@deployfoundation/foundation-deploy 0.2.39 → 0.2.41

package/dist/bin/app.js

@@ -8,13 +8,13 @@ import {
   FoundationPages,
   FoundationPipeline,
   NewsletterStack
-} from "../chunk-6v0ewr5b.js";
+} from "../chunk-sgh3agpa.js";
 import {
   adminsFor,
   instanceNames,
   loadInstanceFile,
   provisionsIntegration
-} from "../chunk-7rd8c1th.js";
+} from "../chunk-g56mqyfx.js";
 import"../chunk-dz91avrb.js";
 
 // bin/app.ts

package/dist/bin/foundation-deploy.js

@@ -22,7 +22,7 @@ import {
   slackCommandPrefix,
   verifyManifest,
   versionedConfigKey
-} from "../chunk-7rd8c1th.js";
+} from "../chunk-g56mqyfx.js";
 import {
   DEFAULT_RELEASE_BUCKET,
   expectedKeyArn
@@ -1043,6 +1043,21 @@ var SMOKE_EXPECT = ['"authenticated":true', '"config_source":"s3"'];
 function smokeSessionId(prefix) {
   return `${prefix}-${Math.floor(Date.now() / 1000)}`.padEnd(40, "0");
 }
+async function stopSmokeSession(runner, opts, log) {
+  try {
+    await runner.capture([
+      "bedrock-agentcore",
+      "stop-runtime-session",
+      "--agent-runtime-arn",
+      opts.arn,
+      ...opts.qualifier === undefined ? [] : ["--qualifier", opts.qualifier],
+      "--runtime-session-id",
+      opts.sessionId
+    ]);
+  } catch {
+    log("  (could not stop the smoke session; it ends at the idle timeout)");
+  }
+}
 async function smokeInvoke(runner, opts) {
   const log = opts.log ?? ((line) => console.log(line));
   const payload = opts.payload ?? '{"_smoke_test":true}';
@@ -1070,6 +1085,7 @@ async function smokeInvoke(runner, opts) {
   } finally {
     await rm(dir, { recursive: true, force: true });
   }
+  await stopSmokeSession(runner, opts, log);
   log(`  ${out.replaceAll(`
 `, `
   `)}`);

package/dist/{chunk-7rd8c1th.js → chunk-g56mqyfx.js}

@@ -1261,31 +1261,45 @@ var contextStorage = new AsyncLocalStorage;
 // ../core/src/config.ts
 import { createHash as createHash2 } from "node:crypto";
 import { parse as parseYaml2 } from "yaml";
-import { z as z4 } from "zod";
+import { z as z5 } from "zod";
 
