@jsonstudio/llms 0.6.1399 → 0.6.1403

package/dist/conversion/hub/pipeline/stages/req_inbound/req_inbound_stage3_context_capture/index.js

@@ -1,7 +1,29 @@
 import { captureResponsesContext, buildChatRequestFromResponses } from '../../../../../responses/responses-openai-bridge.js';
 import { captureResponsesRequestContext } from '../../../../../shared/responses-conversation-store.js';
+import { readRuntimeMetadata } from '../../../../../shared/runtime-metadata.js';
 import { recordStage } from '../../../stages/utils.js';
 export async function runReqInboundStage3ContextCapture(options) {
+    // 对由 server-side 工具触发的二跳/内部跳转请求（例如 stopMessage followup），
+    // 跳过工具输出扫描与 apply_patch 诊断日志，避免在内部流中重复放大客户端已收到的
+    // 工具错误信息。此类请求的工具治理在 chat-process 阶段完成，这里仅保留最小快照。
+    try {
+        const ctx = options.adapterContext;
+        const rt = readRuntimeMetadata(ctx);
+        const followupFlag = rt?.serverToolFollowup;
+        const isFollowup = followupFlag === true ||
+            (typeof followupFlag === 'string' && followupFlag.trim().toLowerCase() === 'true');
+        if (isFollowup) {
+            const snapshot = {
+                providerProtocol: options.adapterContext.providerProtocol ?? 'unknown',
+                tool_outputs: []
+            };
+            recordStage(options.stageRecorder, 'chat_process.req.stage3.context_capture', snapshot);
+            return snapshot;
+        }
+    }
+    catch {
+        // best-effort: 若检测 serverToolFollowup 失败，则继续走通用路径
+    }
     let context;
     if (options.captureContext) {
         try {

