@wardby/cli 0.5.3 → 0.6.0

package/dist/sandbox/user-tool.js

@@ -0,0 +1,32 @@
+/**
+ * One user-tool call, end to end: parse the model's arguments, validate them
+ * against the tool's Zod schema (in its own sandbox), run the body, and format
+ * the result string fed back to the model. Never throws — every failure is a
+ * JSON error result. Worker-safe: it touches no database or provider, only the
+ * privileged host it is given, so the native sandbox worker runs it as is.
+ */
+import { runToolCode } from "./run-in-sandbox.js";
+import { validateParams } from "./zod-params.js";
+export async function runUserToolCall(tool, argsJson, privileged, limits) {
+    let parsedArgs;
+    try {
+        // Some providers stream no JSON delta at all for a zero-parameter
+        // tool call, yielding an empty argsJson rather than "{}".
+        parsedArgs = JSON.parse(argsJson || "{}");
+    }
+    catch (err) {
+        return JSON.stringify({
+            error: "invalid_arguments_json",
+            message: err instanceof Error ? err.message : String(err),
+        });
+    }
+    const validation = await validateParams(tool.paramsZod, parsedArgs);
+    if (!validation.ok) {
+        return JSON.stringify({ error: "validation_failed", message: validation.errorMessage });
+    }
+    const result = await runToolCode({ code: tool.code, params: validation.value, limits, privileged });
+    if (!result.ok) {
+        return JSON.stringify({ error: result.errorKind, message: result.errorMessage });
+    }
+    return JSON.stringify(result.value);
+}

package/dist/serve.js

@@ -19,10 +19,15 @@
 import { loadCodingConcurrencyConfig, loadMcpConfig } from "./config/providers.js";
 import { drainCodingQueue } from "./core/coding-queue.js";
 import { prisma } from "./core/db.js";
+import { logger } from "./core/logger.js";
+import { DISABLED_POWER } from "./core/power.js";
+import { startNotifications } from "./core/notifications.js";
 import { startReconciler } from "./core/reconciler.js";
 import { createRepoAccessGate } from "./core/repo-access.js";
 import { startScheduler } from "./core/scheduler.js";
+import { drainDeferredRuns } from "./providers/executor/deferred.js";
 import { buildMcpProviders, startMcp } from "./mcp/index.js";
+const serveLog = logger.child({ module: "serve" });
 export async function startServe(options = {}) {
     const transport = loadMcpConfig().transport;
     if (transport !== "http") {
@@ -33,6 +38,13 @@ export async function startServe(options = {}) {
     // CODING_QUEUE_TIMEOUT_SEC fails fast instead of after HTTP is listening.
     const concurrency = loadCodingConcurrencyConfig();
     const providers = options.providers ?? buildMcpProviders().providers;
+    const power = providers.power ?? DISABLED_POWER;
+    if (power.enabled) {
+        void power.checkAccess().then((missing) => {
+            if (missing.length)
+                serveLog.warn(`low_power_scale_forbidden: missing ${missing.join(", ")}`);
+        }, (err) => serveLog.warn({ err }, "low_power_access_check_failed"));
+    }
     const mcp = await startMcp({ providers, schedulerAttached: true });
     const reviewHosts = providers.reviewHosts;
     const reconciler = startReconciler({
@@ -51,6 +63,7 @@ export async function startServe(options = {}) {
             }
             : {}),
     });
+    const notifications = await startNotifications({ db: prisma, chat: providers.chat ?? {} });
     const selfDefects = { db: prisma, issueTrackers: providers.issueTrackers };
     const scheduler = startScheduler({
         executor: providers.executor,
@@ -58,6 +71,10 @@ export async function startServe(options = {}) {
         selfDefects,
         onLeaderTick: async () => {
             await drainCodingQueue({ db: prisma, executor: providers.executor, ...concurrency, selfDefects });
+            // Runs the native sandbox gateway dispatched (or asked to stop): it holds no executor of its own.