 // ../core/src/jobs.ts
 import { createHash, createHmac, timingSafeEqual } from "node:crypto";
+import { z as z4 } from "zod";
+
+// ../core/src/thinking-level.ts
 import { z as z3 } from "zod";
+var ThinkingLevel = z3.enum(["none", "low", "medium", "high", "xhigh", "max"]);
+
+// ../core/src/jobs.ts
 var JOB_ID_PATTERN = /^[a-z0-9][a-z0-9-]{0,40}$/;
-var RawScheduleTrigger = z3.object({
-  schedule: z3.string().min(1)
+var RawScheduleTrigger = z4.object({
+  schedule: z4.string().min(1)
 }).strict();
-var RawWebhookTrigger = z3.object({
-  webhook: z3.object({}).strict()
+var RawWebhookTrigger = z4.object({
+  webhook: z4.object({}).strict()
 }).strict();
-var RawJob = z3.object({
-  id: z3.string().regex(JOB_ID_PATTERN),
-  name: z3.string().min(1).max(80),
-  channel: z3.string().regex(/^[CG][A-Z0-9]{8,}$/, "a job runs in a Slack channel id"),
-  on: z3.union([RawScheduleTrigger, RawWebhookTrigger]),
-  prompt: z3.string().min(1).max(4000)
-}).strict();
-var RawJobs = z3.array(RawJob).default([]).superRefine((jobs, ctx) => {
+var RawJob = z4.object({
+  id: z4.string().regex(JOB_ID_PATTERN),
+  name: z4.string().min(1).max(80),
+  channel: z4.string().regex(/^[CG][A-Z0-9]{8,}$/, "a job runs in a Slack channel id"),
+  on: z4.union([RawScheduleTrigger, RawWebhookTrigger]),
+  prompt: z4.string().min(1).max(4000),
+  thinkingLevel: ThinkingLevel.optional()
+}).strict().superRefine((job, ctx) => {
+  if (job.thinkingLevel !== undefined && !("schedule" in job.on))
+    ctx.addIssue({
+      code: z4.ZodIssueCode.custom,
+      path: ["thinkingLevel"],
+      message: "thinkingLevel applies to scheduled jobs only"
+    });
+});
+var RawJobs = z4.array(RawJob).default([]).superRefine((jobs, ctx) => {
   const seen = new Set;
   for (const [index, job] of jobs.entries()) {
     if (seen.has(job.id)) {
       ctx.addIssue({
-        code: z3.ZodIssueCode.custom,
+        code: z4.ZodIssueCode.custom,
         path: [index, "id"],
         message: `duplicate job id: ${job.id}`
       });
@@ -1306,9 +1320,13 @@ function scheduledJobFingerprint(input) {
   return createHash("sha256").update(JSON.stringify([input.routineId, input.name, input.channel, input.prompt])).digest("hex");
 }
 function scheduledJobBindings(jobs) {
-  return jobs.filter(isScheduledJob).map(({ id, name, channel, prompt }) => {
+  return jobs.filter(isScheduledJob).map(({ id, name, channel, prompt, thinkingLevel }) => {
     const input = { routineId: `job-${id}`, name, channel, prompt };
-    return { routineId: input.routineId, fingerprint: scheduledJobFingerprint(input) };
+    return {
+      routineId: input.routineId,
+      fingerprint: scheduledJobFingerprint(input),
+      ...thinkingLevel === undefined ? {} : { thinkingLevel }
+    };
   });
 }
 function jobWebhookKey(instanceSecret, jobId) {
@@ -1374,11 +1392,10 @@ function capabilityScheduledJobs(config) {
 