package/dist/conversion/hub/policy/policy-engine.js

@@ -20,6 +20,43 @@ function shouldSample(rate) {
 function isJsonRecord(value) {
     return Boolean(value) && typeof value === 'object' && !Array.isArray(value);
 }
+function hasGeminiCliRequestWrapper(payload) {
+    const request = payload.request;
+    if (!isJsonRecord(request)) {
+        return false;
+    }
+    const req = request;
+    return (Object.prototype.hasOwnProperty.call(req, 'contents') ||
+        Object.prototype.hasOwnProperty.call(req, 'systemInstruction') ||
+        Object.prototype.hasOwnProperty.call(req, 'tools') ||
+        Object.prototype.hasOwnProperty.call(req, 'toolConfig') ||
+        Object.prototype.hasOwnProperty.call(req, 'generationConfig') ||
+        Object.prototype.hasOwnProperty.call(req, 'safetySettings'));
+}
+function isGeminiCliPayload(payload) {
+    const record = payload;
+    return (typeof record.project === 'string' ||
+        typeof record.requestType === 'string' ||
+        typeof record.userAgent === 'string' ||
+        typeof record.requestId === 'string' ||
+        hasGeminiCliRequestWrapper(payload));
+}
+function extendGeminiAllowlistIfNeeded(providerProtocol, payload, allowedTopLevelKeys) {
+    if (providerProtocol !== 'gemini-chat') {
+        return allowedTopLevelKeys;
+    }
+    if (!isGeminiCliPayload(payload)) {
+        return allowedTopLevelKeys;
+    }
+    const extended = allowedTopLevelKeys ? new Set(allowedTopLevelKeys) : new Set();
+    extended.add('request');
+    extended.add('project');
+    extended.add('requestId');
+    extended.add('requestType');
+    extended.add('userAgent');
+    extended.add('action');
+    return extended;
+}
 function applyProviderOutboundPolicy(providerProtocol, payload) {
     const removedTopLevelKeys = [];
     const flattenedWrappers = [];
@@ -43,6 +80,8 @@ function applyProviderOutboundPolicy(providerProtocol, payload) {
         spec.providerOutbound.enforceAllowedTopLevelKeys === true
         ? new Set(spec.providerOutbound.allowedTopLevelKeys)
         : undefined;
+    const isGeminiCli = providerProtocol === 'gemini-chat' && isGeminiCliPayload(payload);
+    const allowedTopLevelKeysResolved = extendGeminiAllowlistIfNeeded(providerProtocol, payload, allowedTopLevelKeys);
     // Reserved/private keys must never be sent upstream.
     for (const key of Object.keys(payload)) {
         if (spec.providerOutbound.reservedKeyPrefixes.some((prefix) => key.startsWith(prefix))) {
@@ -57,6 +96,9 @@ function applyProviderOutboundPolicy(providerProtocol, payload) {
         if (!wrapperKey || typeof wrapperKey !== 'string') {
             continue;
         }
+        if (isGeminiCli && wrapperKey === 'request') {
+            continue;
+        }
         if (!isJsonRecord(out[wrapperKey])) {
             continue;
         }
@@ -84,9 +126,9 @@ function applyProviderOutboundPolicy(providerProtocol, payload) {
     // Enforce protocol allowlist (top-level). Only runs when explicitly enabled
     // for this protocol, and only after wrapper flatten so allowed fields are
     // present at the correct level.
-    if (allowedTopLevelKeys) {
+    if (allowedTopLevelKeysResolved) {
         for (const key of Object.keys(out)) {
-            if (allowedTopLevelKeys.has(key))
+            if (allowedTopLevelKeysResolved.has(key))
                 continue;
             ensureOutClone();
             delete out[key];
@@ -106,13 +148,18 @@ function observeProviderPayload(options) {
     // Do NOT modify payload here.
     const spec = resolveHubProtocolSpec(options.providerProtocol);
     const allowlistEnabled = options.phase === 'provider_outbound';
+    const isGeminiCli = options.providerProtocol === 'gemini-chat' && isGeminiCliPayload(options.payload);
     const allowedTopLevelKeys = allowlistEnabled && Array.isArray(spec.providerOutbound.allowedTopLevelKeys)
         ? new Set(spec.providerOutbound.allowedTopLevelKeys)
         : undefined;
+    const allowedTopLevelKeysResolved = extendGeminiAllowlistIfNeeded(options.providerProtocol, options.payload, allowedTopLevelKeys);
     for (const rule of spec.providerOutbound.forbidWrappers) {
         if (rule.code !== 'forbid_wrapper') {
             continue;
         }
+        if (isGeminiCli && rule.path === 'request') {
+            continue;
+        }
         if (rule.path in options.payload && isJsonRecord(options.payload[rule.path])) {
             violations.push({
                 code: 'unexpected_wrapper',
@@ -130,7 +177,7 @@ function observeProviderPayload(options) {
             });
             continue;
         }
-        if (allowedTopLevelKeys && !allowedTopLevelKeys.has(key)) {
+        if (allowedTopLevelKeysResolved && !allowedTopLevelKeysResolved.has(key)) {
             violations.push({
                 code: 'unexpected_field',
                 path: key,

