test(hermes): add bounded managed Bedrock qualification

Co-Authored-By: Paperclip <noreply@paperclip.ing>

doc/evals.md

@@ -10,6 +10,12 @@ and Daytona, with independent account/model attribution. Its registered cells
 remain pending until their live evidence passes; its API-only scope leaves the
 subscription and lifecycle gates separate.
 
+The same guide registers the separate explicit-only `hermes-bedrock-connections`
+suite for one exact `us-east-1` inference profile on local and Daytona. Its two
+cells remain pending. They verify the explicitly selected personal grant and
+public region/protocol/model routing before and after native completion, using
+a short-lived AWS bearer credential and the existing budget/evidence pipeline.
+
 Paperclip has two live eval families with different questions, owners, and
 evidence. Choose the family before selecting a model, profile, or case.
 

doc/plans/2026-10-06-hermes-native-runner.md

@@ -709,3 +709,31 @@ because GitHub returned HTTP 429 for the pinned source archive. No native fixtur
 or provider execution occurred. This is retained as a download infrastructure
 failure; it is not evidence of a closure mismatch. The separate credential-free
 AWS build continues against its recorded immutable source.
+
+### Corrected cloud image and Bedrock fixture preparation, 2026-10-07
+
+The corrected credential-free AWS build completed successfully in 11.73 minutes
+at source `d389750b90787ea5240d0f8f4e92396a1d9d20a1`. Its immutable image digest is
+`sha256:5f457b7aed4dc224c77125edb7b17fd1aedc9e7bad0a9ce92d454cecc660e767`.
+The independent collected proof verifies fresh Linux closure
+`6166fadd24dae41b9fdfd994e6c7bc129e9317ed252b7dabdd777b691ab47771`
+and the original frozen dependency-lock SHA
+`f5ee14ee77b1dc7771fe455d619c880fc64b1e62e40a15704addc9f7430e5c50`.
+No provider credentials or local Docker were used. This proves the corrected
+Linux distribution; it does not prove execution-host namespace compatibility,
+browser behavior or Daytona. The ledger reserves $19 of the $25 cap after the
+additional $2 cloud-build reservation; actual billing remains incomplete.
+
+The new explicit-only `hermes-bedrock-connections` suite registers two pending
+completion cells on local and Daytona. Read-only AWS discovery confirmed the
+exact `us.anthropic.claude-haiku-4-5-20251001-v1:0` inference profile is active
+in `us-east-1`. The fixture uses an ephemeral region-bound bearer, a public
+personal Connections account and its explicitly selected grant. It verifies
+the public account's owner, grant, region, protocol, authentication and model
+catalog before task creation and after completion, and grades the actual native
+run's selected account, grant and exact inference profile. The server inherits
+no ambient AWS environment settings. It retains the existing 200-cent company
+and agent budget admission and one-attempt policy. All 129 affected fixture and
+catalog tests, Product E2E TypeScript, two-cell discovery and diff checks passed.
+No Bedrock inference, credential refresh or Daytona execution is claimed from
+this fixture preparation. Hermes remains pending qualification.

tests/runner-e2e/README.md

@@ -246,6 +246,8 @@ Shell variables take precedence over the local file. The recognized names are:
 - `KIMI_MODEL_API_KEY` (local-only pending Kimi CLI/ACP profiles)
 - `XAI_API_KEY` (local Grok API-key profile)
 - `GROK_AUTH_JSON` (local native Grok subscription profile)
+- `GEMINI_API_KEY` (Hermes Google API profile)
+- `AWS_BEARER_TOKEN_BEDROCK` (explicit Hermes Bedrock profile; use a short-lived token for the declared region)
 - `DAYTONA_API_KEY`
 - `CURSOR_AUTH_TOKEN` (extended Cursor candidate)
 - `COPILOT_GITHUB_TOKEN` (extended Copilot candidate)
@@ -1541,6 +1543,26 @@ protocols, Bedrock, refresh/revocation, images and live controls need their own
 qualification. The selected candidate metadata drives pre-credential Hermes
 provisioning and image/provider-pack selection for every Hermes profile.
 
+The separate explicit-only `hermes-bedrock-connections` suite declares two
+pending one-turn cells, on local and Daytona, for
+`us.anthropic.claude-haiku-4-5-20251001-v1:0` in `us-east-1`.
+Supply `AWS_BEARER_TOKEN_BEDROCK` ephemerally to the launcher after verifying
+the runtime and AWS identity. This suite creates a personal managed connection
+with Bedrock routing and selects its returned grant explicitly; routed accounts
+do not become responsible-user provider defaults. Public readback before and
+after the task must match the region, protocol, authentication, model catalog
+and account owner. The native run must match the selected connection, grant,
+owner, Hermes harness and exact inference profile. Both 200-cent budget checks,
+the single-attempt rule and normal evidence/cleanup contracts also apply.
+This scope proves Bedrock completion only; it does not qualify credential
+refresh or make a local AWS run equivalent to Daytona. The paid GitHub workflow
+does not provision this short-lived credential; use the explicit CLI path.
+
+```sh
+pnpm test:e2e:runner -- --list --suite hermes-bedrock-connections
+pnpm test:e2e:runner -- --id hermes-bedrock-connections.runner-acpx-hermes-bedrock.local.hello-complete --max-automatic-retries 0
+```
+
 The separate Runner Evals `extended-harnesses` campaign lives in the private
 `paperclip-evals` repository and grades semantic protocol behavior against the
 mock control plane. Neither suite substitutes for the other.

tests/runner-e2e/SECURITY.md

@@ -30,6 +30,18 @@ public structured evidence. Treat changes to the workflow, harness, fixture
 prompts, evidence packager, and publisher as security-sensitive production
 changes.
 