 // ../core/src/config.ts
 var MODEL_RE = /^([a-z0-9-]+)\/(.+)$/;
-var ThinkingLevel = z4.enum(["none", "low", "medium", "high", "xhigh", "max"]);
-var RawImages = z4.object({
-  backend: z4.enum(["gemini"]).default("gemini"),
-  model: z4.string().min(1).default("gemini-2.5-flash-image"),
-  dailyCap: z4.number().int().min(0).default(40)
+var RawImages = z5.object({
+  backend: z5.enum(["gemini"]).default("gemini"),
+  model: z5.string().min(1).default("gemini-2.5-flash-image"),
+  dailyCap: z5.number().int().min(0).default(40)
 }).strict().default({});
 var OMP_MODEL_ROLES = [
   "default",
@@ -1391,40 +1408,40 @@ var OMP_MODEL_ROLES = [
   "task",
   "advisor"
 ];
-var RawModelRoles = z4.record(z4.enum(OMP_MODEL_ROLES), z4.string().regex(MODEL_RE)).default({});
-var RawChannelModels = z4.record(z4.string().regex(/^[CDG][A-Z0-9]+$/, "channelModels keys must be Slack conversation ids"), z4.string().regex(MODEL_RE, "channel model must be provider/modelId")).default({});
-var RawSubagents = z4.object({
-  enabled: z4.boolean().default(true),
-  maxConcurrency: z4.number().int().min(1).max(8).default(3),
-  maxRecursionDepth: z4.number().int().min(1).max(3).default(2),
-  effort: z4.boolean().default(true)
+var RawModelRoles = z5.record(z5.enum(OMP_MODEL_ROLES), z5.string().regex(MODEL_RE)).default({});
+var RawChannelModels = z5.record(z5.string().regex(/^[CDG][A-Z0-9]+$/, "channelModels keys must be Slack conversation ids"), z5.string().regex(MODEL_RE, "channel model must be provider/modelId")).default({});
+var RawSubagents = z5.object({
+  enabled: z5.boolean().default(true),
+  maxConcurrency: z5.number().int().min(1).max(8).default(3),
+  maxRecursionDepth: z5.number().int().min(1).max(3).default(2),
+  effort: z5.boolean().default(true)
 }).strict().default({});
-var RawMemory = z4.object({ ompBackend: z4.enum(["off", "local"]).default("off") }).strict().default({});
+var RawMemory = z5.object({ ompBackend: z5.enum(["off", "local"]).default("off") }).strict().default({});
 var SKILL_SOURCE_SCOPES = ["shared", "instance"];
 var RawSkillSourceMetadata = {
-  owner: z4.string().min(1).optional(),
-  scope: z4.enum(SKILL_SOURCE_SCOPES).optional(),
-  requires: z4.array(z4.enum(CAPABILITY_IDS)).optional()
+  owner: z5.string().min(1).optional(),
+  scope: z5.enum(SKILL_SOURCE_SCOPES).optional(),
+  requires: z5.array(z5.enum(CAPABILITY_IDS)).optional()
 };
-var RawSkillSource = z4.union([
-  z4.object({
-    github: z4.string().min(1),
-    ref: z4.string().min(1).optional(),
-    path: z4.string().optional(),
+var RawSkillSource = z5.union([
+  z5.object({
+    github: z5.string().min(1),
+    ref: z5.string().min(1).optional(),
+    path: z5.string().optional(),
     ...RawSkillSourceMetadata
   }).strict(),
-  z4.object({ s3: z4.string().min(1), ...RawSkillSourceMetadata }).strict()
+  z5.object({ s3: z5.string().min(1), ...RawSkillSourceMetadata }).strict()
 ]);
-var RawSkills = z4.object({
-  sources: z4.array(RawSkillSource).optional(),
-  promoteTo: z4.string().min(1).optional()
+var RawSkills = z5.object({
+  sources: z5.array(RawSkillSource).optional(),
+  promoteTo: z5.string().min(1).optional()
 }).strict().default({});
-var RawInstanceConfig = z4.object({
-  name: z4.string().min(1),
-  model: z4.string().regex(MODEL_RE, "model must be provider/modelId"),
-  admins: z4.array(z4.string().min(1)).min(1, "admins must list at least one Slack user id"),
-  instructions: z4.string().default(""),
-  timezone: z4.string().default("UTC").refine(isValidTimezone, (tz) => ({ message: `unknown timezone: ${tz}` })),
+var RawInstanceConfig = z5.object({
+  name: z5.string().min(1),
+  model: z5.string().regex(MODEL_RE, "model must be provider/modelId"),
+  admins: z5.array(z5.string().min(1)).min(1, "admins must list at least one Slack user id"),
+  instructions: z5.string().default(""),
+  timezone: z5.string().default("UTC").refine(isValidTimezone, (tz) => ({ message: `unknown timezone: ${tz}` })),
   thinkingLevel: ThinkingLevel.optional(),
   skills: RawSkills,
   images: RawImages,
@@ -1501,24 +1518,24 @@ function parseInstanceConfig(yamlText) {
 }
 // ../core/src/company-capabilities.ts
 import { parse as parseYaml3 } from "yaml";