package/dist/conversion/hub/process/chat-process.js

@@ -112,7 +112,7 @@ async function applyRequestToolGovernance(request, context) {
     // Server-side web_search tool injection (config-driven, best-effort).
     merged = applyHubOperations(merged, buildWebSearchOperations(merged, metadata));
     // Server-side clock tool + scheduled reminders injection (config-driven, best-effort).
-    merged = applyHubOperations(merged, buildClockOperations(merged, metadata));
+    merged = applyHubOperations(merged, buildClockOperations(metadata));
     merged = await maybeInjectClockRemindersAndApplyDirectives(merged, metadata, context.requestId);
     const { request: sanitized, summary } = toolGovernanceEngine.governRequest(merged, providerProtocol);
     if (summary.applied) {
@@ -124,129 +124,12 @@ async function applyRequestToolGovernance(request, context) {
             }
         };
     }
-    return pruneUndeclaredToolHistory(sanitized);
+    return sanitized;
 }
 function resolveSessionIdForPending(metadata, request) {
     const candidate = readString(metadata.sessionId) ?? readString(request.metadata?.sessionId);
     return candidate && candidate.trim() ? candidate.trim() : null;
 }
-function pruneUndeclaredToolHistory(request) {
-    const tools = Array.isArray(request.tools) ? request.tools : [];
-    const allowedToolNames = new Set();
-    for (const tool of tools) {
-        const name = tool && typeof tool === 'object' && tool.function && typeof tool.function.name === 'string'
-            ? String(tool.function.name).trim()
-            : '';
-        if (name) {
-            allowedToolNames.add(name);
-        }
-    }
-    // If no tools are declared, do not carry any tool calls/results in history.
-    const hasAnyTools = allowedToolNames.size > 0;
-    let prunedToolCalls = 0;
-    let prunedToolResults = 0;
-    const prunedToolNames = new Set();
-    const keptToolCallIds = new Set();
-    const nextMessages = [];
-    for (const msg of request.messages) {
-        if (!msg || typeof msg !== 'object')
-            continue;
-        if (msg.role === 'assistant' && Array.isArray(msg.tool_calls) && msg.tool_calls.length) {
-            if (!hasAnyTools) {
-                prunedToolCalls += msg.tool_calls.length;
-                for (const tc of msg.tool_calls) {
-                    const n = tc && tc.function && typeof tc.function.name === 'string' ? tc.function.name.trim() : '';
-                    if (n)
-                        prunedToolNames.add(n);
-                }
-                // Drop tool_calls when no tools are declared.
-                const keptContent = msg.content;
-                const hasContent = typeof keptContent === 'string'
-                    ? keptContent.trim().length > 0
-                    : Array.isArray(keptContent)
-                        ? keptContent.length > 0
-                        : false;
-                if (hasContent) {
-                    nextMessages.push({ ...msg, tool_calls: [] });
-                }
-                continue;
-            }
-            const keptCalls = [];
-            for (const tc of msg.tool_calls) {
-                if (!tc || typeof tc !== 'object')
-                    continue;
-                const name = tc.function && typeof tc.function.name === 'string' ? tc.function.name.trim() : '';
-                if (!name || !allowedToolNames.has(name)) {
-                    prunedToolCalls += 1;
-                    if (name)
-                        prunedToolNames.add(name);
-                    continue;
-                }
-                keptCalls.push(tc);
-                if (typeof tc.id === 'string' && tc.id.trim()) {
-                    keptToolCallIds.add(tc.id.trim());
-                }
-            }
-            const keptContent = msg.content;
-            const hasContent = typeof keptContent === 'string'
-                ? keptContent.trim().length > 0
-                : Array.isArray(keptContent)
-                    ? keptContent.length > 0