+            await drainDeferredRuns({ db: prisma, executor: providers.executor });
+            // Scale idle coding-proxy/native-gateway Deployments to zero (no-op unless WARDBY_LOW_POWER=1).
+            await power.sleepIfIdle().catch((err) => serveLog.warn({ err }, "power_sleep_failed"));
         },
     });
     return {
@@ -67,6 +84,7 @@ export async function startServe(options = {}) {
             // stop reaping, then let startMcp close HTTP and drain the executor.
             scheduler.stop();
             reconciler.stop();
+            notifications.stop();
             await mcp.close();
         },
     };

package/dist/viewer/api-schema.d.ts

@@ -192,6 +192,10 @@ export declare const GraphRunSchema: z.ZodObject<{
     model: z.ZodString;
     /** The coding worker (e.g. `codex`, `claude-code`); null for native runs. */
     codingProvider: z.ZodNullable<z.ZodString>;
+    /** Where a native run executed, snapshotted at start; null for coding runs and runs from before the field existed. */
+    nativeExecutionMode: z.ZodNullable<z.ZodEnum<["control-plane", "sandbox"]>>;
+    /** The warm pool worker (container or pod name) a sandbox run claimed, while its record lasts; null otherwise. */
+    warmWorkerName: z.ZodNullable<z.ZodString>;
     status: z.ZodEnum<["pending", "running", "succeeded", "failed", "refused", "lost", "budget_exhausted", "cancelled"]>;
     trigger: z.ZodDiscriminatedUnion<"kind", [z.ZodObject<{
         kind: z.ZodLiteral<"manual">;
@@ -432,10 +436,12 @@ export declare const GraphRunSchema: z.ZodObject<{
     finishedAt: string | null;
     heartbeatAt: string | null;
     turns: number;
+    nativeExecutionMode: "sandbox" | "control-plane" | null;
     parentRunId: string | null;
-    agentKind: "native" | "coding";
     agentName: string;
+    agentKind: "native" | "coding";
     codingProvider: string | null;
+    warmWorkerName: string | null;
     outcomes: ({
         number: number;
         at: string | null;
@@ -513,10 +519,12 @@ export declare const GraphRunSchema: z.ZodObject<{
     finishedAt: string | null;
     heartbeatAt: string | null;
     turns: number;
+    nativeExecutionMode: "sandbox" | "control-plane" | null;
     parentRunId: string | null;
-    agentKind: "native" | "coding";
     agentName: string;
+    agentKind: "native" | "coding";
     codingProvider: string | null;
+    warmWorkerName: string | null;
     outcomes: ({
         number: number;
         at: string | null;
@@ -581,6 +589,10 @@ export declare const GraphSnapshotSchema: z.ZodObject<{
         model: z.ZodString;
         /** The coding worker (e.g. `codex`, `claude-code`); null for native runs. */
         codingProvider: z.ZodNullable<z.ZodString>;
+        /** Where a native run executed, snapshotted at start; null for coding runs and runs from before the field existed. */
+        nativeExecutionMode: z.ZodNullable<z.ZodEnum<["control-plane", "sandbox"]>>;
+        /** The warm pool worker (container or pod name) a sandbox run claimed, while its record lasts; null otherwise. */
+        warmWorkerName: z.ZodNullable<z.ZodString>;
         status: z.ZodEnum<["pending", "running", "succeeded", "failed", "refused", "lost", "budget_exhausted", "cancelled"]>;
         trigger: z.ZodDiscriminatedUnion<"kind", [z.ZodObject<{
             kind: z.ZodLiteral<"manual">;
@@ -821,10 +833,12 @@ export declare const GraphSnapshotSchema: z.ZodObject<{
         finishedAt: string | null;
         heartbeatAt: string | null;
         turns: number;
+        nativeExecutionMode: "sandbox" | "control-plane" | null;
         parentRunId: string | null;
-        agentKind: "native" | "coding";
         agentName: string;
+        agentKind: "native" | "coding";
         codingProvider: string | null;
+        warmWorkerName: string | null;
         outcomes: ({
             number: number;
             at: string | null;
@@ -902,10 +916,12 @@ export declare const GraphSnapshotSchema: z.ZodObject<{