-import { z as z5 } from "zod";
+import { z as z6 } from "zod";
 var COMPANY_CAPABILITY_DECLARATION_VERSION = 1;
 var COMPANY_CAPABILITY_DECLARATION_MAX_COUNT = CAPABILITY_IDS.length;
 var COMPANY_CAPABILITY_DESCRIPTION_MAX_CHARS = 280;
 var COMPANY_CAPABILITY_DECLARATION_MAX_BYTES = 16 * 1024;
-var RawCompanyCapabilityDeclaration = z5.object({
-  id: z5.enum(CAPABILITY_IDS),
-  description: z5.string().min(1).max(COMPANY_CAPABILITY_DESCRIPTION_MAX_CHARS).refine((value) => value.trim() === value && !Array.from(value).some((char) => char.charCodeAt(0) < 32 || char.charCodeAt(0) === 127), "description must not have surrounding whitespace or control characters")
+var RawCompanyCapabilityDeclaration = z6.object({
+  id: z6.enum(CAPABILITY_IDS),
+  description: z6.string().min(1).max(COMPANY_CAPABILITY_DESCRIPTION_MAX_CHARS).refine((value) => value.trim() === value && !Array.from(value).some((char) => char.charCodeAt(0) < 32 || char.charCodeAt(0) === 127), "description must not have surrounding whitespace or control characters")
 }).strict();
-var RawCompanyCapabilityDeclarations = z5.object({
-  version: z5.literal(COMPANY_CAPABILITY_DECLARATION_VERSION),
-  capabilities: z5.array(RawCompanyCapabilityDeclaration).max(COMPANY_CAPABILITY_DECLARATION_MAX_COUNT)
+var RawCompanyCapabilityDeclarations = z6.object({
+  version: z6.literal(COMPANY_CAPABILITY_DECLARATION_VERSION),
+  capabilities: z6.array(RawCompanyCapabilityDeclaration).max(COMPANY_CAPABILITY_DECLARATION_MAX_COUNT)
 }).strict().superRefine((document, ctx) => {
   const seen = new Set;
   for (const [index, declaration] of document.capabilities.entries()) {
     if (seen.has(declaration.id)) {
       ctx.addIssue({
-        code: z5.ZodIssueCode.custom,
+        code: z6.ZodIssueCode.custom,
         path: ["capabilities", index, "id"],
         message: `duplicate capability declaration: ${declaration.id}`
       });
@@ -1585,13 +1602,13 @@ var LIVE_ENDPOINT_NAME = "live";
 // src/release/manifest.ts
 import { createHash as createHash3 } from "node:crypto";
 import { readFileSync as readFileSync2, statSync } from "node:fs";
-import { z as z6 } from "zod";
+import { z as z7 } from "zod";
 var RELEASE_VERSION_RE = /^v\d+\.\d+\.\d+$/;
-var Sha256 = z6.string().regex(/^[0-9a-f]{64}$/, "must be a lowercase hex sha256");
-var ReleaseArtifactSchema = z6.object({
-  key: z6.string().min(1),
+var Sha256 = z7.string().regex(/^[0-9a-f]{64}$/, "must be a lowercase hex sha256");
+var ReleaseArtifactSchema = z7.object({
+  key: z7.string().min(1),
   sha256: Sha256,
-  bytes: z6.number().int().nonnegative()
+  bytes: z7.number().int().nonnegative()
 }).strict();
 var SIGNING_ALGORITHMS = [
   "RSASSA_PSS_SHA_256",
@@ -1604,21 +1621,21 @@ var SIGNING_ALGORITHMS = [
   "ECDSA_SHA_384",
   "ECDSA_SHA_512"
 ];
-var ReleaseManifestSchema = z6.object({
-  version: z6.string().regex(RELEASE_VERSION_RE, "must be vMAJOR.MINOR.PATCH"),
-  gitCommit: z6.string().regex(/^[0-9a-f]{40}$/, "must be a full commit sha"),
-  createdAt: z6.string().datetime(),
-  agentImage: z6.object({
-    repository: z6.string().min(1),
-    tag: z6.string().min(1),
-    digest: z6.string().regex(/^sha256:[0-9a-f]{64}$/)
+var ReleaseManifestSchema = z7.object({
+  version: z7.string().regex(RELEASE_VERSION_RE, "must be vMAJOR.MINOR.PATCH"),
+  gitCommit: z7.string().regex(/^[0-9a-f]{40}$/, "must be a full commit sha"),
+  createdAt: z7.string().datetime(),
+  agentImage: z7.object({
+    repository: z7.string().min(1),
+    tag: z7.string().min(1),
+    digest: z7.string().regex(/^sha256:[0-9a-f]{64}$/)
   }).strict(),
-  lambda: z6.record(ReleaseArtifactSchema),
-  skills: z6.object({ key: z6.string().min(1), sha256: Sha256 }).strict(),
-  signature: z6.object({
-    kmsKeyArn: z6.string().min(1),
-    algorithm: z6.enum(SIGNING_ALGORITHMS),
-    value: z6.string().min(1)
+  lambda: z7.record(ReleaseArtifactSchema),
+  skills: z7.object({ key: z7.string().min(1), sha256: Sha256 }).strict(),
+  signature: z7.object({
+    kmsKeyArn: z7.string().min(1),
+    algorithm: z7.enum(SIGNING_ALGORITHMS),
+    value: z7.string().min(1)
   }).strict().optional()
 }).strict();
 function parseManifest(json) {