+## Short-lived Bedrock qualification
+
+The explicit-only `hermes-bedrock-connections` suite accepts a short-lived,
+region-bound `AWS_BEARER_TOKEN_BEDROCK` from the launcher. The browser fixture
+posts it through the sensitive public Connections API into encrypted storage;
+the agent selects that exact personal grant. The isolated server inherits no
+ambient `AWS_*` values, including profile, credential-file paths, static keys,
+session tokens or container-role endpoints. The selected bearer joins the
+attempt redaction set. Do not persist it in catalog data or evidence, copy an
+operator's AWS files, create a long-lived key for this fixture, or widen IAM
+permissions. The paid GitHub workflow does not supply this credential.
+
 ## GitHub authorization
 
 Set `RUNNER_E2E_ALLOWED_ACTOR_IDS` to a non-empty JSON array of numeric GitHub

tests/runner-e2e/catalog.test.ts

@@ -146,10 +146,10 @@ describe("runner E2E catalog", () => {
     expect(localIntegrityTasks).toHaveLength(2);
     expect(openRouterBreadthTasks).toHaveLength(3);
     expect(runnerSuites.map((suite) => suite.expectedMatrixSize)).toEqual([
-      63, 12, 6, 8, 2, 2, 2, 40, 10, 3, 16, 16, 2, 6, 8, 46, 23, 15, 52, 6, 6, 20, 26, 52, 28, 18, 2, 6, 6, 12, 10, 48, 16, 10, 2, 1, 1, 116,
+      63, 12, 6, 8, 2, 2, 2, 40, 10, 2, 3, 16, 16, 2, 6, 8, 46, 23, 15, 52, 6, 6, 20, 26, 52, 28, 18, 2, 6, 6, 12, 10, 48, 16, 10, 2, 1, 1, 116,
     ]);
-    expect(validateRunnerCatalog()).toHaveLength(718);
-    expect(new Set(runnerMatrix.map((entry) => entry.id)).size).toBe(718);
+    expect(validateRunnerCatalog()).toHaveLength(720);
+    expect(new Set(runnerMatrix.map((entry) => entry.id)).size).toBe(720);
     expect(
       runnerMatrix.filter((entry) => entry.suite.id === "core-compatibility"),
     ).toHaveLength(48);

tests/runner-e2e/catalog.ts

@@ -5,7 +5,7 @@ import { nativeCompletionProfile, NATIVE_COMPLETION_BUDGET_CENTS } from "./nativ
 import { chatConfirmationTasks } from "./chat-cases.js";
 import { buildConnectionSuite } from "./connection-cases.js";
 import { hiringTemplateTasks, hiringTemplateProfile, hiringTemplateDefinitionDigest } from "./hiring-template-cases.js";
-import { HERMES_API_CONNECTION_BUDGET_CENTS, hermesApiConnectionChoices, hermesApiConnectionDefinitionDigest } from "./hermes-api-connections.js";
+import { HERMES_API_CONNECTION_BUDGET_CENTS, hermesApiConnectionChoices, hermesApiConnectionDefinitionDigest, hermesBedrockConnectionChoice } from "./hermes-api-connections.js";
 import { nativeActiveStopTasks } from "./native-active-stop-tasks.js";
 import { cursorNativeTasks } from "./cursor-native-cases.js";
 import { instructionPersistenceTask } from "./instruction-persistence.js";
@@ -180,6 +180,7 @@ function nativeProfile(input: {
   supportedEnvironments?: readonly (typeof ENVIRONMENT_IDS)[number][];
   modelQualification?: RunnerProfileFixture["modelQualification"];
   ranking?: RunnerProfileFixture["ranking"];
+  managedConnectionRouting?: RunnerProfileFixture["managedConnectionRouting"];
 }): RunnerProfileFixture {
   return {
     ...input,
@@ -375,6 +376,15 @@ export const hermesApiConnectionProfiles: readonly RunnerProfileFixture[] = [
   })),
 ];
 
