llm-cli-gateway 2.16.0 → 2.17.0

package/dist/config.d.ts

@@ -1,6 +1,7 @@
 import type { Logger } from "./logger.js";
 import type { RemoteOAuthConfig } from "./auth.js";
 import type { ApiProviderKind } from "./api-provider.js";
+import type { QualityTier } from "./least-cost-types.js";
 export interface DatabaseConfig {
     connectionString: string;
     pool: {
@@ -119,6 +120,36 @@ export interface CompressionConfig {
 }
 export declare const DEFAULT_COMPRESSION_CONFIG: CompressionConfig;
 export declare function loadCompressionConfig(logger?: Logger): CompressionConfig;
+export declare const LEAST_COST_PRIORS_SCOPES: readonly ["global", "principal", "off"];
+export type LeastCostPriorsScope = (typeof LEAST_COST_PRIORS_SCOPES)[number];
+export declare const DEFAULT_LEAST_COST_MAX_COST_USD = 0.5;
+export declare const DEFAULT_LEAST_COST_EXPECTED_OUTPUT_TOKENS = 800;
+export declare const DEFAULT_LEAST_COST_BUDGET_SAFETY_FACTOR = 1.5;
+export declare const DEFAULT_LEAST_COST_MAX_REROUTES = 2;
+export declare const DEFAULT_LEAST_COST_TIERS: Readonly<Record<string, QualityTier>>;
+export interface LeastCostConfig {
+    enabled: boolean;
+    minTier: QualityTier;
+    maxCostUsd: number;
+    defaultExpectedOutputTokens: number;
+    budgetOutputSafetyFactor: number;
+    priorsScope: LeastCostPriorsScope;
+    allowUnpriced: boolean;
+    maxReroutes: number;
+    preferCatalogPrice: boolean;
+    preferenceOrder: string[];
+    tiers: Record<string, QualityTier>;
+    candidates: {
+        allow: string[];
+        deny: string[];
+    };
+    sources: {
+        configFile: string | null;
+        envOverrides: string[];
+    };
+}
+export declare function defaultLeastCostConfig(sourcePath?: string | null): LeastCostConfig;
+export declare function loadLeastCostConfig(logger?: Logger): LeastCostConfig;
 export declare function minStableTokensForModel(config: CacheAwarenessConfig, modelName: string): number;
 export declare const DEFAULT_XAI_API_KEY_ENV = "XAI_API_KEY";
 export declare const DEFAULT_XAI_BASE_URL = "https://api.x.ai/v1";