package/dist/{chunk-6v0ewr5b.js → chunk-sgh3agpa.js}

@@ -24,7 +24,7 @@ import {
   scheduledJobBindings,
   slackCommandPrefixes,
   webhookJobBindings
-} from "./chunk-7rd8c1th.js";
+} from "./chunk-g56mqyfx.js";
 
 // src/stacks/agent-stack.ts
 import * as cdk from "aws-cdk-lib";
@@ -1104,6 +1104,14 @@ class FoundationApi extends cdk2.Stack {
         `arn:${this.partition}:bedrock-agentcore:${this.region}:${this.account}:runtime/${names.runtimeName}*/runtime-endpoint/*`
       ]
     }));
+    invokerRole.addToPolicy(new iam2.PolicyStatement({
+      sid: "StopSingleUseRuntimeSessions",
+      actions: ["bedrock-agentcore:StopRuntimeSession"],
+      resources: [
+        `arn:${this.partition}:bedrock-agentcore:${this.region}:${this.account}:runtime/${names.runtimeName}*`,
+        `arn:${this.partition}:bedrock-agentcore:${this.region}:${this.account}:runtime/${names.runtimeName}*/runtime-endpoint/*`
+      ]
+    }));
     invokerRole.addToPolicy(new iam2.PolicyStatement({
       sid: "ReadRuntimeEndpointVersion",
       actions: ["bedrock-agentcore:GetAgentRuntimeEndpoint"],
@@ -1937,6 +1945,7 @@ function deployStatements(stack, names, targets, instance) {
       sid: "SmokeTestRuntime",
       actions: [
         "bedrock-agentcore:InvokeAgentRuntime",
+        "bedrock-agentcore:StopRuntimeSession",
         "bedrock-agentcore:GetAgentRuntime",
         "bedrock-agentcore:ListAgentRuntimeVersions",
         "bedrock-agentcore:GetAgentRuntimeEndpoint",

package/dist/src/index.js

@@ -10,7 +10,7 @@ import {
   FoundationPipeline,
   NewsletterStack,
   deployStatements
-} from "../chunk-6v0ewr5b.js";
+} from "../chunk-sgh3agpa.js";
 import {
   BUN_VERSION,
   LAMBDA_ENTRY_POINTS,
@@ -44,7 +44,7 @@ import {
   skillsKey,
   skipBundle,
   verifyManifest
-} from "../chunk-7rd8c1th.js";
+} from "../chunk-g56mqyfx.js";
 import"../chunk-dz91avrb.js";
 export {
   BUILD_TIMEOUT,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@deployfoundation/foundation-deploy",
-  "version": "0.2.39",
+  "version": "0.2.41",
   "description": "Deploys one Foundation instance from a published, signed release: the CDK app and the deploy tool.",
   "license": "UNLICENSED",
   "private": false,

package/src/deploy/endpoint.ts

@@ -214,6 +214,32 @@ export interface SmokeOptions {
   log?: (line: string) => void;
 }
 