+/** Region-bound Bedrock credentials enter only through a managed connection. */
+export const hermesBedrockConnectionProfile = nativeProfile({
+  id: "runner-acpx-hermes-bedrock", label: "Hermes Bedrock (candidate)",
+  provider: "acpx", acpxAgent: "hermes", qualificationCandidate: "hermes",
+  credential: hermesBedrockConnectionChoice.credential, model: hermesBedrockConnectionChoice.model,
+  managedConnectionRouting: structuredClone(hermesBedrockConnectionChoice.routing),
+  modelQualification: { source: "candidate_runner_profile", qualificationId: "hermes:v2026.9.24:bedrock:us-east-1:discovery-2026-10-07:pending" },
+});
+
 /** Narrow legacy ACP lanes used only by the explicit context-integrity matrix. */
 export const legacyAcpxProfiles: readonly RunnerProfileFixture[] = [
   legacyProfile({
@@ -1215,6 +1225,20 @@ export const runnerSuites: readonly RunnerSuiteFixture[] = [
       coverage: "api-account-native-completion-only", sourceDigest: hermesApiConnectionDefinitionDigest,
     },
   },
+  {
+    id: "hermes-bedrock-connections", label: "Hermes managed Bedrock connection", manualOnly: true,
+    description: "One native browser task using an explicitly selected personal Bedrock account, exact inference profile and independently verified region.",
+    groups: ["native"], profiles: [hermesBedrockConnectionProfile], environments: runnerEnvironments,
+    tasks: [{ ...openRouterBreadthTasks.find(task => task.id === "hello-complete")!, automaticRetryPolicy: "single_attempt" }],
+    expectedMatrixSize: 2,
+    definitionMetadata: {
+      version: 1, qualification: "pending", scheduling: "explicit-only", authenticatedDiscoveryDate: "2026-10-07",
+      accountMethod: "api_key", accountMode: "delegated", providerTurns: 1, expectedUserSource: "public-account-owner-before-task",
+      routingSource: "public-selected-account-before-and-after-task", credentialScope: "ephemeral-region-bound-bearer",
+      budgetMonthlyCents: HERMES_API_CONNECTION_BUDGET_CENTS, maximumAttemptsPerCell: 1,
+      coverage: "bedrock-account-native-completion-only", sourceDigest: hermesApiConnectionDefinitionDigest,
+    },
+  },
   {
     id: "instruction-persistence", label: "Instruction Persistence",
     description: "Agent-owned text and binary files round trip through the editor, survive a server restart and fresh task, and synchronize concurrent edits per file with last-sync-wins.",
@@ -1595,6 +1619,7 @@ export function suiteDefinitionHash(suite: RunnerSuiteFixture) {
           id: profile.id,
           model: profile.model,
           qualification: profile.modelQualification,
+          ...(profile.managedConnectionRouting ? { managedConnectionRouting: profile.managedConnectionRouting } : {}),
         })),
         environments: suite.environments.map((environment) => ({
           id: environment.id,
@@ -1691,7 +1716,7 @@ function assertNoRawSecretValues(value: unknown, label: string) {
 
 export function validateRunnerCatalog(): MatrixExecution[] {
   const connectionSuite = runnerSuites.find(suite => suite.id === "provider-connections")!;
-  const allProfiles = [...connectionSuite.profiles, ...extendedHarnessProfiles, ...hermesApiConnectionProfiles.filter(p => !extendedHarnessProfiles.some(existing => existing.id === p.id)), ...runnerProfiles, ...legacyAcpxProfiles, ...pendingContextIntegrityProfiles, ...openRouterBreadthProfiles, ...everydayProfiles.filter(p => !runnerProfiles.some(existing => existing.id === p.id))];
+  const allProfiles = [...connectionSuite.profiles, ...extendedHarnessProfiles, hermesBedrockConnectionProfile, ...hermesApiConnectionProfiles.filter(p => !extendedHarnessProfiles.some(existing => existing.id === p.id)), ...runnerProfiles, ...legacyAcpxProfiles, ...pendingContextIntegrityProfiles, ...openRouterBreadthProfiles, ...everydayProfiles.filter(p => !runnerProfiles.some(existing => existing.id === p.id))];
   const allTasks = [
     ...connectionSuite.tasks,
     extendedHarnessFileTask,

tests/runner-e2e/harness-env.ts

@@ -25,7 +25,7 @@ const AMBIENT_EXTERNAL_STATE_KEYS = [
   "PAPERCLIP_STORAGE_S3_PREFIX",
   "PAPERCLIP_STORAGE_S3_FORCE_PATH_STYLE",
 ] as const;
-const PROVIDER_SECRET_KEY = /^(?:OPENAI|ANTHROPIC|OPENROUTER|DAYTONA|XAI|GROK|CURSOR|COPILOT|GITHUB|GH)(?:_|$)/;
+const PROVIDER_SECRET_KEY = /^(?:AWS|OPENAI|ANTHROPIC|OPENROUTER|DAYTONA|XAI|GROK|CURSOR|COPILOT|GITHUB|GH)(?:_|$)/;
 
 export function runnerE2EServerControlPaths(temporaryRoot: string) {
   const controlDirectory = path.join(temporaryRoot, "control");
@@ -116,7 +116,7 @@ export function buildRunnerE2EProcessEnvironment(
     const agent = execution.profile.qualificationCandidate;
     if (!agent || agent === "cursor") continue;
     const admittedSuite = execution.suite.id === "extended-harnesses"
-      || (execution.suite.id === "hermes-api-connections" && agent === "hermes")
+      || (["hermes-api-connections", "hermes-bedrock-connections"].includes(execution.suite.id) && agent === "hermes")
       || execution.suite.id === "rich-acp-warm-continuity";
     if (!admittedSuite || !execution.suite.manualOnly) {
       throw new Error("Candidate qualification requires an explicit provider qualification suite");