         finishedAt: string | null;
         heartbeatAt: string | null;
         turns: number;
+        nativeExecutionMode: "sandbox" | "control-plane" | null;
         parentRunId: string | null;
-        agentKind: "native" | "coding";
         agentName: string;
+        agentKind: "native" | "coding";
         codingProvider: string | null;
+        warmWorkerName: string | null;
         outcomes: ({
             number: number;
             at: string | null;
@@ -1020,10 +1036,12 @@ export declare const GraphSnapshotSchema: z.ZodObject<{
         finishedAt: string | null;
         heartbeatAt: string | null;
         turns: number;
+        nativeExecutionMode: "sandbox" | "control-plane" | null;
         parentRunId: string | null;
-        agentKind: "native" | "coding";
         agentName: string;
+        agentKind: "native" | "coding";
         codingProvider: string | null;
+        warmWorkerName: string | null;
         outcomes: ({
             number: number;
             at: string | null;
@@ -1116,10 +1134,12 @@ export declare const GraphSnapshotSchema: z.ZodObject<{
         finishedAt: string | null;
         heartbeatAt: string | null;
         turns: number;
+        nativeExecutionMode: "sandbox" | "control-plane" | null;
         parentRunId: string | null;
-        agentKind: "native" | "coding";
         agentName: string;
+        agentKind: "native" | "coding";
         codingProvider: string | null;
+        warmWorkerName: string | null;
         outcomes: ({
             number: number;
             at: string | null;
@@ -1177,6 +1197,10 @@ export declare const RunDetailSchema: z.ZodObject<{
     model: z.ZodString;
     /** The coding worker (e.g. `codex`, `claude-code`); null for native runs. */
     codingProvider: z.ZodNullable<z.ZodString>;
+    /** Where a native run executed, snapshotted at start; null for coding runs and runs from before the field existed. */
+    nativeExecutionMode: z.ZodNullable<z.ZodEnum<["control-plane", "sandbox"]>>;
+    /** The warm pool worker (container or pod name) a sandbox run claimed, while its record lasts; null otherwise. */
+    warmWorkerName: z.ZodNullable<z.ZodString>;
     status: z.ZodEnum<["pending", "running", "succeeded", "failed", "refused", "lost", "budget_exhausted", "cancelled"]>;
     trigger: z.ZodDiscriminatedUnion<"kind", [z.ZodObject<{
         kind: z.ZodLiteral<"manual">;
@@ -1473,6 +1497,7 @@ export declare const RunDetailSchema: z.ZodObject<{
     heartbeatAt: string | null;
     finalText: string | null;
     turns: number;
+    nativeExecutionMode: "sandbox" | "control-plane" | null;
     parentRunId: string | null;
     coding: {
         repository: string;
@@ -1488,9 +1513,10 @@ export declare const RunDetailSchema: z.ZodObject<{
         failureCategory: string | null;
         queuedAt: string | null;
     } | null;
-    agentKind: "native" | "coding";
     agentName: string;
+    agentKind: "native" | "coding";
     codingProvider: string | null;
+    warmWorkerName: string | null;
     outcomes: ({
         number: number;
         at: string | null;
@@ -1571,6 +1597,7 @@ export declare const RunDetailSchema: z.ZodObject<{
     heartbeatAt: string | null;
     finalText: string | null;
     turns: number;
+    nativeExecutionMode: "sandbox" | "control-plane" | null;
     parentRunId: string | null;
     coding: {
         repository: string;
@@ -1586,9 +1613,10 @@ export declare const RunDetailSchema: z.ZodObject<{
         failureCategory: string | null;
         queuedAt: string | null;
     } | null;
-    agentKind: "native" | "coding";
     agentName: string;
+    agentKind: "native" | "coding";
     codingProvider: string | null;
+    warmWorkerName: string | null;
     outcomes: ({
         number: number;
         at: string | null;
@@ -1726,6 +1754,56 @@ export declare const InfraInfoSchema: z.ZodObject<{
         componentLabel: Record<string, string>;
