeof previous.repeatCount === 'number' && Number.isFinite(previous.repeatCount)
+        ? Math.max(0, Math.floor(previous.repeatCount))
+        : 0;
+    const repeatCount = sameFlow && samePayload ? prevCount + 1 : 1;
+    return {
+        ...(flowId ? { flowId } : {}),
+        payloadHash,
+        repeatCount
+    };
+}
+function readServerToolLoopState(adapterContext) {
+    if (!adapterContext || typeof adapterContext !== 'object') {
+        return null;
+    }
+    const raw = adapterContext.serverToolLoopState;
+    if (!raw || typeof raw !== 'object' || Array.isArray(raw)) {
+        return null;
+    }
+    const record = raw;
+    const flowId = typeof record.flowId === 'string' ? record.flowId.trim() : undefined;
+    const payloadHash = typeof record.payloadHash === 'string' ? record.payloadHash.trim() : undefined;
+    const repeatCount = typeof record.repeatCount === 'number' && Number.isFinite(record.repeatCount)
+        ? Math.max(0, Math.floor(record.repeatCount))
+        : undefined;
+    if (!payloadHash) {
+        return null;
+    }
+    return {
+        ...(flowId ? { flowId } : {}),
+        payloadHash,
+        ...(repeatCount !== undefined ? { repeatCount } : {})
+    };
+}
+function hashPayload(payload) {
+    try {
+        const stable = stableStringify(payload);
+        return createHash('sha1').update(stable).digest('hex');
+    }
+    catch {
+        return null;
+    }
+}
+function stableStringify(value) {
+    if (value === null || typeof value !== 'object') {
+        return JSON.stringify(value);
+    }
+    if (Array.isArray(value)) {
+        return `[${value.map((entry) => stableStringify(entry)).join(',')}]`;
+    }
+    const record = value;
+    const keys = Object.keys(record).sort();
+    const entries = keys.map((key) => `${JSON.stringify(key)}:${stableStringify(record[key])}`);
+    return `{${entries.join(',')}}`;
+}
+function buildFollowupRequestId(baseRequestId, suffix) {
+    const requestId = typeof baseRequestId === 'string' ? baseRequestId : '';
+    const suffixText = typeof suffix === 'string' ? suffix : '';
+    if (!suffixText) {
+        return requestId;
+    }
+    if (!requestId) {
+        return suffixText;
+    }
+    const normalized = normalizeFollowupRequestId(requestId, suffixText);
+    if (!normalized) {
+        return suffixText;
+    }
+    return normalized.endsWith(suffixText) ? normalized : `${normalized}${suffixText}`;
+}
+function getStopMessageSource(adapterContext) {
+    if (!adapterContext || typeof adapterContext !== 'object') {
+        return undefined;
+    }
+    const raw = adapterContext.stopMessageState;
+    if (!raw || typeof raw !== 'object' || Array.isArray(raw)) {
+        return undefined;
+    }
+    const record = raw;
+    const source = typeof record.stopMessageSource === 'string' && record.stopMessageSource.trim()
+        ? record.stopMessageSource.trim()
+        : '';
+    return source || undefined;
+}
+function isAdapterClientDisconnected(adapterContext) {
+    if (!adapterContext || typeof adapterContext !== 'object') {
+        return false;
+    }
+    const state = adapterContext.clientConnectionState;
+    if (state && typeof state === 'object' && !Array.isArray(state)) {
+        const disconnected = state.disconnected;
+        if (disconnected === true) {
+            return true;
+        }
+        if (typeof disconnected === 'string' && disconnected.trim().toLowerCase() === 'true') {
+            return true;
+        }
+    }
+    const raw = adapterContext.clientDisconnected;
+    if (raw === true) {
+        return true;
+    }
+    if (typeof raw === 'string' && raw.trim().toLowerCase() === 'true') {
+        return true;
+    }
+    return false;
+}
+function normalizeFollowupRequestId(requestId, suffixText) {
+    if (!requestId) {
+        return '';
+    }
+    const token = suffixText.startsWith(':') ? suffixText.slice(1) : suffixText;
+    if (!token) {
+        return requestId;
+    }
+    const delimiterIndex = requestId.indexOf(':');
+    if (delimiterIndex === -1) {
+        return requestId;
+    }
+    const base = requestId.slice(0, delimiterIndex);
+    const rawSuffix = requestId.slice(delimiterIndex + 1);
+    if (!rawSuffix) {
+        return requestId;
+    }
+    const tokens = rawSuffix.split(':').filter((entry) => entry.length > 0 && entry !== token);
+    const rebuilt = tokens.length > 0 ? `${base}:${tokens.join(':')}` : base;
+    return rebuilt;
+}

package/dist/servertool/handlers/compaction-detect.d.ts

@@ -0,0 +1 @@
+export { isCompactionRequest } from '../../conversion/shared/compaction-detect.js';

package/dist/servertool/handlers/compaction-detect.js

@@ -0,0 +1 @@
+export { isCompactionRequest } from '../../conversion/shared/compaction-detect.js';

