llm-cli-gateway 2.11.0 → 2.12.0

package/dist/async-job-manager.js

@@ -2,12 +2,12 @@ import { randomUUID } from "crypto";
 import { envWithExtendedPath, getExtendedPath, killProcessGroup, providerCommandName, spawnCliProcess, unregisterProcessGroup, } from "./executor.js";
 import { noopLogger, logWarn } from "./logger.js";
 import { ProcessMonitor } from "./process-monitor.js";
-import { computeRequestKey } from "./job-store.js";
+import { computeRequestKey, isValidationRunStore } from "./job-store.js";
 import { NoopFlightRecorder, } from "./flight-recorder.js";
 import { codexFrResponse } from "./codex-json-parser.js";
 import { getRequestContext, resolveOwnerPrincipal } from "./request-context.js";
 import { runApiRequest, ApiHttpError, } from "./api-provider.js";
-function extractApiHttpStatus(error) {
+export function extractApiHttpStatus(error) {
     for (const candidate of [error, error?.cause]) {
         if (candidate instanceof ApiHttpError && typeof candidate.status === "number") {
             return candidate.status;
@@ -18,6 +18,14 @@ function extractApiHttpStatus(error) {
     }
     return null;
 }
+export function extractApiErrorBody(error) {
+    for (const candidate of [error, error?.cause]) {
+        if (candidate instanceof ApiHttpError && candidate.responseText) {
+            return candidate.responseText;
+        }
+    }
+    return undefined;
+}
 const MAX_OUTPUT_SIZE = 50 * 1024 * 1024;
 const JOB_TTL_MS = 60 * 60 * 1000;
 const EVICTION_INTERVAL_MS = 5 * 60 * 1000;
@@ -173,6 +181,9 @@ export class AsyncJobManager {
     hasStore() {
         return this.store !== null;
     }
+    getValidationRunStore() {
+        return this.store && isValidationRunStore(this.store) ? this.store : null;
+    }
     emitMetrics(job) {
         if (job.metricsRecorded)
             return;
@@ -404,6 +415,9 @@ export class AsyncJobManager {
             job.stdout = result.text;
             job.httpStatus = result.httpStatus;
             job.exitCode = 0;
+            job.apiUsage = result.usage;
+            job.apiResponseId = result.responseId ?? null;
+            job.apiModel = result.model;
         }
         else {
             job.status = "failed";
@@ -413,6 +427,7 @@ export class AsyncJobManager {
             job.stderr = message;
             job.error = message;
             job.exitCode = 1;
+            job.apiErrorBody = extractApiErrorBody(error);
         }
         job.finishedAt = new Date().toISOString();
         job.exited = true;
@@ -443,7 +458,13 @@ export class AsyncJobManager {
         if (job.flightRecorderComplete)
             return;
         const durationMs = Math.max(0, Date.now() - new Date(job.startedAt).getTime());
-        const usage = finalStatus === "completed" && job.extractUsage ? this.safeExtractUsage(job) : {};
+        const usage = finalStatus !== "completed"
+            ? {}
+            : job.transport === "http"
+                ? this.httpUsage(job)
+                : job.extractUsage
+                    ? this.safeExtractUsage(job)
+                    : {};
         const isFailure = finalStatus === "failed";
         let response;
         if (job.transport === "process" && job.cli === "codex") {
@@ -491,6 +512,17 @@ export class AsyncJobManager {
             return {};
         }
     }
+    httpUsage(job) {
+        const u = job.apiUsage;
+        if (!u)
+            return {};
+        return {
+            inputTokens: u.inputTokens,
+            outputTokens: u.outputTokens,
+            cacheReadTokens: u.cacheReadTokens,
+            costUsd: u.costUsd,
+        };
+    }
     armFlightCompleteForDeferral(jobId) {
         const job = this.jobs.get(jobId);
         if (!job)
@@ -846,6 +878,15 @@ export class AsyncJobManager {
             stderr: stderr.text,
             stdoutTruncated: stdout.truncated,
             stderrTruncated: stderr.truncated,
+            ...(job.transport === "http"
+                ? {
+                    apiUsage: job.apiUsage,
+                    httpStatus: job.httpStatus,
+                    responseId: job.apiResponseId,
+                    model: job.apiModel,
+                    errorBody: job.apiErrorBody,
+                }
+                : {}),
         };
     }
     cancelJob(jobId) {