         managedByLabel: Record<string, string>;
     }>>;
+    /** How sandbox-mode native agents run; null when NATIVE_SANDBOX_LAUNCHER is unset. */
+    native: z.ZodNullable<z.ZodObject<{
+        launcher: z.ZodEnum<["docker", "kubernetes"]>;
+        /** Idle workers kept ready for runs to claim (0 = no pool). */
+        warmPoolSize: z.ZodNumber;
+        kubernetes: z.ZodNullable<z.ZodObject<{
+            namespace: z.ZodString;
+            runtimeClass: z.ZodNullable<z.ZodString>;
+            runLabel: z.ZodString;
+            componentLabel: z.ZodRecord<z.ZodString, z.ZodString>;
+            warmPoolLabel: z.ZodRecord<z.ZodString, z.ZodString>;
+            warmWorkerLabel: z.ZodString;
+        }, "strip", z.ZodTypeAny, {
+            namespace: string;
+            runtimeClass: string | null;
+            runLabel: string;
+            componentLabel: Record<string, string>;
+            warmPoolLabel: Record<string, string>;
+            warmWorkerLabel: string;
+        }, {
+            namespace: string;
+            runtimeClass: string | null;
+            runLabel: string;
+            componentLabel: Record<string, string>;
+            warmPoolLabel: Record<string, string>;
+            warmWorkerLabel: string;
+        }>>;
+    }, "strip", z.ZodTypeAny, {
+        kubernetes: {
+            namespace: string;
+            runtimeClass: string | null;
+            runLabel: string;
+            componentLabel: Record<string, string>;
+            warmPoolLabel: Record<string, string>;
+            warmWorkerLabel: string;
+        } | null;
+        launcher: "docker" | "kubernetes";
+        warmPoolSize: number;
+    }, {
+        kubernetes: {
+            namespace: string;
+            runtimeClass: string | null;
+            runLabel: string;
+            componentLabel: Record<string, string>;
+            warmPoolLabel: Record<string, string>;
+            warmWorkerLabel: string;
+        } | null;
+        launcher: "docker" | "kubernetes";
+        warmPoolSize: number;
+    }>>;
 }, "strip", z.ZodTypeAny, {
     kubernetes: {
         namespace: string;
@@ -1737,6 +1815,18 @@ export declare const InfraInfoSchema: z.ZodObject<{
         componentLabel: Record<string, string>;
         managedByLabel: Record<string, string>;
     } | null;
+    native: {
+        kubernetes: {
+            namespace: string;
+            runtimeClass: string | null;
+            runLabel: string;
+            componentLabel: Record<string, string>;
+            warmPoolLabel: Record<string, string>;
+            warmWorkerLabel: string;
+        } | null;
+        launcher: "docker" | "kubernetes";
+        warmPoolSize: number;
+    } | null;
     launcher: "local" | "docker" | "kubernetes";
 }, {
     kubernetes: {
@@ -1749,6 +1839,18 @@ export declare const InfraInfoSchema: z.ZodObject<{
         componentLabel: Record<string, string>;
         managedByLabel: Record<string, string>;
     } | null;
+    native: {
+        kubernetes: {
+            namespace: string;
+            runtimeClass: string | null;
+            runLabel: string;
+            componentLabel: Record<string, string>;
+            warmPoolLabel: Record<string, string>;
+            warmWorkerLabel: string;
+        } | null;
+        launcher: "docker" | "kubernetes";
+        warmPoolSize: number;
+    } | null;
     launcher: "local" | "docker" | "kubernetes";
 }>;
 export type InfraInfo = z.infer<typeof InfraInfoSchema>;
@@ -1768,6 +1870,10 @@ export declare const VIEWER_SCHEMAS: {
             model: z.ZodString;
             /** The coding worker (e.g. `codex`, `claude-code`); null for native runs. */
             codingProvider: z.ZodNullable<z.ZodString>;
+            /** Where a native run executed, snapshotted at start; null for coding runs and runs from before the field existed. */
+            nativeExecutionMode: z.ZodNullable<z.ZodEnum<["control-plane", "sandbox"]>>;