package/dist/config.js

@@ -33,7 +33,7 @@ export function loadConfig() {
         DatabaseUrlSchema.parse(databaseUrl);
     }
     catch (error) {
-        throw new Error(`Invalid database URL: ${error instanceof Error ? error.message : String(error)}`);
+        throw new Error(`Invalid database URL: ${error instanceof Error ? error.message : String(error)}`, { cause: error });
     }
     return {
         database: {
@@ -87,6 +87,13 @@ const PersistenceSchema = z
             message: `httpJobGraceMs (${cfg.httpJobGraceMs}) must be >= instanceLeaseTtlMs (${cfg.instanceLeaseTtlMs})`,
         });
     }
+    if (cfg.instanceGcMs < cfg.instanceLeaseTtlMs) {
+        ctx.addIssue({
+            code: z.ZodIssueCode.custom,
+            path: ["instanceGcMs"],
+            message: `instanceGcMs (${cfg.instanceGcMs}) must be >= instanceLeaseTtlMs (${cfg.instanceLeaseTtlMs})`,
+        });
+    }
 });
 const DEFAULT_SQLITE_PATH = path.join(os.homedir(), ".llm-cli-gateway", "logs.db");
 function defaultPersistenceConfigPath() {
@@ -108,7 +115,9 @@ export function loadSkillsConfig(logger = noopLogger) {
         skillsParsed = SkillsSchema.parse(rawSkills);
     }
     catch (err) {
-        throw new Error(`Invalid [skills] config: ${err instanceof Error ? err.message : String(err)}`);
+        throw new Error(`Invalid [skills] config: ${err instanceof Error ? err.message : String(err)}`, {
+            cause: err,
+        });
     }
     const paths = [...skillsParsed.paths];
     const envPaths = process.env.LLM_GATEWAY_SKILLS_PATH;
@@ -220,7 +229,7 @@ export function loadPersistenceConfig(logger = noopLogger) {
         parsed = PersistenceSchema.parse(merged);
     }
     catch (err) {
-        throw new Error(`Invalid [persistence] config: ${err instanceof Error ? err.message : String(err)}`);
+        throw new Error(`Invalid [persistence] config: ${err instanceof Error ? err.message : String(err)}`, { cause: err });
     }
     const backend = parsed.backend;
     const resolvedPath = backend === "sqlite" ? expandHome(parsed.path ?? DEFAULT_SQLITE_PATH) : null;
@@ -315,14 +324,16 @@ export function loadLimitsConfig(logger = noopLogger) {
         httpParsed = HttpSessionLimitsSchema.parse(rawHttp);
     }
     catch (err) {
-        throw new Error(`Invalid [http] session-limit config: ${err instanceof Error ? err.message : String(err)}`);
+        throw new Error(`Invalid [http] session-limit config: ${err instanceof Error ? err.message : String(err)}`, { cause: err });
     }
     let limitsParsed;
     try {
         limitsParsed = JobLimitsSchema.parse(rawLimits);
     }
     catch (err) {
-        throw new Error(`Invalid [limits] config: ${err instanceof Error ? err.message : String(err)}`);
+        throw new Error(`Invalid [limits] config: ${err instanceof Error ? err.message : String(err)}`, {
+            cause: err,
+        });
     }
     return {
         http: {
@@ -398,7 +409,7 @@ export function loadCacheAwarenessConfig(logger = noopLogger) {
         parsed = CacheAwarenessSchema.parse(raw ?? {});
     }
     catch (err) {
-        throw new Error(`Invalid [cache_awareness] config: ${err instanceof Error ? err.message : String(err)}`);
+        throw new Error(`Invalid [cache_awareness] config: ${err instanceof Error ? err.message : String(err)}`, { cause: err });
     }
     return {
         emitAnthropicCacheControl: parsed.emit_anthropic_cache_control,
@@ -431,13 +442,137 @@ export function loadCompressionConfig(logger = noopLogger) {
         parsedCompression = CompressionSchema.parse(raw);
     }
     catch (err) {
-        throw new Error(`Invalid [compression] config: ${err instanceof Error ? err.message : String(err)}`);
+        throw new Error(`Invalid [compression] config: ${err instanceof Error ? err.message : String(err)}`, { cause: err });
     }
     return {
         enabled: parsedCompression.enabled,
         sources: { configFile: sourcePath },
     };
 }
+export const LEAST_COST_PRIORS_SCOPES = ["global", "principal", "off"];
+export const DEFAULT_LEAST_COST_MAX_COST_USD = 0.5;
+export const DEFAULT_LEAST_COST_EXPECTED_OUTPUT_TOKENS = 800;
+export const DEFAULT_LEAST_COST_BUDGET_SAFETY_FACTOR = 1.5;
+export const DEFAULT_LEAST_COST_MAX_REROUTES = 2;
+export const DEFAULT_LEAST_COST_TIERS = {
+    "claude:claude-haiku": "economy",
+    "claude:claude-sonnet": "standard",
+    "claude:claude-opus": "frontier",
+    "codex:openai-gpt5": "standard",
+    "gemini:gemini-2.5-flash": "economy",
+    "gemini:gemini-2.5-pro": "standard",
+    "gemini:gemini-3-pro": "frontier",
+    "grok:grok-build": "economy",
+    "grok:grok-4": "standard",