package/dist/config.d.ts

@@ -76,6 +76,7 @@ export interface ApiProviderConfig {
     baseUrl: string;
     defaultModel: string;
     models?: string[];
+    usageInclude?: boolean;
 }
 export interface ApiProviderRuntime {
     name: string;
@@ -83,6 +84,7 @@ export interface ApiProviderRuntime {
     baseUrl: string;
     defaultModel: string;
     models?: string[];
+    usageInclude?: boolean;
     apiKey: string;
 }
 export interface ProvidersConfig {

package/dist/config.js

@@ -301,6 +301,7 @@ const ApiProviderSchema = z
     api_key_env: z.string().min(1).optional(),
     default_model: z.string().min(1),
     models: z.array(z.string().min(1)).nonempty().optional(),
+    usage_include: z.boolean().optional(),
 })
     .strict();
 function readProvidersFile(configPath, logger) {
@@ -369,6 +370,7 @@ export function loadProvidersConfig(logger = noopLogger) {
             baseUrl: parsed.data.base_url,
             defaultModel: parsed.data.default_model,
             models: parsed.data.models ? [...parsed.data.models] : undefined,
+            usageInclude: parsed.data.usage_include,
         };
     }
     return { xai, providers, sources: { configFile: sourcePath } };
@@ -395,6 +397,7 @@ export function enabledApiProviders(config, env = process.env) {
             baseUrl: provider.baseUrl,
             defaultModel: provider.defaultModel,
             models: provider.models,
+            usageInclude: provider.usageInclude,
             apiKey: resolveProviderKey(provider.apiKeyEnv, env) ?? "",
         });
     }