+            /** The warm pool worker (container or pod name) a sandbox run claimed, while its record lasts; null otherwise. */
+            warmWorkerName: z.ZodNullable<z.ZodString>;
             status: z.ZodEnum<["pending", "running", "succeeded", "failed", "refused", "lost", "budget_exhausted", "cancelled"]>;
             trigger: z.ZodDiscriminatedUnion<"kind", [z.ZodObject<{
                 kind: z.ZodLiteral<"manual">;
@@ -2008,10 +2114,12 @@ export declare const VIEWER_SCHEMAS: {
             finishedAt: string | null;
             heartbeatAt: string | null;
             turns: number;
+            nativeExecutionMode: "sandbox" | "control-plane" | null;
             parentRunId: string | null;
-            agentKind: "native" | "coding";
             agentName: string;
+            agentKind: "native" | "coding";
             codingProvider: string | null;
+            warmWorkerName: string | null;
             outcomes: ({
                 number: number;
                 at: string | null;
@@ -2089,10 +2197,12 @@ export declare const VIEWER_SCHEMAS: {
             finishedAt: string | null;
             heartbeatAt: string | null;
             turns: number;
+            nativeExecutionMode: "sandbox" | "control-plane" | null;
             parentRunId: string | null;
-            agentKind: "native" | "coding";
             agentName: string;
+            agentKind: "native" | "coding";
             codingProvider: string | null;
+            warmWorkerName: string | null;
             outcomes: ({
                 number: number;
                 at: string | null;
@@ -2207,10 +2317,12 @@ export declare const VIEWER_SCHEMAS: {
             finishedAt: string | null;
             heartbeatAt: string | null;
             turns: number;
+            nativeExecutionMode: "sandbox" | "control-plane" | null;
             parentRunId: string | null;
-            agentKind: "native" | "coding";
             agentName: string;
+            agentKind: "native" | "coding";
             codingProvider: string | null;
+            warmWorkerName: string | null;
             outcomes: ({
                 number: number;
                 at: string | null;
@@ -2303,10 +2415,12 @@ export declare const VIEWER_SCHEMAS: {
             finishedAt: string | null;
             heartbeatAt: string | null;
             turns: number;
+            nativeExecutionMode: "sandbox" | "control-plane" | null;
             parentRunId: string | null;
-            agentKind: "native" | "coding";
             agentName: string;
+            agentKind: "native" | "coding";
             codingProvider: string | null;
+            warmWorkerName: string | null;
             outcomes: ({
                 number: number;
                 at: string | null;
@@ -2364,6 +2478,10 @@ export declare const VIEWER_SCHEMAS: {
         model: z.ZodString;
         /** The coding worker (e.g. `codex`, `claude-code`); null for native runs. */
         codingProvider: z.ZodNullable<z.ZodString>;
+        /** Where a native run executed, snapshotted at start; null for coding runs and runs from before the field existed. */
+        nativeExecutionMode: z.ZodNullable<z.ZodEnum<["control-plane", "sandbox"]>>;
+        /** The warm pool worker (container or pod name) a sandbox run claimed, while its record lasts; null otherwise. */
+        warmWorkerName: z.ZodNullable<z.ZodString>;
         status: z.ZodEnum<["pending", "running", "succeeded", "failed", "refused", "lost", "budget_exhausted", "cancelled"]>;
         trigger: z.ZodDiscriminatedUnion<"kind", [z.ZodObject<{
             kind: z.ZodLiteral<"manual">;
@@ -2660,6 +2778,7 @@ export declare const VIEWER_SCHEMAS: {
         heartbeatAt: string | null;
         finalText: string | null;
         turns: number;
+        nativeExecutionMode: "sandbox" | "control-plane" | null;
         parentRunId: string | null;
         coding: {
             repository: string;
@@ -2675,9 +2794,10 @@ export declare const VIEWER_SCHEMAS: {