+    "mistral:mistral-devstral": "economy",
+    "mistral:mistral-medium": "standard",
+};
+const QualityTierSchema = z.enum(["economy", "standard", "frontier"]);
+const LeastCostCandidatesSchema = z
+    .object({
+    allow: z.array(z.string().min(1)).default([]),
+    deny: z.array(z.string().min(1)).default([]),
+})
+    .strict()
+    .default({ allow: [], deny: [] });
+const LeastCostSchema = z
+    .object({
+    enabled: z.boolean().default(false),
+    min_tier: QualityTierSchema.default("standard"),
+    max_cost_usd: z.number().positive().default(DEFAULT_LEAST_COST_MAX_COST_USD),
+    default_expected_output_tokens: z
+        .number()
+        .int()
+        .positive()
+        .default(DEFAULT_LEAST_COST_EXPECTED_OUTPUT_TOKENS),
+    budget_output_safety_factor: z
+        .number()
+        .positive()
+        .default(DEFAULT_LEAST_COST_BUDGET_SAFETY_FACTOR),
+    priors_scope: z.enum(LEAST_COST_PRIORS_SCOPES).default("global"),
+    allow_unpriced: z.boolean().default(false),
+    max_reroutes: z.number().int().nonnegative().default(DEFAULT_LEAST_COST_MAX_REROUTES),
+    prefer_catalog_price: z.boolean().default(true),
+    preference_order: z.array(z.string().min(1)).default([]),
+    tiers: z.record(z.string(), QualityTierSchema).default({}),
+    candidates: LeastCostCandidatesSchema,
+})
+    .strict();
+export function defaultLeastCostConfig(sourcePath = null) {
+    return {
+        enabled: false,
+        minTier: "standard",
+        maxCostUsd: DEFAULT_LEAST_COST_MAX_COST_USD,
+        defaultExpectedOutputTokens: DEFAULT_LEAST_COST_EXPECTED_OUTPUT_TOKENS,
+        budgetOutputSafetyFactor: DEFAULT_LEAST_COST_BUDGET_SAFETY_FACTOR,
+        priorsScope: "global",
+        allowUnpriced: false,
+        maxReroutes: DEFAULT_LEAST_COST_MAX_REROUTES,
+        preferCatalogPrice: true,
+        preferenceOrder: [],
+        tiers: { ...DEFAULT_LEAST_COST_TIERS },
+        candidates: { allow: [], deny: [] },
+        sources: { configFile: sourcePath, envOverrides: [] },
+    };
+}
+function readLeastCostFile(configPath, logger) {
+    if (!existsSync(configPath)) {
+        return { raw: undefined, sourcePath: null };
+    }
+    try {
+        const require = createRequire(import.meta.url);
+        const TOML = require("smol-toml");
+        const text = readFileSync(configPath, "utf-8");
+        const parsed = TOML.parse(text);
+        return { raw: parsed?.least_cost, sourcePath: configPath };
+    }
+    catch (err) {
+        logger.error(`Failed to parse gateway config at ${configPath}; using least_cost defaults`, err);
+        return { raw: undefined, sourcePath: null };
+    }
+}
+export function loadLeastCostConfig(logger = noopLogger) {
+    const configPath = defaultGatewayConfigPath();
+    const { raw, sourcePath } = readLeastCostFile(configPath, logger);
+    const envOverrides = [];
+    const envSentinel = process.env.LLM_GATEWAY_LEAST_COST;
+    if (envSentinel !== undefined && envSentinel.length > 0) {
+        envOverrides.push("LLM_GATEWAY_LEAST_COST");
+        logWarn(logger, "LLM_GATEWAY_LEAST_COST is not supported and is ignored; configure [least_cost] in " +
+            "~/.llm-cli-gateway/config.toml (routing cannot be enabled via an environment variable).");
+    }
+    if (raw === undefined) {
+        const cfg = defaultLeastCostConfig(sourcePath);
+        cfg.sources.envOverrides = envOverrides;
+        return cfg;
+    }
+    let parsed;
+    try {
+        parsed = LeastCostSchema.parse(raw);
+    }
+    catch (err) {
+        throw new Error(`Invalid [least_cost] config: ${err instanceof Error ? err.message : String(err)}`, { cause: err });
+    }
+    const tiers = { ...DEFAULT_LEAST_COST_TIERS, ...parsed.tiers };
+    return {
+        enabled: parsed.enabled,
+        minTier: parsed.min_tier,
+        maxCostUsd: parsed.max_cost_usd,
+        defaultExpectedOutputTokens: parsed.default_expected_output_tokens,
+        budgetOutputSafetyFactor: parsed.budget_output_safety_factor,
+        priorsScope: parsed.priors_scope,
+        allowUnpriced: parsed.allow_unpriced,
+        maxReroutes: parsed.max_reroutes,
+        preferCatalogPrice: parsed.prefer_catalog_price,
+        preferenceOrder: [...parsed.preference_order],
+        tiers,
+        candidates: {
+            allow: [...parsed.candidates.allow],
+            deny: [...parsed.candidates.deny],
+        },
+        sources: { configFile: sourcePath, envOverrides },
+    };
+}
 export function minStableTokensForModel(config, modelName) {
     const lower = modelName.toLowerCase();
     const table = config.minStableTokensForCacheControl;

