rivet-dev · NathanFlurry · Jul 8, 2026
diff --git a/docs-internal/registry-parity-worklist.md b/docs-internal/registry-parity-worklist.md
@@ -701,6 +701,19 @@ real e2e tests that prove Linux-parity behavior — not smoke tests.
     `2026-07-08T15-00-00-0700-item12-pi-cli-sidecar-build-after-cc-restore-final.log`;
     `2026-07-08T15-01-00-0700-item12-pi-cli-focused-final-after-cc-restore.log`.
     Rev: `xqtkmsyn`.
+  - **claude — DONE.** Enabled the existing real `createSession('claude')`
+    session suite in default core Vitest coverage, projected the actual
+    `@agentos-software/claude-code` agent package, and replaced the missing
+    test-only `xu` binary dependency with a real PATH-backed `sh` command from
+    registry coreutils. The suite proves Claude ACP shell/tool flow, text-only
+    responses, nested Node `execSync`/`spawn`, session metadata/lifecycle,
+    modes, and raw ACP sends. The flat node_modules fixture cache now lives
+    outside root `node_modules` so VM mounts survive dependency refreshes.
+    Proof:
+    `2026-07-08T15-17-00-0700-item12-claude-cc-cache-restored-after-cache-move.log`;
+    `2026-07-08T15-17-00-0700-item12-claude-sidecar-build-after-cache-move.log`;
+    `2026-07-08T15-19-00-0700-item12-claude-session-after-cache-move.log`.
+    Rev: `rxmoulty`.
 - **Objective:** write real e2e tests proving each behaves like its Linux
  counterpart (jq processes real JSON, sed edits streams, tar round-trips archives,
   gzip round-trips, etc.); agents exercise the real ACP

diff --git a/packages/core/tests/claude-session.test.ts b/packages/core/tests/claude-session.test.ts
@@ -1,6 +1,6 @@
 import { resolve } from "node:path";
+import claude from "@agentos-software/claude-code";
 import type { Fixture, LLMock, ToolCall } from "@copilotkit/llmock";
-import { moduleAccessMounts } from "./helpers/node-modules-mount.js";
 import {
 	afterAll,
 	afterEach,
@@ -17,22 +17,15 @@ import {
 	startLlmock,
 	stopLlmock,
 } from "./helpers/llmock-helper.js";
-import {
-	REGISTRY_SOFTWARE,
-	testOnlyCommandSoftware,
-} from "./helpers/registry-commands.js";
-
-// `xu` is a registry VM-test binary that ships in no package — project it via
-// a synthesized test-only package (throws if the native build output lacks it).
-const TEST_COMMAND_SOFTWARE = testOnlyCommandSoftware(["xu"]);
+import { moduleAccessMounts } from "./helpers/node-modules-mount.js";
+import { REGISTRY_SOFTWARE } from "./helpers/registry-commands.js";
 
 const MODULE_ACCESS_CWD = resolve(import.meta.dirname, "..");
-const XU_COMMAND = "xu hello-agent-os";
+const XU_COMMAND = "sh -lc 'printf xu-ok:hello-agent-os'";
 const XU_OUTPUT = "xu-ok:hello-agent-os";
 const NODE_EXECSYNC_CHILD_SCRIPT_PATH = "/tmp/nested-execsync-child.cjs";
 const NODE_EXECSYNC_SCRIPT_PATH = "/tmp/nested-execsync.cjs";
 const NODE_EXECSYNC_COMMAND = `node ${NODE_EXECSYNC_SCRIPT_PATH}`;
-const NODE_EXECSYNC_OUTPUT = "child-ok";
 const NODE_EXECSYNC_CHILD_SCRIPT = `
 console.log("child-ok");
 `.trimStart();
@@ -153,15 +146,15 @@ describe("full createSession('claude')", () => {
 		vm = await AgentOs.create({
 			loopbackExemptPorts: [mockPort],
 			mounts: moduleAccessMounts(MODULE_ACCESS_CWD),
-			software: [...REGISTRY_SOFTWARE, TEST_COMMAND_SOFTWARE],
+			software: [claude, ...REGISTRY_SOFTWARE],
 		});
 	});
 
 	afterEach(async () => {
 		await vm.dispose();
 	});
 