-                    : false;
-            if (!hasContent && keptCalls.length === 0) {
-                continue;
-            }
-            if (keptCalls.length === msg.tool_calls.length) {
-                nextMessages.push(msg);
-            }
-            else {
-                nextMessages.push({ ...msg, tool_calls: keptCalls.length ? keptCalls : undefined });
-            }
-            continue;
-        }
-        if (msg.role === 'tool') {
-            const name = typeof msg.name === 'string' ? msg.name.trim() : '';
-            const toolCallId = typeof msg.tool_call_id === 'string' ? msg.tool_call_id.trim() : '';
-            if (!hasAnyTools) {
-                prunedToolResults += 1;
-                if (name)
-                    prunedToolNames.add(name);
-                continue;
-            }
-            if (name && !allowedToolNames.has(name)) {
-                prunedToolResults += 1;
-                prunedToolNames.add(name);
-                continue;
-            }
-            // If the result references a tool_call_id that no longer exists, drop it to avoid
-            // provider-side strict pairing errors (Gemini/Anthropic tool pairing).
-            if (toolCallId && !keptToolCallIds.has(toolCallId)) {
-                prunedToolResults += 1;
-                if (name)
-                    prunedToolNames.add(name);
-                continue;
-            }
-            nextMessages.push(msg);
-            continue;
-        }
-        nextMessages.push(msg);
-    }
-    if (prunedToolCalls === 0 && prunedToolResults === 0) {
-        return request;
-    }
-    return {
-        ...request,
-        messages: nextMessages,
-        metadata: {
-            ...request.metadata,
-            toolHistoryPrune: {
-                prunedToolCalls,
-                prunedToolResults,
-                prunedToolNames: Array.from(prunedToolNames).sort()
-            }
-        }
-    };
-}
 function extractToolCallIdsFromToolMessages(messages) {
     const ids = new Set();
     for (const msg of messages) {
@@ -613,32 +496,8 @@ function buildToolChoiceOperations(toolChoice) {
         }
     ];
 }
-function isGeminiAntigravityLike(metadata) {
-    const providerProtocol = readString(metadata.providerProtocol) ?? readString(metadata.provider) ?? '';
-    if (!providerProtocol.toLowerCase().includes('gemini')) {
-        return false;
-    }
-    const rt = readRuntimeMetadata(metadata);
-    const providerId = readString(metadata.providerId) ??
-        readString(rt?.providerId) ??
-        '';
-    const providerKey = readString(metadata.providerKey) ??
-        readString(rt?.providerKey) ??
-        '';
-    const loweredId = providerId.toLowerCase();
-    const loweredKey = providerKey.toLowerCase();
-    return (loweredId.includes('antigravity') ||
-        loweredId === 'gemini-cli' ||
-        loweredKey.startsWith('antigravity.') ||
-        loweredKey.startsWith('gemini-cli.'));
-}
 function buildWebSearchOperations(request, metadata) {
     const rt = readRuntimeMetadata(metadata);
-    // gcli2api-style: do not force-inject server-side web_search tools for Gemini/Antigravity.
-    // Tools must be present only when supplied by the client payload/config, not auto-appended.
-    if (isGeminiAntigravityLike(metadata)) {
-        return [];
-    }
     // ServerTool 二/三跳（serverToolFollowup=true）不再注入 web_search 工具，
     // 以避免在 web_search 流程内部形成循环命中。
     if (rt?.serverToolFollowup === true) {
@@ -750,10 +609,10 @@ function buildWebSearchOperations(request, metadata) {
     });
     return ops;
 }
-function buildClockOperations(request, metadata) {
+function buildClockOperations(metadata) {
     const rt = readRuntimeMetadata(metadata);
-    // gcli2api-style: do not force-inject server-side clock tools for Gemini/Antigravity.
-    if (isGeminiAntigravityLike(metadata)) {
+    // Do not inject clock tool into internal servertool followup hops.
+    if (rt?.serverToolFollowup === true) {
         return [];
     }
     const rawConfig = rt?.clock;