             failureCategory: string | null;
             queuedAt: string | null;
         } | null;
-        agentKind: "native" | "coding";
         agentName: string;
+        agentKind: "native" | "coding";
         codingProvider: string | null;
+        warmWorkerName: string | null;
         outcomes: ({
             number: number;
             at: string | null;
@@ -2758,6 +2878,7 @@ export declare const VIEWER_SCHEMAS: {
         heartbeatAt: string | null;
         finalText: string | null;
         turns: number;
+        nativeExecutionMode: "sandbox" | "control-plane" | null;
         parentRunId: string | null;
         coding: {
             repository: string;
@@ -2773,9 +2894,10 @@ export declare const VIEWER_SCHEMAS: {
             failureCategory: string | null;
             queuedAt: string | null;
         } | null;
-        agentKind: "native" | "coding";
         agentName: string;
+        agentKind: "native" | "coding";
         codingProvider: string | null;
+        warmWorkerName: string | null;
         outcomes: ({
             number: number;
             at: string | null;
@@ -2905,6 +3027,56 @@ export declare const VIEWER_SCHEMAS: {
             componentLabel: Record<string, string>;
             managedByLabel: Record<string, string>;
         }>>;
+        /** How sandbox-mode native agents run; null when NATIVE_SANDBOX_LAUNCHER is unset. */
+        native: z.ZodNullable<z.ZodObject<{
+            launcher: z.ZodEnum<["docker", "kubernetes"]>;
+            /** Idle workers kept ready for runs to claim (0 = no pool). */
+            warmPoolSize: z.ZodNumber;
+            kubernetes: z.ZodNullable<z.ZodObject<{
+                namespace: z.ZodString;
+                runtimeClass: z.ZodNullable<z.ZodString>;
+                runLabel: z.ZodString;
+                componentLabel: z.ZodRecord<z.ZodString, z.ZodString>;
+                warmPoolLabel: z.ZodRecord<z.ZodString, z.ZodString>;
+                warmWorkerLabel: z.ZodString;
+            }, "strip", z.ZodTypeAny, {
+                namespace: string;
+                runtimeClass: string | null;
+                runLabel: string;
+                componentLabel: Record<string, string>;
+                warmPoolLabel: Record<string, string>;
+                warmWorkerLabel: string;
+            }, {
+                namespace: string;
+                runtimeClass: string | null;
+                runLabel: string;
+                componentLabel: Record<string, string>;
+                warmPoolLabel: Record<string, string>;
+                warmWorkerLabel: string;
+            }>>;
+        }, "strip", z.ZodTypeAny, {
+            kubernetes: {
+                namespace: string;
+                runtimeClass: string | null;
+                runLabel: string;
+                componentLabel: Record<string, string>;
+                warmPoolLabel: Record<string, string>;
+                warmWorkerLabel: string;
+            } | null;
+            launcher: "docker" | "kubernetes";
+            warmPoolSize: number;
+        }, {
+            kubernetes: {
+                namespace: string;
+                runtimeClass: string | null;
+                runLabel: string;
+                componentLabel: Record<string, string>;
+                warmPoolLabel: Record<string, string>;
+                warmWorkerLabel: string;
+            } | null;
+            launcher: "docker" | "kubernetes";
+            warmPoolSize: number;
+        }>>;
     }, "strip", z.ZodTypeAny, {
         kubernetes: {
             namespace: string;
@@ -2916,6 +3088,18 @@ export declare const VIEWER_SCHEMAS: {
             componentLabel: Record<string, string>;
             managedByLabel: Record<string, string>;
         } | null;
+        native: {
+            kubernetes: {
+                namespace: string;
+                runtimeClass: string | null;
+                runLabel: string;