package/dist/db.js

@@ -28,7 +28,7 @@ export class DatabaseConnection {
         }
         catch (error) {
             this.logger.error("Failed to connect to PostgreSQL", { error });
-            throw new Error(`Failed to connect to PostgreSQL: ${error instanceof Error ? error.message : String(error)}`);
+            throw new Error(`Failed to connect to PostgreSQL: ${error instanceof Error ? error.message : String(error)}`, { cause: error });
         }
     }
     async disconnect() {
@@ -40,7 +40,7 @@ export class DatabaseConnection {
                 this.pool = null;
             }
             catch (error) {
-                errors.push(new Error(`PostgreSQL disconnect error: ${error instanceof Error ? error.message : String(error)}`));
+                errors.push(new Error(`PostgreSQL disconnect error: ${error instanceof Error ? error.message : String(error)}`, { cause: error }));
             }
         }
         if (errors.length > 0) {
@@ -60,7 +60,7 @@ export class DatabaseConnection {
                 result.postgres.connected = true;
                 result.postgres.latency = Date.now() - pgStart;
             }
-            catch (error) {
+            catch {
                 result.postgres.connected = false;
             }
             finally {
@@ -87,7 +87,7 @@ async function importOptionalPg() {
     }
     catch (error) {
         if (error?.code === "ERR_MODULE_NOT_FOUND" || error?.code === "MODULE_NOT_FOUND") {
-            throw new Error("PostgreSQL sessions require optional peer dependency 'pg'. Install it alongside llm-cli-gateway to use DATABASE_URL-backed sessions.");
+            throw new Error("PostgreSQL sessions require optional peer dependency 'pg'. Install it alongside llm-cli-gateway to use DATABASE_URL-backed sessions.", { cause: error });
         }
         throw error;
     }

package/dist/doctor.d.ts

@@ -2,7 +2,9 @@ import { type EndpointExposureReport } from "./endpoint-exposure.js";
 import { type ProviderLoginStatus } from "./provider-status.js";
 import { type ApiProviderLoginGuidance } from "./provider-login-guidance.js";
 import type { FlightRecorderQuery } from "./flight-recorder.js";
-import { type CacheAwarenessConfig, type ProvidersConfig, type RemoteOAuthConfigDiagnostics } from "./config.js";
+import { type CacheAwarenessConfig, type LeastCostConfig, type ProvidersConfig, type RemoteOAuthConfigDiagnostics } from "./config.js";
+import { type TelemetryTier } from "./lcr-telemetry.js";
+import type { Confidence, PriceSource, QualityTier } from "./least-cost-types.js";
 import type { AuthConfig } from "./auth.js";
 import { type RemoteSafeWorkspaceSummary } from "./workspace-registry.js";
 import { type ProviderCapabilityId, type ProviderKind } from "./provider-tool-capabilities.js";
@@ -79,6 +81,35 @@ export interface GeminiConfigStatus {
 }
 export declare function checkVibeSessionLogging(home?: string): VibeSessionLoggingStatus;
 export declare function checkGeminiConfig(cwd?: string, home?: string, whitelist?: readonly string[]): GeminiConfigStatus;
+export interface LeastCostCandidateReport {
+    model: string;
+    family: string;
+    priced: boolean;
+    priceSource: PriceSource;
+    tier: QualityTier | null;
+}
+export interface LeastCostProviderReport {
+    provider: CliType;
+    telemetryTier: TelemetryTier;
+    candidates: LeastCostCandidateReport[];
+}
+export interface LeastCostCalibrationBucketReport {
+    bucket: string;
+    k: number;
+    samples: number;
+    confidence: Confidence;
+}
+export interface LeastCostReport {
+    enabled: boolean;
+    pricing: {
+        tableAsOf: string;
+        apiCatalogAsOf: string;
+        staleDays: number;
+    };
+    providers: LeastCostProviderReport[];
+    untieredModels: string[];
+    calibrationQuality: LeastCostCalibrationBucketReport[];
+}
 export interface DoctorReport {
     schema_version: "1.0";
     ok: boolean;
@@ -158,6 +189,7 @@ export interface DoctorReport {
     };
     cache_awareness: CacheAwarenessReport;
     provider_capabilities: ProviderCapabilitySummaryReport;
+    least_cost: LeastCostReport;
     api_providers?: ApiProviderHealthReport;
     upstream: {
         note: string;
@@ -215,6 +247,7 @@ export interface CreateDoctorReportOptions {
         reachable: boolean;
         error?: string;
     }>;
+    leastCost?: LeastCostConfig;
 }
 export declare function createDoctorReport(envOrOptions?: NodeJS.ProcessEnv | CreateDoctorReportOptions): DoctorReport;
 export declare function printDoctorJson(opts?: {