tests/runner-e2e/hermes-api-connections.ts

@@ -1,25 +1,41 @@
 import { createHash } from "node:crypto";
 import { readFileSync } from "node:fs";
+import { isDeepStrictEqual } from "node:util";
+import type { AiProviderRouting } from "../../packages/shared/src/ai-provider-routing.js";
 import type { RunnerProfileFixture } from "./types.js";
 
 const record = (value: unknown): Record<string, unknown> =>
   value && typeof value === "object" && !Array.isArray(value) ? value as Record<string, unknown> : {};
 const present = (value: unknown) => typeof value === "string" && value.trim().length > 0;
 export const HERMES_API_CONNECTION_BUDGET_CENTS = 200;
+export const isHermesConnectionSuite = (suiteId: string) =>
+  suiteId === "hermes-api-connections" || suiteId === "hermes-bedrock-connections";
+export const hermesBedrockConnectionChoice = {
+  credential: "AWS_BEARER_TOKEN_BEDROCK",
+  model: "us.anthropic.claude-haiku-4-5-20251001-v1:0",
+  routing: { kind: "bedrock", protocol: "bedrock", auth: "bearer", region: "us-east-1",
+    models: [{ id: "us.anthropic.claude-haiku-4-5-20251001-v1:0" }] },
+} as const satisfies { credential: RunnerProfileFixture["credential"]; model: string; routing: AiProviderRouting };
 
 /** Establish the intended personal-account user independently, before a paid task. */
 export async function captureHermesApiAccountOwner(input: {
   api: { get<T>(path: string): Promise<T> }; companyId: string; connectionId: string; provider: string;
+  expectedRouting?: AiProviderRouting; expectedGrantId?: string;
 }) {
   const listed = await input.api.get<{ currentUserId: string; connections: {
     id: string; companyId: string; provider: string; method: string; ownership: string; ownerUserId?: string; status: string;
+    routing?: AiProviderRouting; grantId?: string;
   }[] }>(`/api/companies/${input.companyId}/ai-connections`);
   const matches = listed.connections.filter(connection => connection.id === input.connectionId);
   const selected = matches[0];
-  return { expectedResponsibleUserId: selected?.ownerUserId ?? null, checks: [
+  return { expectedResponsibleUserId: selected?.ownerUserId ?? null,
+    ...(input.expectedRouting ? { routing: { expected: input.expectedRouting, observed: selected?.routing ?? null } } : {}),
+    checks: [
     { id: "selected-account-public-readback", passed: matches.length === 1 && selected?.companyId === input.companyId && selected.provider === input.provider && selected.method === "api_key" && selected.ownership === "personal" && selected.status === "connected" },
     { id: "authenticated-caller-known", passed: present(listed.currentUserId) },
     { id: "selected-account-owned-by-caller", passed: present(selected?.ownerUserId) && selected?.ownerUserId === listed.currentUserId },
+    ...(input.expectedRouting ? [{ id: "selected-account-routing", passed: isDeepStrictEqual(selected?.routing, input.expectedRouting) }] : []),
+    ...(input.expectedGrantId !== undefined ? [{ id: "selected-account-grant", passed: present(input.expectedGrantId) && selected?.grantId === input.expectedGrantId }] : []),
   ] };
 }
 
@@ -40,6 +56,7 @@ export async function captureHermesApiBudgets(input: {
 /** Grade public run/account/model metadata; a model's completion claim cannot supply it. */
 export function gradeHermesApiConnection(input: {
   companyId: string; agentId: string; issueId: string; connectionId: string; provider: string; model: string; expectedResponsibleUserId: string;
+  accountMode?: "responsible_user" | "delegated"; expectedGrantId?: string;
   runs: readonly {
     companyId: string; agentId: string; status: string; runtimeMode?: string;
     issueId?: string | null; responsibleUserId?: string | null;
@@ -53,7 +70,8 @@ export function gradeHermesApiConnection(input: {
   return [
     { id: "one-successful-native-run", passed: input.runs.length === 1 && run?.status === "succeeded" && run.runtimeMode === "native" },
     { id: "company-agent-task-scope", passed: run?.companyId === input.companyId && run.agentId === input.agentId && run.issueId === input.issueId },
-    { id: "selected-managed-api-account", passed: account.connectionId === input.connectionId && account.provider === input.provider && account.method === "api_key" && account.mode === "responsible_user" },
+    { id: "selected-managed-api-account", passed: account.connectionId === input.connectionId && account.provider === input.provider && account.method === "api_key" && account.mode === (input.accountMode ?? "responsible_user") },
+    ...(input.expectedGrantId !== undefined ? [{ id: "selected-managed-account-grant", passed: present(input.expectedGrantId) && account.grantId === input.expectedGrantId }] : []),
     { id: "responsible-user-attribution", passed: present(input.expectedResponsibleUserId) && run?.responsibleUserId === input.expectedResponsibleUserId && account.responsibleUserId === input.expectedResponsibleUserId },
     { id: "native-hermes-provider", passed: provider.kind === "acpx" && provider.agent === "hermes" && provider.model === input.model },
     { id: "exact-native-model", passed: identity.kind === "acpx" && identity.requestedModel === input.model && identity.effectiveModel === input.model },