+                componentLabel: Record<string, string>;
+                warmPoolLabel: Record<string, string>;
+                warmWorkerLabel: string;
+            } | null;
+            launcher: "docker" | "kubernetes";
+            warmPoolSize: number;
+        } | null;
         launcher: "local" | "docker" | "kubernetes";
     }, {
         kubernetes: {
@@ -2928,6 +3112,18 @@ export declare const VIEWER_SCHEMAS: {
             componentLabel: Record<string, string>;
             managedByLabel: Record<string, string>;
         } | null;
+        native: {
+            kubernetes: {
+                namespace: string;
+                runtimeClass: string | null;
+                runLabel: string;
+                componentLabel: Record<string, string>;
+                warmPoolLabel: Record<string, string>;
+                warmWorkerLabel: string;
+            } | null;
+            launcher: "docker" | "kubernetes";
+            warmPoolSize: number;
+        } | null;
         launcher: "local" | "docker" | "kubernetes";
     }>;
 };

package/dist/viewer/api-schema.js

@@ -88,6 +88,10 @@ export const GraphRunSchema = z.object({
     model: z.string(),
     /** The coding worker (e.g. `codex`, `claude-code`); null for native runs. */
     codingProvider: z.string().nullable(),
+    /** Where a native run executed, snapshotted at start; null for coding runs and runs from before the field existed. */
+    nativeExecutionMode: z.enum(["control-plane", "sandbox"]).nullable(),
+    /** The warm pool worker (container or pod name) a sandbox run claimed, while its record lasts; null otherwise. */
+    warmWorkerName: z.string().nullable(),
     status: RunStatusSchema,
     trigger: RunTriggerSchema,
     turns: z.number().int(),
@@ -181,6 +185,24 @@ export const InfraInfoSchema = z.object({
         managedByLabel: z.record(z.string(), z.string()),
     })
         .nullable(),
+    /** How sandbox-mode native agents run; null when NATIVE_SANDBOX_LAUNCHER is unset. */
+    native: z
+        .object({
+        launcher: z.enum(["docker", "kubernetes"]),
+        /** Idle workers kept ready for runs to claim (0 = no pool). */
+        warmPoolSize: z.number().int(),
+        kubernetes: z
+            .object({
+            namespace: z.string(),
+            runtimeClass: z.string().nullable(),
+            runLabel: z.string(),
+            componentLabel: z.record(z.string(), z.string()),
+            warmPoolLabel: z.record(z.string(), z.string()),
+            warmWorkerLabel: z.string(),
+        })
+            .nullable(),
+    })
+        .nullable(),
 });
 export const VIEWER_SCHEMAS = {
     "graph-snapshot": GraphSnapshotSchema,

package/dist/viewer/build-schemas.js

@@ -1,9 +1,10 @@
 /** Writes src/viewer/schemas/<name>.schema.json from VIEWER_SCHEMAS (npm run build:viewer-schemas). */
-import { mkdir, writeFile } from "node:fs/promises";
+import { mkdir, readFile, writeFile } from "node:fs/promises";
 import { dirname, join } from "node:path";
 import { fileURLToPath } from "node:url";
 import { zodToJsonSchema } from "zod-to-json-schema";
 import { VIEWER_SCHEMAS } from "./api-schema.js";
+import { OPENAPI_PATH, renderViewerOpenApi } from "./openapi.js";
 export const SCHEMA_DIR = join(dirname(fileURLToPath(import.meta.url)), "schemas");
 export function renderSchemas() {
     return Object.fromEntries(Object.entries(VIEWER_SCHEMAS).map(([name, schema]) => [
@@ -15,4 +16,6 @@ if (process.argv[1] === fileURLToPath(import.meta.url)) {
     await mkdir(SCHEMA_DIR, { recursive: true });
     for (const [file, text] of Object.entries(renderSchemas()))
         await writeFile(join(SCHEMA_DIR, file), text);
+    const pkg = JSON.parse(await readFile(new URL("../../package.json", import.meta.url), "utf8"));
+    await writeFile(OPENAPI_PATH, renderViewerOpenApi(pkg.version));
 }

package/dist/viewer/event-bus.d.ts

@@ -14,6 +14,7 @@
  *   layer can tell its clients to refetch whatever they may have missed.