-	test("createSession('claude') runs PATH-backed xu commands end-to-end", async () => {
+	test("createSession('claude') runs PATH-backed shell commands end-to-end", async () => {
 		let sessionId: string | undefined;
 
 		try {
@@ -173,8 +166,13 @@ describe("full createSession('claude')", () => {
 				},
 			});
 			sessionId = session.sessionId;
-			vm.onPermissionRequest(sessionId, (request) => {
-				void vm.respondPermission(sessionId!, request.permissionId, "once");
+			const activeSessionId = sessionId;
+			vm.onPermissionRequest(activeSessionId, (request) => {
+				void vm.respondPermission(
+					activeSessionId,
+					request.permissionId,
+					"once",
+				);
 			});
 
 			const events: { method: string; params?: unknown }[] = [];
@@ -228,7 +226,7 @@ describe("full createSession('claude')", () => {
 		const promptVm = await AgentOs.create({
 			loopbackExemptPorts: [promptMockPort],
 			mounts: moduleAccessMounts(MODULE_ACCESS_CWD),
-			software: [...REGISTRY_SOFTWARE, TEST_COMMAND_SOFTWARE],
+			software: [claude, ...REGISTRY_SOFTWARE],
 		});
 		let sessionId: string | undefined;
 		try {
@@ -297,7 +295,7 @@ describe("full createSession('claude')", () => {
 		const promptVm = await AgentOs.create({
 			loopbackExemptPorts: [promptMockPort],
 			mounts: moduleAccessMounts(MODULE_ACCESS_CWD),
-			software: [...REGISTRY_SOFTWARE, TEST_COMMAND_SOFTWARE],
+			software: [claude, ...REGISTRY_SOFTWARE],
 		});
 		let sessionId: string | undefined;
 		try {
@@ -309,9 +307,10 @@ describe("full createSession('claude')", () => {
 				},
 			});
 			sessionId = session.sessionId;
-			promptVm.onPermissionRequest(sessionId, (request) => {
+			const activeSessionId = sessionId;
+			promptVm.onPermissionRequest(activeSessionId, (request) => {
 				void promptVm.respondPermission(
-					sessionId!,
+					activeSessionId,
 					request.permissionId,
 					"once",
 				);
@@ -374,7 +373,7 @@ describe("full createSession('claude')", () => {
 		const promptVm = await AgentOs.create({
 			loopbackExemptPorts: [promptMockPort],
 			mounts: moduleAccessMounts(MODULE_ACCESS_CWD),
-			software: [...REGISTRY_SOFTWARE, TEST_COMMAND_SOFTWARE],
+			software: [claude, ...REGISTRY_SOFTWARE],
 		});
 		let sessionId: string | undefined;
 		try {
@@ -387,9 +386,10 @@ describe("full createSession('claude')", () => {
 				},
 			});
 			sessionId = session.sessionId;
-			promptVm.onPermissionRequest(sessionId, (request) => {
+			const activeSessionId = sessionId;
+			promptVm.onPermissionRequest(activeSessionId, (request) => {
 				void promptVm.respondPermission(
-					sessionId!,
+					activeSessionId,
 					request.permissionId,
 					"once",
 				);

diff --git a/packages/core/tests/helpers/fixture-node-modules.ts b/packages/core/tests/helpers/fixture-node-modules.ts
@@ -13,6 +13,7 @@ import {
 	unlinkSync,
 	writeFileSync,
 } from "node:fs";
+import { tmpdir } from "node:os";
 import { dirname, join, resolve, sep } from "node:path";
 
 /**
@@ -129,12 +130,11 @@ export function ensureFlatNodeModules(cwd: string): string {
 
 	const repoRoot = findRepoRoot(cwd);
 	const safe = packageName.replace(/[^a-z0-9]+/gi, "_");
-	const cacheRoot = join(
-		repoRoot,
-		"node_modules",
-		".cache",
-		"agentos-flat-fixtures",
-	);
+	const repoSafe = repoRoot.replace(/[^a-z0-9]+/gi, "_");
+	const cacheBase =
+		process.env.AGENTOS_FLAT_FIXTURE_CACHE_ROOT ??
+		join(tmpdir(), "agentos-flat-fixtures");
+	const cacheRoot = join(cacheBase, repoSafe);
 	mkdirSync(cacheRoot, { recursive: true });
 	const target = join(cacheRoot, safe);
 	const readyMarker = join(target, ".ready");

diff --git a/packages/core/vitest.config.ts b/packages/core/vitest.config.ts
@@ -10,7 +10,6 @@ import { configDefaults, defineConfig } from "vitest/config";
 const SLOW_E2E_FILES = [
 	"tests/wasm-commands.test.ts", // ~24m
 	"tests/session-cleanup.test.ts", // ~12m
-	"tests/claude-session.test.ts", // ~7.5m
 	"tests/execute.test.ts",
 	"tests/filesystem.test.ts",
 	"tests/native-sidecar-process.test.ts",