package/dist/conversion/hub/response/provider-response.js

@@ -8,7 +8,6 @@ import { OpenAIChatResponseMapper, ResponsesResponseMapper, AnthropicResponseMap
 import { runRespInboundStage1SseDecode } from '../pipeline/stages/resp_inbound/resp_inbound_stage1_sse_decode/index.js';
 import { runRespInboundStage2FormatParse } from '../pipeline/stages/resp_inbound/resp_inbound_stage2_format_parse/index.js';
 import { runRespInboundStage3SemanticMap } from '../pipeline/stages/resp_inbound/resp_inbound_stage3_semantic_map/index.js';
-import { runRespInboundStage3ThoughtSignatureCapture } from '../pipeline/stages/resp_inbound/resp_inbound_stage3_thought_signature_capture/index.js';
 import { runRespInboundStageCompatResponse } from '../pipeline/stages/req_outbound/req_outbound_stage3_compat/index.js';
 import { runRespProcessStage1ToolGovernance } from '../pipeline/stages/resp_process/resp_process_stage1_tool_governance/index.js';
 import { runRespProcessStage2Finalize } from '../pipeline/stages/resp_process/resp_process_stage2_finalize/index.js';
@@ -260,14 +259,21 @@ export async function convertProviderResponse(options) {
     // Client protocol remapping happens only on the outermost request.
     const clientProtocol = isFollowup ? 'openai-chat' : resolveClientProtocol(options.entryEndpoint);
     const hasServerToolSupport = Boolean(options.providerInvoker) || Boolean(options.reenterPipeline);
-    // ServerTool followup hop: always return a fully materialized ChatCompletion JSON (non-SSE).
-    // This is an internal hop used by the servertool engine, and it must be deterministic.
+    // 是否跳过 ServerTool 编排：
+    // - Provider 不支持 ServerTool（没有 invoker/reenterPipeline）时跳过；
+    // - 对于 serverToolFollowup=true 的二/三跳内部请求，必须跳过：
+    //   followup 的语义是“本轮 ServerTool 已决定并发起下一跳”，下一跳响应不应再次触发 ServerTool，
+    //   否则会形成嵌套/循环（例如 stop_message_flow 反复触发）。
+    const skipServerTools = isFollowup || !hasServerToolSupport;
+    // 对于由 server-side 工具触发的内部跳转（二跳/三跳），统一禁用 SSE 聚合输出，
+    // 始终返回完整的 ChatCompletion JSON，便于在 llms 内部直接解析，而不是拿到
+    // __sse_responses 可读流。
     const wantsStream = isFollowup ? false : options.wantsStream;
     try {
         // eslint-disable-next-line no-console
         console.log(`\x1b[38;5;33m[servertool][orchestrator][debug] requestId=${options.context.requestId} ` +
             `protocol=${options.providerProtocol} endpoint=${options.entryEndpoint} ` +
-            `serverTools=${hasServerToolSupport} hasInvoker=${Boolean(options.providerInvoker)} ` +
+            `skipServerTools=${skipServerTools} hasInvoker=${Boolean(options.providerInvoker)} ` +
             `hasReenter=${Boolean(options.reenterPipeline)}\x1b[0m`);
     }
     catch {
@@ -300,11 +306,6 @@ export async function convertProviderResponse(options) {
         stageRecorder: options.stageRecorder
     });
     stripInternalPolicyDebugFields(formatEnvelope.payload);
-    formatEnvelope.payload = await runRespInboundStage3ThoughtSignatureCapture({
-        payload: formatEnvelope.payload,
-        adapterContext: options.context,
-        stageRecorder: options.stageRecorder
-    });
     // Phase 2 (shadow): response tool surface mismatch detection (provider inbound).
     // Only records diffs; does not rewrite payload.
     try {
@@ -352,7 +353,7 @@ export async function convertProviderResponse(options) {
     // 检查是否需要进行 ServerTool 编排
     // 使用新的 ChatEnvelope 级别的 servertool 实现
     let effectiveChatResponse = chatResponse;
-    if (hasServerToolSupport) {
+    if (!skipServerTools && options.reenterPipeline) {
         try {
             // eslint-disable-next-line no-console
             console.log(`\x1b[38;5;33m[servertool][orchestrator] start requestId=${options.context.requestId} ` +

package/dist/conversion/hub/response/response-mappers.d.ts

@@ -20,7 +20,5 @@ export declare class AnthropicResponseMapper implements ResponseMapper {
     }): ChatCompletionLike;
 }
 export declare class GeminiResponseMapper implements ResponseMapper {
-    toChatCompletion(format: FormatEnvelope, ctx: AdapterContext, options?: {
-        requestSemantics?: ChatSemantics | JsonObject;
-    }): ChatCompletionLike;
+    toChatCompletion(format: FormatEnvelope, _ctx: AdapterContext): ChatCompletionLike;
 }