+/**
+ * End a smoke session once it has answered. Its id is minted per invoke and
+ * never used again, so otherwise the microVM idles out its whole idle timeout
+ * billed for its memory (~2 GB-hours each; a deploy runs three). Best-effort:
+ * a deployer without the permission still deploys.
+ */
+async function stopSmokeSession(
+  runner: EndpointRunner,
+  opts: SmokeOptions,
+  log: (line: string) => void,
+): Promise<void> {
+  try {
+    await runner.capture([
+      "bedrock-agentcore",
+      "stop-runtime-session",
+      "--agent-runtime-arn",
+      opts.arn,
+      ...(opts.qualifier === undefined ? [] : ["--qualifier", opts.qualifier]),
+      "--runtime-session-id",
+      opts.sessionId,
+    ]);
+  } catch {
+    log("  (could not stop the smoke session; it ends at the idle timeout)");
+  }
+}
+
 /** Invoke the runtime and assert the response says what a healthy runtime says. */
 export async function smokeInvoke(runner: EndpointRunner, opts: SmokeOptions): Promise<void> {
   const log = opts.log ?? ((line: string) => console.log(line));
@@ -250,6 +276,7 @@ export async function smokeInvoke(runner: EndpointRunner, opts: SmokeOptions): P
   } finally {
     await rm(dir, { recursive: true, force: true });
   }
+  await stopSmokeSession(runner, opts, log);
   log(`  ${out.replaceAll("\n", "\n  ")}`);
   for (const needle of opts.expect ?? SMOKE_EXPECT)
     if (!out.includes(needle)) throw new Error(`smoke test: response does not contain ${needle}`);

package/src/deploy-permissions.ts

@@ -175,11 +175,13 @@ export function deployStatements(
     // `foundation-deploy post-deploy` (and `deploy`'s inline promotion): the
     // smoke test (`_smoke_test`, `_fs_probe`) against DEFAULT, then promotion
     // of the new version onto `live`, which CREATES `live` when an instance
-    // brought up by `deploy` alone has none yet (`promoteEndpoint`).
+    // brought up by `deploy` alone has none yet (`promoteEndpoint`). Each
+    // smoke session is stopped once it answers: nothing invokes it again.
     new iam.PolicyStatement({
       sid: "SmokeTestRuntime",
       actions: [
         "bedrock-agentcore:InvokeAgentRuntime",
+        "bedrock-agentcore:StopRuntimeSession",
         "bedrock-agentcore:GetAgentRuntime",
         "bedrock-agentcore:ListAgentRuntimeVersions",
         "bedrock-agentcore:GetAgentRuntimeEndpoint",

package/src/stacks/api-stack.ts

@@ -529,6 +529,19 @@ export class FoundationApi extends cdk.Stack {
         ],
       }),
     );
+    // A routine fire's session is single-use: the invoker ends it as soon as
+    // the fire answers instead of paying for its memory through the idle
+    // timeout (`isSingleUseSession` in gateway/src/invoker.ts).
+    invokerRole.addToPolicy(
+      new iam.PolicyStatement({
+        sid: "StopSingleUseRuntimeSessions",
+        actions: ["bedrock-agentcore:StopRuntimeSession"],
+        resources: [
+          `arn:${this.partition}:bedrock-agentcore:${this.region}:${this.account}:runtime/${names.runtimeName}*`,
+          `arn:${this.partition}:bedrock-agentcore:${this.region}:${this.account}:runtime/${names.runtimeName}*/runtime-endpoint/*`,
+        ],
+      }),
+    );
     // Which version the endpoint serves: the invoker names it in the session
     // id so a thread moves off a pre-deploy microVM on its next turn.
     invokerRole.addToPolicy(