tests/runner-e2e/hermes-bedrock-connections.test.ts

@@ -0,0 +1,61 @@
+import { describe, expect, it } from "vitest";
+import { runnerMatrix, runnerSuites, suiteDefinitionHash } from "./catalog.js";
+import { captureHermesApiAccountOwner, gradeHermesApiConnection, hermesBedrockConnectionChoice } from "./hermes-api-connections.js";
+import { buildPaperclipServerEnvironment, buildRunnerE2EProcessEnvironment } from "./harness-env.js";
+import { parseRunnerSelectors, selectRunnerExecutions } from "./selectors.js";
+
+describe("Hermes managed Bedrock qualification", () => {
+  const suite = runnerSuites.find(value => value.id === "hermes-bedrock-connections")!;
+  const cells = runnerMatrix.filter(value => value.suite.id === suite.id);
+  const routing = structuredClone(hermesBedrockConnectionChoice.routing);
+  const model = hermesBedrockConnectionChoice.model;
+
+  it("registers only two explicit pending single-attempt cells with exact routing", () => {
+    expect(cells).toHaveLength(2);
+    expect(new Set(cells.map(value => value.environment.id))).toEqual(new Set(["local", "daytona"]));
+    expect(suite).toMatchObject({ manualOnly: true, definitionMetadata: { qualification: "pending", accountMode: "delegated", providerTurns: 1, maximumAttemptsPerCell: 1, budgetMonthlyCents: 200 } });
+    for (const cell of cells) {
+      expect(cell.profile).toMatchObject({ credential: "AWS_BEARER_TOKEN_BEDROCK", model, managedConnectionRouting: routing, modelQualification: { source: "candidate_runner_profile" } });
+      expect(cell.requiredCredentials).toContain("AWS_BEARER_TOKEN_BEDROCK");
+      expect(cell.task).toMatchObject({ id: "hello-complete", expectedRunCount: 1, automaticRetryPolicy: "single_attempt" });
+      expect(JSON.parse(buildRunnerE2EProcessEnvironment({}, [cell]).PAPERCLIP_RUNNER_ACPX_QUALIFICATION!)).toEqual([{ agent: "hermes", model }]);
+    }
+    expect(selectRunnerExecutions(parseRunnerSelectors(["--all"])).some(value => value.suite.id === suite.id)).toBe(false);
+    expect(() => buildRunnerE2EProcessEnvironment({}, [{ ...cells[0]!, suite: { ...suite, manualOnly: false } }])).toThrow("explicit");
+  });
+
+  it("records routing changes in the historical definition digest", () => {
+    const changed = { ...suite, profiles: suite.profiles.map(profile => ({ ...profile, managedConnectionRouting: { ...routing, region: "us-west-2" } })) };
+    expect(suiteDefinitionHash(changed)).not.toBe(suiteDefinitionHash(suite));
+  });
+
+  it("keeps ambient AWS identities and the selected bearer out of the server environment", () => {
+    const source = { PATH: "/fixture/bin", AWS_PROFILE: "operator", AWS_ACCESS_KEY_ID: "fixture-access", AWS_SECRET_ACCESS_KEY: "fixture-secret", AWS_SESSION_TOKEN: "fixture-session", AWS_BEARER_TOKEN_BEDROCK: "fixture-bearer", AWS_SHARED_CREDENTIALS_FILE: "/operator/credentials", AWS_CONFIG_FILE: "/operator/config", AWS_CONTAINER_CREDENTIALS_RELATIVE_URI: "/operator/role" };
+    const server = buildPaperclipServerEnvironment(buildRunnerE2EProcessEnvironment(source, [cells[0]!]));
+    expect(server.PATH).toBe(source.PATH);
+    for (const key of Object.keys(source).filter(key => key.startsWith("AWS_"))) expect(server[key]).toBeUndefined();
+  });
+
+  it.each(["valid", "region", "protocol", "auth", "model", "missing-routing", "grant", "missing-grant"])("rejects incorrect public routing before a paid task: %s", async fault => {
+    const observed: Record<string, unknown> = structuredClone(routing);
+    if (fault === "region") observed.region = "us-west-2";
+    if (fault === "protocol") observed.protocol = "messages";
+    if (fault === "auth") observed.auth = "api_key";
+    if (fault === "model") observed.models = [{ id: "wrong-model" }];
+    const receipt = await captureHermesApiAccountOwner({ companyId: "company", connectionId: "account", provider: "anthropic", expectedRouting: routing, expectedGrantId: "grant", api: {
+      async get<T>() { return { currentUserId: "user", connections: [{ id: "account", companyId: "company", provider: "anthropic", method: "api_key", ownership: "personal", ownerUserId: "user", status: "connected", routing: fault === "missing-routing" ? undefined : observed, grantId: fault === "missing-grant" ? undefined : fault === "grant" ? "foreign" : "grant" }] } as T; },
+    } });
+    expect(receipt.checks.every(check => check.passed)).toBe(fault === "valid");
+    expect(receipt.routing?.observed).toEqual(fault === "missing-routing" ? null : observed);
+  });
+
+  it.each(["valid", "grant", "missing-grant", "mode", "owner", "model"])("grades the actual explicitly selected native account and inference profile: %s", fault => {
+    const account = { connectionId: "account", grantId: "grant", provider: "anthropic", method: "api_key", mode: "delegated", responsibleUserId: "user" };
+    if (fault === "grant") account.grantId = "foreign";
+    if (fault === "missing-grant") account.grantId = "";
+    if (fault === "mode") account.mode = "responsible_user";
+    if (fault === "owner") account.responsibleUserId = "foreign";
+    const checks = gradeHermesApiConnection({ companyId: "company", agentId: "agent", issueId: "task", connectionId: "account", expectedGrantId: "grant", provider: "anthropic", model, expectedResponsibleUserId: "user", accountMode: "delegated", runs: [{ companyId: "company", agentId: "agent", issueId: "task", responsibleUserId: "user", status: "succeeded", runtimeMode: "native", contextSnapshot: { aiConnection: account }, runnerProfileJson: { nativeExecutionInput: { provider: { kind: "acpx", agent: "hermes", model } }, sessionCheckpoint: { providerIdentity: { kind: "acpx", requestedModel: model, effectiveModel: fault === "model" ? "wrong-model" : model } } } }] });
+    expect(checks.every(check => check.passed)).toBe(fault === "valid");
+  });
+});