package/dist/conversion/hub/response/response-mappers.js

@@ -18,9 +18,8 @@ export class AnthropicResponseMapper {
     }
 }
 export class GeminiResponseMapper {
-    toChatCompletion(format, ctx, options) {
-        const aliasMap = extractToolNameAliasMapFromSemantics(options?.requestSemantics) ?? extractGeminiToolNameAliasMapFromContext(ctx);
-        return buildOpenAIChatFromGeminiResponse(format.payload ?? {}, { aliasMap });
+    toChatCompletion(format, _ctx) {
+        return buildOpenAIChatFromGeminiResponse(format.payload ?? {});
     }
 }
 function extractToolNameAliasMapFromSemantics(semantics) {
@@ -48,20 +47,3 @@ function extractToolNameAliasMapFromSemantics(semantics) {
     }
     return Object.keys(normalized).length ? normalized : undefined;
 }
-function extractGeminiToolNameAliasMapFromContext(ctx) {
-    const candidate = ctx?.geminiToolNameAliasMap;
-    if (!candidate || typeof candidate !== 'object' || Array.isArray(candidate)) {
-        return undefined;
-    }
-    const normalized = {};
-    for (const [key, value] of Object.entries(candidate)) {
-        if (typeof key !== 'string' || typeof value !== 'string')
-            continue;
-        const trimmedKey = key.trim();
-        const trimmedValue = value.trim();
-        if (!trimmedKey.length || !trimmedValue.length)
-            continue;
-        normalized[trimmedKey] = trimmedValue;
-    }
-    return Object.keys(normalized).length ? normalized : undefined;
-}

package/dist/conversion/hub/tool-surface/tool-surface-engine.js

@@ -477,7 +477,8 @@ export function applyProviderOutboundToolSurface(args) {
         // best-effort only
     }
     const totalDiffCount = (schemaDiff.diffCount || 0) + (historyDiff.diffCount || 0);
-    if (totalDiffCount > 0) {
+    const shouldRecord = totalDiffCount > 0 || Boolean(candidate?.candidateTools) || Boolean(historyReason);
+    if (shouldRecord) {
         const diffHeadMerged = [];
         for (const entry of schemaDiff.diffHead) {
             if (diffHeadMerged.length >= 10)

package/dist/conversion/responses/responses-openai-bridge.js

@@ -170,9 +170,10 @@ function stripRoutingTagsFromText(text) {
     if (typeof text !== 'string') {
         return typeof text === 'string' ? text : String(text ?? '');
     }
-    // 移除形如 <** ... **> 的路由指令标记（例如 stopMessage/!provider 等），
-    // 避免泄露到上游 provider 或持久化的 Responses 历史中。
-    return text.replace(/<\*\*[^*]+\*\*>/g, '').trim();
+    // NOTE: Do not strip <**...**> here. Routing instructions must reach the
+    // virtual router so allowlist/sticky directives can be applied. The router
+    // will clean them from messages after parsing.
+    return text;
 }
 function sanitizeBridgeHistory(history) {
     if (!history || typeof history !== 'object') {

package/dist/conversion/shared/gemini-tool-utils.d.ts

@@ -1,12 +1,7 @@
 import type { BridgeToolDefinition } from './bridge-message-types.js';
 import type { MissingField } from '../hub/types/chat-envelope.js';
 import type { JsonValue, JsonObject } from '../hub/types/json.js';
-export declare function isGeminiFunctionNameSafe(name: string): boolean;
-export declare function sanitizeGeminiFunctionName(name: string, options?: {
-    maxLen?: number;
-}): string;
-export type GeminiToolSchemaMode = 'default' | 'antigravity';
 export declare function prepareGeminiToolsForBridge(rawTools: JsonValue | undefined, missing?: MissingField[]): BridgeToolDefinition[] | undefined;
 export declare function buildGeminiToolsFromBridge(defs: BridgeToolDefinition[] | undefined, options?: {
-    mode?: GeminiToolSchemaMode;
+    mode?: 'antigravity' | 'default';
 }): JsonObject[] | undefined;