package/dist/doctor.js

@@ -7,7 +7,11 @@ import { createEndpointExposureReport, redactDiagnosticUrl, } from "./endpoint-e
 import { getApiProviderStatus, listProviderRuntimeStatuses, } from "./provider-status.js";
 import { getApiProviderLoginGuidance, } from "./provider-login-guidance.js";
 import { CLAUDE_MCP_SERVER_NAMES } from "./claude-mcp-config.js";
-import { diagnoseRemoteOAuthConfig, enabledApiProviders, loadCacheAwarenessConfig, loadProvidersConfig, } from "./config.js";
+import { defaultLeastCostConfig, diagnoseRemoteOAuthConfig, enabledApiProviders, loadCacheAwarenessConfig, loadLeastCostConfig, loadProvidersConfig, } from "./config.js";
+import { telemetryTierFor } from "./lcr-telemetry.js";
+import { API_CATALOG_AS_OF, PRICING_AS_OF, getModelCost, modelIdToFamily } from "./pricing.js";
+import { computeLcrPriorsFromDb } from "./lcr-priors.js";
+import { getCliInfo } from "./model-registry.js";
 import { loadWorkspaceRegistry, remoteSafeWorkspaceSummary, } from "./workspace-registry.js";
 import { buildRemoteConnectorUrls, resolveConfiguredRemoteOrigin } from "./remote-url.js";
 import { computeGlobalCacheStats } from "./cache-stats.js";
@@ -58,7 +62,7 @@ export function checkVibeSessionLogging(home = homedir()) {
 function parseVibeSessionLoggingEnabled(text) {
     const lines = text.split(/\r?\n/);
     let inSection = false;
-    for (let raw of lines) {
+    for (const raw of lines) {
         const line = raw.replace(/#.*$/, "").trim();
         if (!line)
             continue;
@@ -426,6 +430,70 @@ function buildCacheAwarenessReport(opts) {
         per_cli: perCli,
     };
 }
+const MS_PER_DAY = 86_400_000;
+function computePricingStaleDays(runIso) {
+    const run = Date.parse(runIso);
+    const table = Date.parse(PRICING_AS_OF);
+    const catalog = Date.parse(API_CATALOG_AS_OF);
+    const newer = Math.max(table, catalog);
+    if (!Number.isFinite(run) || !Number.isFinite(newer))
+        return 0;
+    const days = Math.floor((run - newer) / MS_PER_DAY);
+    return days > 0 ? days : 0;
+}
+function buildLeastCostReport(cfg, generatedAt, flightRecorder) {
+    const pricing = {
+        tableAsOf: PRICING_AS_OF,
+        apiCatalogAsOf: API_CATALOG_AS_OF,
+        staleDays: computePricingStaleDays(generatedAt),
+    };
+    if (!cfg.enabled) {
+        return { enabled: false, pricing, providers: [], untieredModels: [], calibrationQuality: [] };
+    }
+    const tierMap = new Map(Object.entries(cfg.tiers));
+    const cliInfo = getCliInfo();
+    const providers = [];
+    const untieredModels = [];
+    for (const provider of CLI_TYPES) {
+        const candidates = Object.keys(cliInfo[provider].models).map(model => {
+            const family = modelIdToFamily(model);
+            const cost = getModelCost(provider, model, { preferCatalog: cfg.preferCatalogPrice });
+            const tier = tierMap.get(`${provider}:${family}`) ?? null;
+            if (tier === null) {
+                untieredModels.push(`${provider}:${model}`);
+            }
+            return {
+                model,
+                family,
+                priced: cost.source !== "unknown",
+                priceSource: cost.source,
+                tier,