package/dist/servertool/handlers/gemini-empty-reply-continue.js

@@ -1,5 +1,6 @@
 import { registerServerToolHandler } from '../registry.js';
 import { cloneJson } from '../server-side-tools.js';
+import { isCompactionRequest } from './compaction-detect.js';
 const FLOW_ID = 'gemini_empty_reply_continue';
 const handler = async (ctx) => {
     if (!ctx.options.reenterPipeline) {
@@ -11,6 +12,9 @@ const handler = async (ctx) => {
     if (followupRaw === true || (typeof followupRaw === 'string' && followupRaw.trim().toLowerCase() === 'true')) {
         return null;
     }
+    if (hasCompactionFlag(adapterRecord)) {
+        return null;
+    }
     // 仅针对 gemini-chat 协议 + antigravity.* providerKey 的 /v1/responses 路径启用。
     if (ctx.options.providerProtocol !== 'gemini-chat') {
         return null;
@@ -22,7 +26,9 @@ const handler = async (ctx) => {
     const providerKey = typeof adapterRecord.providerKey === 'string' && adapterRecord.providerKey.trim()
         ? adapterRecord.providerKey.trim().toLowerCase()
         : '';
-    if (!providerKey.startsWith('antigravity.')) {
+    const isAntigravity = providerKey.startsWith('antigravity.');
+    const isGeminiCli = providerKey.startsWith('gemini-cli.');
+    if (!isAntigravity && !isGeminiCli) {
         return null;
     }
     // 仅在 finish_reason=stop 且第一条消息内容为空、无 tool_calls 时触发。
@@ -36,10 +42,13 @@ const handler = async (ctx) => {
         return null;
     }
     const first = firstRaw;
-    const finishReason = typeof first.finish_reason === 'string' && first.finish_reason.trim()
+    const finishReasonRaw = typeof first.finish_reason === 'string' && first.finish_reason.trim()
         ? first.finish_reason.trim()
         : '';
-    if (finishReason !== 'stop') {
+    const finishReason = finishReasonRaw.toLowerCase();
+    const isStop = finishReason === 'stop';
+    const isMaxTokens = finishReason === 'length'; // 映射自 Gemini 的 MAX_TOKENS
+    if (!isStop && !isMaxTokens) {
         return null;
     }
     const message = first.message && typeof first.message === 'object' && !Array.isArray(first.message)
@@ -50,7 +59,9 @@ const handler = async (ctx) => {
     }
     const contentRaw = message.content;
     const contentText = typeof contentRaw === 'string' ? contentRaw.trim() : '';
-    if (contentText.length > 0) {
+    // 对于 finish_reason=stop，仅在真正“空回复”时触发；
+    // 对于 finish_reason=length（MAX_TOKENS 截断），允许已有内容，视为需要自动续写。
+    if (isStop && contentText.length > 0) {
         return null;
     }
     const toolCalls = Array.isArray(message.tool_calls) ? message.tool_calls : [];
@@ -67,6 +78,9 @@ const handler = async (ctx) => {
     if (!captured) {
         return null;
     }
+    if (isCompactionRequest(captured)) {
+        return null;
+    }
     // 超过最多 3 次空回复：返回一个 HTTP_HANDLER_ERROR 形状的错误，交由上层错误中心处理。
     if (nextCount > 3) {
         const errorChat = {
@@ -129,7 +143,7 @@ function buildContinueFollowupPayload(source) {
     const messages = Array.isArray(rawMessages) ? cloneJson(rawMessages) : [];
     messages.push({
         role: 'user',
-        content: '继续'
+        content: '继续执行'
     });
     payload.messages = messages;
     if (Array.isArray(source.tools) && source.tools.length) {
@@ -142,3 +156,13 @@ function buildContinueFollowupPayload(source) {
     }
     return payload;
 }
+function hasCompactionFlag(record) {
+    const flag = record.compactionRequest;
+    if (flag === true) {
+        return true;
+    }
+    if (typeof flag === 'string' && flag.trim().toLowerCase() === 'true') {
+        return true;
+    }
+    return false;
+}

package/dist/servertool/handlers/iflow-model-error-retry.js

@@ -1,5 +1,6 @@
 import { registerServerToolHandler } from '../registry.js';
 import { cloneJson } from '../server-side-tools.js';
+import { isCompactionRequest } from './compaction-detect.js';
 const FLOW_ID = 'iflow_model_error_retry';
 const handler = async (ctx) => {
     if (!ctx.options.reenterPipeline) {
@@ -11,6 +12,9 @@ const handler = async (ctx) => {
     if (followupRaw === true || (typeof followupRaw === 'string' && followupRaw.trim().toLowerCase() === 'true')) {
         return null;
     }
+    if (hasCompactionFlag(adapterRecord)) {
+        return null;
+    }
     // 仅针对 openai-chat 协议 + iflow.* providerKey 的 /v1/responses 路径启用。
     if (ctx.options.providerProtocol !== 'openai-chat') {
         return null;
@@ -39,6 +43,9 @@ const handler = async (ctx) => {
     if (!captured) {
         return null;
     }
+    if (isCompactionRequest(captured)) {
+        return null;
+    }
     const followupPayload = buildRetryFollowupPayload(captured);
     if (!followupPayload) {
         return null;
@@ -91,3 +98,13 @@ function buildRetryFollowupPayload(source) {
     }
     return payload;
 }
+function hasCompactionFlag(record) {
+    const flag = record.compactionRequest;
+    if (flag === true) {
+        return true;
+    }
+    if (typeof flag === 'string' && flag.trim().toLowerCase() === 'true') {
+        return true;
+    }
+    return false;
+}