package/dist/index.d.ts

@@ -5,7 +5,6 @@ import { ISessionManager, type CliType, type ProviderType } from "./session-mana
 import { ResourceProvider } from "./resources.js";
 import { PerformanceMetrics } from "./metrics.js";
 import { type PersistenceConfig, type CacheAwarenessConfig, type ProvidersConfig, type ApiProviderRuntime, type AcpConfig } from "./config.js";
-import { type XaiReasoningEffort } from "./xai-api-provider.js";
 import { DatabaseConnection } from "./db.js";
 import { AsyncJobManager } from "./async-job-manager.js";
 import { ApprovalManager, ApprovalRecord } from "./approval-manager.js";
@@ -63,6 +62,7 @@ export declare const WORKTREE_SCHEMA: z.ZodUnion<[z.ZodBoolean, z.ZodObject<{
 export declare const WORKSPACE_ALIAS_SCHEMA: z.ZodString;
 export declare const SESSION_PROVIDER_VALUES: readonly ["claude", "codex", "gemini", "grok", "mistral", "devin", "grok-api"];
 export declare const SESSION_PROVIDER_ENUM: z.ZodEnum<["claude", "codex", "gemini", "grok", "mistral", "devin", "grok-api"]>;
+export declare function sessionProviderValuesFor(providers: ProvidersConfig): string[];
 export type SessionProvider = ProviderType;
 export interface GatewayServerDeps {
     sessionManager?: ISessionManager;
@@ -120,6 +120,25 @@ export declare function resolveWorktreeForRequest(worktreeOpt: boolean | {
     workspaceRoot?: string;
 }): Promise<ResolvedWorktree>;
 export declare function formatWorktreePrefix(worktreePath?: string): string;
+export declare function createErrorResponse(cli: string, code: number, stderr: string, correlationId?: string, error?: Error, apiError?: {
+    httpStatus?: number | null;
+    responseBody?: string;
+}): {
+    content: {
+        type: "text";
+        text: string;
+    }[];
+    isError: boolean;
+    structuredContent: {
+        httpStatus?: number | undefined;
+        responseBody?: string | undefined;
+        response: string;
+        correlationId: string | null;
+        cli: string;
+        exitCode: number;
+        errorCategory: string;
+    };
+};
 export declare function extractUsageAndCost(cli: "claude" | "codex" | "gemini" | "grok" | "mistral" | "devin", output: string, outputFormat?: string, ctx?: {
     sessionId?: string;
     home?: string;
@@ -232,6 +251,8 @@ export declare function prepareGeminiRequest(params: {
     attachments?: string[];
     skipTrust?: boolean;
     yolo?: boolean;
+    project?: string;
+    newProject?: boolean;
 }, runtime?: GatewayServerRuntime): CliRequestPrep | ExtendedToolResponse;
 export declare function prepareGrokRequest(params: {
     prompt?: string;
@@ -278,6 +299,9 @@ export declare function prepareGrokRequest(params: {
     restoreCode?: boolean;
     leaderSocket?: string;
     nativeWorktree?: boolean | string;
+    worktreeRef?: string;
+    forkSession?: boolean;
+    jsonSchema?: string | Record<string, unknown>;
 }, runtime?: GatewayServerRuntime): CliRequestPrep | ExtendedToolResponse;
 export declare function prepareMistralRequest(params: {
     prompt?: string;
@@ -309,6 +333,7 @@ export declare function buildMistralRetryPrep(params: Pick<MistralRequestParams,
     env: Record<string, string>;
     ignoredDisallowedTools: boolean;
 };
+export declare function buildCliResponse(cli: "claude" | "codex" | "gemini" | "grok" | "mistral" | "devin", stdout: string, optimizeResponse: boolean, corrId: string, sessionId: string | undefined, prep: CliRequestPrep, durationMs: number, resumable?: boolean, outputFormat?: string, warnings?: WarningEntry[]): ExtendedToolResponse;
 export interface GrokApiRequestParams {
     prompt?: string;
     promptParts?: PromptParts;
@@ -321,12 +346,13 @@ export interface GrokApiRequestParams {
     maxOutputTokens?: number;
     temperature?: number;
     topP?: number;
-    reasoningEffort?: XaiReasoningEffort;
+    reasoningEffort?: "none" | "low" | "medium" | "high";
     timeoutMs?: number;
 }
 export declare function handleGrokApiRequest(deps: HandlerDeps, params: GrokApiRequestParams): Promise<ExtendedToolResponse>;
 interface ApiProviderToolParams {
     prompt?: string;
+    promptParts?: PromptParts;
     system?: string;
     model?: string;
     correlationId?: string;
@@ -335,6 +361,11 @@ interface ApiProviderToolParams {
     topP?: number;
     reasoningEffort?: "none" | "low" | "medium" | "high";
     timeoutMs?: number;
+    optimizePrompt?: boolean;
+    optimizeResponse?: boolean;
+    forceRefresh?: boolean;
+    sessionId?: string;
+    createNewSession?: boolean;
 }
 export declare function handleApiProviderRequest(runtimeArg: GatewayServerRuntime, providerRuntime: ApiProviderRuntime, params: ApiProviderToolParams): Promise<ExtendedToolResponse>;
 export declare function handleApiProviderRequestAsync(runtimeArg: GatewayServerRuntime, providerRuntime: ApiProviderRuntime, params: ApiProviderToolParams): ExtendedToolResponse;
@@ -364,6 +395,8 @@ export interface GeminiRequestParams {
     attachments?: string[];
     skipTrust?: boolean;
     yolo?: boolean;
+    project?: string;
+    newProject?: boolean;
     workspace?: string;
     worktree?: boolean | {
         name?: string;
@@ -437,6 +470,9 @@ export interface GrokRequestParams {
     restoreCode?: boolean;
     leaderSocket?: string;
     nativeWorktree?: boolean | string;
+    worktreeRef?: string;
+    forkSession?: boolean;
+    jsonSchema?: string | Record<string, unknown>;
     workspace?: string;
     worktree?: boolean | {
         name?: string;
@@ -448,7 +484,7 @@ export declare function handleGrokRequestAsync(deps: AsyncHandlerDeps, params: O
 export interface DevinRequestParams {
     prompt?: string;
     model?: string;
-    permissionMode?: "normal" | "auto" | "dangerous" | "yolo" | "bypass";
+    permissionMode?: "auto" | "smart" | "dangerous";
     promptFile?: string;
     transport?: "cli" | "acp";
     sessionId?: string;
@@ -463,7 +499,7 @@ export interface DevinRequestParams {
 export declare function prepareDevinRequest(params: {
     prompt?: string;
     model?: string;
-    permissionMode?: string;
+    permissionMode?: DevinRequestParams["permissionMode"];
     promptFile?: string;
     correlationId?: string;
     optimizePrompt: boolean;