tests/runner-e2e/live-fixtures.test.ts

@@ -97,6 +97,7 @@ describe("live runner fixtures", () => {
     ["runner-acpx-hermes-api-openai", "hermes-api-connections", "hello-complete"],
     ["runner-acpx-hermes-api-xai", "hermes-api-connections", "hello-complete"],
     ["runner-acpx-hermes-api-google", "hermes-api-connections", "hello-complete"],
+    ["runner-acpx-hermes-bedrock", "hermes-bedrock-connections", "hello-complete"],
     ["runner-codex", "hiring-templates", "hire-coder-template-reuse"],
     ["runner-acpx-claude", "hiring-templates", "hire-coder-template-reuse"],
     ["runner-codex", "everyday-workflows", "hire-reuse"],
@@ -116,7 +117,7 @@ describe("live runner fixtures", () => {
       )!;
       const provider =
         profile === "runner-acpx-hermes-api-xai" ? "xai" : profile === "runner-acpx-hermes-api-google" ? "google"
-          : ["runner-acpx-claude", "runner-acpx-hermes-api-anthropic"].includes(profile) ? "anthropic"
+          : ["runner-acpx-claude", "runner-acpx-hermes-api-anthropic", "runner-acpx-hermes-bedrock"].includes(profile) ? "anthropic"
           : ["runner-opencode", "runner-acpx-hermes"].includes(profile) ? "openrouter" : "openai";
       let connected = false;
       let companyBody: any;
@@ -141,7 +142,8 @@ describe("live runner fixtures", () => {
               allAgents: false,
             });
             connected = true;
-            return { connectionId: "managed-account" };
+            expect(data.routing).toEqual(execution.profile.managedConnectionRouting);
+            return { connectionId: "managed-account", grantId: "managed-grant" };
           }
           return { id: "secret" };
         },
@@ -159,7 +161,7 @@ describe("live runner fixtures", () => {
         },
       });
       expect(connected).toBe(true);
-      if (suite === "hermes-api-connections") {
+      if (["hermes-api-connections", "hermes-bedrock-connections"].includes(suite)) {
         expect(companyBody.budgetMonthlyCents).toBe(200);
         expect(agentBody.budgetMonthlyCents).toBe(200);
       }
@@ -167,7 +169,9 @@ describe("live runner fixtures", () => {
       expect(agentBody.runtimeConfig.aiConnection).toEqual({
         provider,
         method: "api_key",
-        mode: "responsible_user",
+        ...(execution.profile.managedConnectionRouting
+          ? { mode: "delegated", connectionId: "managed-account", grantId: "managed-grant" }
+          : { mode: "responsible_user" }),
       });
       expect((fixtures as any).aiConnection.connectionId).toBe(
         "managed-account",

tests/runner-e2e/live-fixtures.ts

@@ -1,5 +1,6 @@
 import { NATIVE_COMPLETION_BUDGET_CENTS } from "./native-completion-defaults.js";
-import { HERMES_API_CONNECTION_BUDGET_CENTS } from "./hermes-api-connections.js";
+import { HERMES_API_CONNECTION_BUDGET_CENTS, isHermesConnectionSuite } from "./hermes-api-connections.js";
+import type { AiConnectionBinding } from "../../packages/shared/src/ai-connections.js";
 import path from "node:path";
 import { installedReleaseDaytonaPlugin } from "./installed-release.js";
 import { isManagedHiringCase } from "./chat-cases.js";
@@ -40,11 +41,7 @@ interface AgentRecord {
 }
 interface ManagedAccountFixture {
   connectionId: string;
-  binding: {
-    provider: "openai" | "anthropic" | "openrouter" | "xai" | "google";
-    method: "api_key";
-    mode: "responsible_user";
-  };
+  binding: Extract<AiConnectionBinding, { mode: "responsible_user" | "delegated" }>;
 }
 