+            };
+        });
+        providers.push({
+            provider,
+            telemetryTier: telemetryTierFor(provider),
+            candidates,
+        });
+    }
+    const calibrationQuality = [];
+    if (flightRecorder && cfg.priorsScope !== "off") {
+        try {
+            const priors = computeLcrPriorsFromDb(flightRecorder, { priorsScope: cfg.priorsScope });
+            for (const [bucket, entry] of priors.calibration) {
+                calibrationQuality.push({
+                    bucket,
+                    k: entry.k,
+                    samples: entry.samples,
+                    confidence: entry.confidence,
+                });
+            }
+        }
+        catch {
+        }
+    }
+    return { enabled: true, pricing, providers, untieredModels, calibrationQuality };
+}
 function buildProviderCapabilitySummary(providerStatuses) {
     const capabilities = getProviderToolCapabilities({
         includeSkills: true,
@@ -502,10 +570,12 @@ export function createDoctorReport(envOrOptions = process.env) {
     if (probeReport) {
         upstream.probe_report = probeReport;
     }
+    const generatedAt = new Date().toISOString();
+    const leastCostConfig = opts.leastCost ?? defaultLeastCostConfig();
     const report = {
         schema_version: "1.0",
         ok: true,
-        generated_at: new Date().toISOString(),
+        generated_at: generatedAt,
         system: {
             os: platform(),
             arch: arch(),
@@ -564,6 +634,7 @@ export function createDoctorReport(envOrOptions = process.env) {
         client_config: clientConfigStatus(),
         cache_awareness: buildCacheAwarenessReport(opts),
         provider_capabilities: buildProviderCapabilitySummary(providerStatuses),
+        least_cost: buildLeastCostReport(leastCostConfig, generatedAt, opts.flightRecorder),
         upstream,
         next_actions: [],
     };
@@ -613,11 +684,17 @@ export async function printDoctorJson(opts = {}) {
     let cacheAwareness;
     let flightRecorder;
     let providersConfig;
+    let leastCost;
     try {
         cacheAwareness = loadCacheAwarenessConfig();
     }
     catch {
     }
+    try {
+        leastCost = loadLeastCostConfig();
+    }
+    catch {
+    }
     try {
         const dbPath = resolveFlightRecorderDbPath();
         if (dbPath)
@@ -644,6 +721,7 @@ export async function printDoctorJson(opts = {}) {
         probeUpstream: opts.probeUpstream,
         providersConfig,
         apiReachability,
+        leastCost,
     });
     process.stdout.write(`${JSON.stringify(report, null, 2)}\n`);
     if (flightRecorder) {
@@ -685,6 +763,10 @@ function isCreateDoctorReportOptions(value) {
         const candidate = value.flightRecorder;
         return candidate === undefined || typeof candidate === "object";
     }
+    if (Object.prototype.hasOwnProperty.call(value, "leastCost")) {
+        const candidate = value.leastCost;
+        return candidate === undefined || typeof candidate === "object";
+    }
     if (Object.prototype.hasOwnProperty.call(value, "env")) {
         const candidate = value.env;
         return candidate === undefined || typeof candidate === "object";