  */
 import pg from "pg";
+import { type DatabaseAuth, type PgAuthOptions } from "../core/pg-auth.js";
 import { type ViewerEvent } from "./api-schema.js";
 export declare const VIEWER_CHANNEL = "wardby_viewer";
 export interface ViewerEventBus {
@@ -29,8 +30,13 @@ export interface ViewerEventBus {
     onState(listener: (live: boolean) => void): () => void;
     close(): Promise<void>;
 }
+/** The LISTEN client's config: auth (pg-auth.ts) plus keepAlive and the pool's connect timeout. */
+export declare function eventBusClientConfig(connectionString: string, auth: DatabaseAuth, pgAuth?: PgAuthOptions): pg.ClientConfig;
 export declare function createViewerEventBus(options: {
     connectionString: string;
+    /** Defaults to DATABASE_AUTH. */
+    auth?: DatabaseAuth;
+    pgAuth?: PgAuthOptions;
     connect?: (connectionString: string) => pg.Client;
     reconnectDelaysMs?: readonly number[];
     healthCheckIntervalMs?: number;

package/dist/viewer/event-bus.js

@@ -16,6 +16,7 @@
 import pg from "pg";
 import { poolSettings } from "../core/db.js";
 import { logger } from "../core/logger.js";
+import { databaseAuthFromEnv, pgConnectionConfig } from "../core/pg-auth.js";
 import { ViewerEventSchema } from "./api-schema.js";
 export const VIEWER_CHANNEL = "wardby_viewer";
 const DEFAULT_RECONNECT_DELAYS_MS = [500, 1000, 2000, 5000, 10_000];
@@ -25,17 +26,25 @@ const DEFAULT_HEALTH_CHECK_INTERVAL_MS = 60_000;
 const DEFAULT_HEALTH_CHECK_TIMEOUT_MS = 10_000;
 /** Upper bound on waiting for a graceful disconnect (a half-open socket never answers). */
 const END_TIMEOUT_MS = 2_000;
+/** The LISTEN client's config: auth (pg-auth.ts) plus keepAlive and the pool's connect timeout. */
+export function eventBusClientConfig(connectionString, auth, pgAuth) {
+    return {
+        ...pgConnectionConfig(connectionString, auth, pgAuth),
+        keepAlive: true,
+        keepAliveInitialDelayMillis: KEEPALIVE_INITIAL_DELAY_MS,
+        connectionTimeoutMillis: poolSettings(connectionString).connectionTimeoutMillis,
+    };
+}
 export function createViewerEventBus(options) {
     const healthIntervalMs = options.healthCheckIntervalMs ?? DEFAULT_HEALTH_CHECK_INTERVAL_MS;
     const healthTimeoutMs = options.healthCheckTimeoutMs ?? DEFAULT_HEALTH_CHECK_TIMEOUT_MS;
     const delays = options.reconnectDelaysMs?.length ? options.reconnectDelaysMs : DEFAULT_RECONNECT_DELAYS_MS;
-    const makeClient = options.connect ??
-        ((connectionString) => new pg.Client({
-            connectionString,
-            keepAlive: true,
-            keepAliveInitialDelayMillis: KEEPALIVE_INITIAL_DELAY_MS,
-            connectionTimeoutMillis: poolSettings(connectionString).connectionTimeoutMillis,
-        }));
+    // Built once, up front: a misconfigured DATABASE_AUTH fails here rather than on every reconnect,
+    // and under rds-iam every reconnect shares one token cache.
+    const clientConfig = options.connect
+        ? undefined
+        : eventBusClientConfig(options.connectionString, options.auth ?? databaseAuthFromEnv(), options.pgAuth);
+    const makeClient = options.connect ?? (() => new pg.Client({ ...clientConfig }));
     const listeners = new Set();
     const stateListeners = new Set();
     let client = null; // the current attempt/connection