 interface ProjectRecord {
@@ -140,7 +137,7 @@ export async function setupLiveFixtures(input: {
       return api.post<CompanyRecord>("/api/companies", {
         name: `Runner E2E ${execution.id} ${input.executionNonce}`,
         description: "Ephemeral paid full-stack runner acceptance fixture",
-        budgetMonthlyCents: execution.suite.id === "hermes-api-connections" ? HERMES_API_CONNECTION_BUDGET_CENTS
+        budgetMonthlyCents: isHermesConnectionSuite(execution.suite.id) ? HERMES_API_CONNECTION_BUDGET_CENTS
           : ["native-completion", "native-instruction-consolidation", "native-connection-guidance"].includes(execution.suite.id)
           || (execution.suite.id === "everyday-workflows" && ["hire-reuse", "delegate-feedback"].includes(execution.task.id)) ? NATIVE_COMPLETION_BUDGET_CENTS
           : execution.suite.id === "task-titles" ? TASK_TITLE_BUDGET_CENTS
@@ -252,7 +249,7 @@ export async function setupLiveFixtures(input: {
       async setup(resolved) {
         const company = value<CompanyRecord>(resolved, "company");
         const key = execution.profile.credential;
-        const provider = key === "ANTHROPIC_API_KEY" ? "anthropic"
+        const provider = key === "ANTHROPIC_API_KEY" || key === "AWS_BEARER_TOKEN_BEDROCK" ? "anthropic"
           : key === "OPENROUTER_API_KEY" ? "openrouter"
           : key === "OPENAI_API_KEY" ? "openai"
           : key === "XAI_API_KEY" ? "xai"
@@ -260,21 +257,26 @@ export async function setupLiveFixtures(input: {
         if (!provider) throw new Error(`Unsupported managed hiring credential ${key}`);
         const apiKey = input.credentials[key];
         if (!apiKey) throw new Error(`Missing credential ${key}`);
-        const account = await api.postSensitive<{ connectionId: string }>(
+        const routing = execution.profile.managedConnectionRouting;
+        const account = await api.postSensitive<{ connectionId: string; grantId: string }>(
           `/api/companies/${company.id}/ai-connections`,
           {
             provider,
             method: "api_key",
             name: `Runner E2E account ${input.executionNonce}`,
             ownership: "personal",
+            ...(routing ? { routing } : {}),
             apiKey,
             agentIds: [],
             allAgents: false,
           },
         );
+        if (routing && !account.grantId) throw new Error("Routed managed connection did not return its personal grant");
         return {
           connectionId: account.connectionId,
-          binding: { provider, method: "api_key", mode: "responsible_user" },
+          binding: routing
+            ? { provider, method: "api_key", mode: "delegated", connectionId: account.connectionId, grantId: account.grantId }
+            : { provider, method: "api_key", mode: "responsible_user" },
         };
       },
     });
@@ -304,7 +306,7 @@ export async function setupLiveFixtures(input: {
         || (execution.suite.id === "everyday-workflows" && ["hire-reuse", "delegate-feedback"].includes(execution.task.id))) {
         agent.budgetMonthlyCents = 1_000;
       }
-      if (execution.suite.id === "hermes-api-connections") agent.budgetMonthlyCents = HERMES_API_CONNECTION_BUDGET_CENTS;
+      if (isHermesConnectionSuite(execution.suite.id)) agent.budgetMonthlyCents = HERMES_API_CONNECTION_BUDGET_CENTS;
       if (managedHiring) {
         const account = value<ManagedAccountFixture>(resolved, "ai-connection");
         const config = agent.adapterConfig as Record<string, unknown>;

tests/runner-e2e/runner.spec.ts

@@ -1,5 +1,5 @@
 import { runPlanTaskFlow } from "./plan-task-flow.js";
-import { captureHermesApiAccountOwner, captureHermesApiBudgets, gradeHermesApiConnection } from "./hermes-api-connections.js";
+import { captureHermesApiAccountOwner, captureHermesApiBudgets, gradeHermesApiConnection, isHermesConnectionSuite } from "./hermes-api-connections.js";
 import { assertNativeCompletionSelection, NATIVE_COMPLETION_PREFLIGHT_ENV, verifyNativeCompletionPreflight } from "./native-completion-admission.js";
 import { assertNativeInstructionSelection, verifyNativeInstructionPreflight, NATIVE_INSTRUCTION_PREFLIGHT_ENV, NATIVE_INSTRUCTION_SUITE, NATIVE_INSTRUCTION_DEFAULT_SHA256 } from "./native-instruction-consolidation.js";
 import { captureNativeDefault, gradeNativeDefault, nativeCompletionWorkspaceDigest } from "./native-completion-defaults.js";
@@ -895,10 +895,12 @@ for (const execution of executions) {
         daytonaImage: process.env.PAPERCLIP_E2E_DAYTONA_IMAGE,
       });
 