package/dist/executor.d.ts

@@ -1,5 +1,7 @@
 import { ChildProcess, type SpawnOptions } from "child_process";
+import { CircuitBreakerState } from "./retry.js";
 import type { Logger } from "./logger.js";
+import type { CliType } from "./provider-types.js";
 export interface ExecuteOptions {
     timeout?: number;
     idleTimeout?: number;
@@ -14,6 +16,9 @@ export interface ExecuteResult {
     code: number;
 }
 export declare function providerCommandName(command: string): string;
+export declare function cliBreakerState(cli: CliType): CircuitBreakerState;
+export declare function resetCliBreakersForTest(): void;
+export declare function primeCliBreakerStateForTest(command: string, state: CircuitBreakerState): void;
 export declare function buildExtendedPath(env?: NodeJS.ProcessEnv, home?: string, nodePath?: string, platform?: NodeJS.Platform): string;
 export declare function getExtendedPath(): string;
 export declare function envWithExtendedPath(baseEnv?: NodeJS.ProcessEnv, extendedPath?: string, platform?: NodeJS.Platform): NodeJS.ProcessEnv;

package/dist/executor.js

@@ -2,7 +2,7 @@ import { spawn, spawnSync } from "child_process";
 import { homedir } from "os";
 import { delimiter, join, dirname, extname, win32 } from "path";
 import { readdirSync, existsSync } from "fs";
-import { createCircuitBreaker, withRetry } from "./retry.js";
+import { createCircuitBreaker, withRetry, CircuitBreakerState } from "./retry.js";
 import { applySpawnEnvIsolation } from "./spawn-env-isolation.js";
 export function providerCommandName(command) {
     if (command === "gemini")
@@ -25,6 +25,16 @@ function getCircuitBreaker(command) {
     circuitBreakers.set(command, circuitBreaker);
     return circuitBreaker;
 }
+export function cliBreakerState(cli) {
+    const command = providerCommandName(cli);
+    return circuitBreakers.get(command)?.state ?? CircuitBreakerState.CLOSED;
+}
+export function resetCliBreakersForTest() {
+    circuitBreakers.clear();
+}
+export function primeCliBreakerStateForTest(command, state) {
+    getCircuitBreaker(command).state = state;
+}
 function getNvmPath() {
     if (cachedNvmPath !== undefined) {
         return cachedNvmPath;

package/dist/flight-recorder.d.ts

@@ -23,6 +23,7 @@ export interface FlightLogResult {
     retryCount: number;
     circuitBreakerState: string;
     costUsd?: number;
+    costBasis?: string;
     approvalDecision?: string;
     optimizationApplied: boolean;
     thinkingBlocks?: string[];
@@ -44,6 +45,13 @@ export interface CompressionTelemetry {
     compressedChars: number;
     estimatedTokensSaved: number;
 }
+export interface RoutingRecord {
+    estCostUsd?: number | null;
+    estConfidence?: string | null;
+    reason?: string | null;
+    considered?: number | null;
+    reroutes?: number | null;
+}
 export declare function resolveFlightRecorderDbPath(): string | null;
 export declare class FlightRecorder {
     private db;
@@ -59,6 +67,7 @@ export declare class FlightRecorder {
     logStart(entry: FlightLogStart): void;
     logComplete(correlationId: string, result: FlightLogResult): void;
     recordCompressionTelemetry(correlationId: string, telemetry: CompressionTelemetry): void;
+    recordRouting(correlationId: string, routing: RoutingRecord): void;
     private redactStart;
     private redactResult;
     queryRequests<T = Record<string, unknown>>(sql: string, ...params: unknown[]): T[];
@@ -69,6 +78,7 @@ export declare class NoopFlightRecorder {
     logStart(_entry: FlightLogStart): void;
     logComplete(_correlationId: string, _result: FlightLogResult): void;
     recordCompressionTelemetry(_correlationId: string, _telemetry: CompressionTelemetry): void;
+    recordRouting(_correlationId: string, _routing: RoutingRecord): void;
     queryRequests<T = Record<string, unknown>>(_sql: string, ..._params: unknown[]): T[];
     flush(): void;
     close(): void;