-      if (execution.suite.id === "hermes-api-connections") {
+      if (isHermesConnectionSuite(execution.suite.id)) {
         if (!fixtures.aiConnection) throw new Error("Hermes API account is missing before task creation");
         hermesApiAccountOwner = await captureHermesApiAccountOwner({ api, companyId: fixtures.company.id,
-          connectionId: fixtures.aiConnection.connectionId, provider: fixtures.aiConnection.binding.provider });
+          connectionId: fixtures.aiConnection.connectionId, provider: fixtures.aiConnection.binding.provider,
+          ...(execution.profile.managedConnectionRouting ? { expectedRouting: execution.profile.managedConnectionRouting,
+            expectedGrantId: fixtures.aiConnection.binding.mode === "delegated" ? fixtures.aiConnection.binding.grantId : "" } : {}) });
         await writeSanitizedJson(snapshotsDir, "hermes-api-owner-before-execution.json", hermesApiAccountOwner, secrets);
         if (!hermesApiAccountOwner.checks.every(check => check.passed)) throw new Error("Hermes API owner failed admission before task creation");
         const receipt = await captureHermesApiBudgets({ api, companyId: fixtures.company.id, agentId: fixtures.agent.id });
@@ -2877,11 +2879,22 @@ for (const execution of executions) {
           }
         }
       }
-      if (execution.suite.id === "hermes-api-connections") {
+      if (isHermesConnectionSuite(execution.suite.id)) {
         if (!fixtures?.aiConnection || !issue?.id || !hermesApiAccountOwner?.expectedResponsibleUserId) throw new Error("Hermes connection qualification is missing its selected account, expected user or task");
         const checks = gradeHermesApiConnection({ companyId: fixtures.company.id, agentId: fixtures.agent.id,
           issueId: issue.id, connectionId: fixtures.aiConnection.connectionId, provider: fixtures.aiConnection.binding.provider,
-          expectedResponsibleUserId: hermesApiAccountOwner.expectedResponsibleUserId, model: execution.profile.model, runs: selectedRuns });
+          expectedResponsibleUserId: hermesApiAccountOwner.expectedResponsibleUserId, model: execution.profile.model, runs: selectedRuns,
+          accountMode: fixtures.aiConnection.binding.mode,
+          ...(execution.profile.managedConnectionRouting ? { expectedGrantId: fixtures.aiConnection.binding.mode === "delegated" ? fixtures.aiConnection.binding.grantId : "" } : {}) });
+        if (execution.profile.managedConnectionRouting) {
+          const after = await captureHermesApiAccountOwner({ api, companyId: fixtures.company.id,
+            connectionId: fixtures.aiConnection.connectionId, provider: fixtures.aiConnection.binding.provider,
+            expectedRouting: execution.profile.managedConnectionRouting,
+            expectedGrantId: fixtures.aiConnection.binding.mode === "delegated" ? fixtures.aiConnection.binding.grantId : "" });
+          checks.push(...after.checks.map(check => ({ ...check, id: `after-task-${check.id}` })),
+            { id: "account-owner-preserved", passed: after.expectedResponsibleUserId === hermesApiAccountOwner.expectedResponsibleUserId });
+          await writeSanitizedJson(snapshotsDir, "hermes-api-owner-after-execution.json", after, secrets);
+        }
         matcherResults.push(...checks.map(check => ({ matcher: { kind: "json_path" as const, path: `hermesConnection.${check.id}`, expected: true }, passed: check.passed, detail: "Public native run metadata must match the selected managed API account and exact model." })));
         await writeSanitizedJson(snapshotsDir, "hermes-api-connection.json", { checks }, secrets);
         expect(checks.every(check => check.passed), "Hermes managed account and native model attribution").toBe(true);

tests/runner-e2e/types.ts

@@ -1,3 +1,5 @@
+import type { AiProviderRouting } from "../../packages/shared/src/ai-provider-routing.js";
+
 export const CREDENTIAL_NAMES = [
   "OPENAI_API_KEY",
   "ANTHROPIC_API_KEY",
@@ -6,6 +8,7 @@ export const CREDENTIAL_NAMES = [
   "XAI_API_KEY",
   "GROK_AUTH_JSON",
   "GEMINI_API_KEY",
+  "AWS_BEARER_TOKEN_BEDROCK",
   "DAYTONA_API_KEY",
   "CURSOR_AUTH_TOKEN",
   "COPILOT_GITHUB_TOKEN",
@@ -87,6 +90,8 @@ export interface RunnerProfileFixture {
     sourceUrl: string;
   };
   credential: Exclude<CredentialName, "DAYTONA_API_KEY">;
+  /** Value-free routing saved on an explicitly selected managed connection. */
+  managedConnectionRouting?: AiProviderRouting;
   supportedEnvironments: readonly RunnerEnvironmentId[];
   expectedRuntimeMode: RunnerGeneration;
   expectedRuntimeMetadata: {