@vellumai/assistant 0.8.5 → 0.8.7-dev.202606052118.34cd356

package/src/__tests__/plugin-bootstrap.test.ts

@@ -31,20 +31,15 @@ mock.module("../security/secure-keys.js", () => ({
   getSecureKeyAsync: getSecureKeyAsyncMock,
 }));
 
-import {
-  _setOverridesForTesting,
-  clearFeatureFlagOverridesCache,
-} from "../config/assistant-feature-flags.js";
-import type { AssistantConfig } from "../config/schema.js";
-import {
-  bootstrapPlugins,
-  type DaemonContext,
-} from "../daemon/external-plugins-bootstrap.js";
+import { clearFeatureFlagOverridesCache } from "../config/assistant-feature-flags.js";
+import { bootstrapPlugins } from "../daemon/external-plugins-bootstrap.js";
 import { runShutdownHooks } from "../daemon/shutdown-registry.js";
 import { RiskLevel } from "../permissions/types.js";
+import { registerDefaultPlugins } from "../plugins/defaults/index.js";
 import {
-  getInjectors,
+  closeRegistration,
   getMiddlewaresFor,
+  getRegisteredPlugins,
   registerPlugin,
   resetPluginRegistryForTests,
 } from "../plugins/registry.js";
@@ -54,6 +49,8 @@ import {
   PluginExecutionError,
   type PluginInitContext,
 } from "../plugins/types.js";
+import { APP_VERSION } from "../version.js";
+import { setOverridesForTesting } from "./feature-flag-test-helpers.js";
 
 // Redirect plugin storage directory creation into a per-process temp tree so
 // the test doesn't touch the developer's real ~/.vellum.
@@ -63,12 +60,6 @@ const TEST_WORKSPACE_DIR = join(
 );
 process.env.VELLUM_WORKSPACE_DIR = TEST_WORKSPACE_DIR;
 
-const fakeConfig = {} as unknown as AssistantConfig;
-const fakeCtx: DaemonContext = {
-  config: fakeConfig,
-  assistantVersion: "9.9.9-test",
-};
-
 /**
  * Test helper. Accepts the new `hooks` bag and ALSO legacy top-level
  * `init` / `onShutdown` for ergonomics — the helper merges them into a
@@ -125,7 +116,7 @@ describe("plugin bootstrap", () => {
     getSecureKeyAsyncMock.mockReset();
     getSecureKeyAsyncMock.mockImplementation(async () => undefined);
     // Reset feature-flag cache so tests start from a known state. Individual
-    // tests that exercise `requiresFlag` use `_setOverridesForTesting(...)`
+    // tests that exercise `requiresFlag` use `setOverridesForTesting(...)`
     // to install their own overrides.
     clearFeatureFlagOverridesCache();
     // Clean storage directory between runs so nothing leaks across cases.
@@ -141,7 +132,7 @@ describe("plugin bootstrap", () => {
     });
     registerPlugin(plugin);
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     expect(received).toBeDefined();
     const ctx = received!;
@@ -157,7 +148,7 @@ describe("plugin bootstrap", () => {
       join(TEST_WORKSPACE_DIR, "plugins-data", "alpha"),
     );
     expect(existsSync(ctx.pluginStorageDir)).toBe(true);
-    expect(ctx.assistantVersion).toBe("9.9.9-test");
+    expect(ctx.assistantVersion).toBe(APP_VERSION);
   });
 
   test("credential resolution: init receives the resolved value under credentials[key]", async () => {
@@ -178,7 +169,7 @@ describe("plugin bootstrap", () => {
     );
     registerPlugin(plugin);
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     expect(getSecureKeyAsyncMock).toHaveBeenCalledTimes(1);
     expect(getSecureKeyAsyncMock).toHaveBeenCalledWith("some-key");
@@ -198,7 +189,7 @@ describe("plugin bootstrap", () => {
 
     let caught: unknown;
     try {
-      await bootstrapPlugins(fakeCtx);
+      await bootstrapPlugins();
     } catch (err) {
       caught = err;
     }
@@ -230,7 +221,7 @@ describe("plugin bootstrap", () => {
 
     let caught: unknown;
     try {
-      await bootstrapPlugins(fakeCtx);
+      await bootstrapPlugins();
     } catch (err) {
       caught = err;
     }
@@ -278,7 +269,7 @@ describe("plugin bootstrap", () => {
 
     let caught: unknown;
     try {
-      await bootstrapPlugins(fakeCtx);
+      await bootstrapPlugins();
     } catch (err) {
       caught = err;
     }
@@ -306,7 +297,7 @@ describe("plugin bootstrap", () => {
       }),
     );
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
     await runShutdownHooks("test-shutdown");
 
     // The last plugin to register must shut down first; the first to register
@@ -322,7 +313,38 @@ describe("plugin bootstrap", () => {
     // tool-result truncate, etc.). Just assert bootstrap completes without
     // throwing — the surface of defaults is verified in each pipeline's own
     // dedicated test file.
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
+  });
+
+  test("startup ordering: defaults registered before the window closes survive bootstrap replay", async () => {
+    /**
+     * Reproduces the daemon startup ordering the registration regression
+     * guards: defaults must register before `loadUserPlugins()` closes the
+     * window, so the `registerDefaultPlugins` replay inside `bootstrapPlugins`
+     * is swallowed by the duplicate-name check instead of throwing on the
+     * closed window and leaving every default unregistered.
+     */
+
+    // GIVEN the first-party defaults have registered while the window is open
+    // (what `initializePlugins()` does at daemon startup)
+    registerDefaultPlugins();
+
+    // AND a user plugin registers after them (what `loadUserPlugins()` does)
+    registerPlugin(buildPlugin("user-after-defaults"));
+
+    // AND the registration window has since closed
+    closeRegistration();
+
+    // WHEN bootstrap runs and replays the default registration
+    await bootstrapPlugins();
+
+    // THEN the defaults are still registered, ahead of the user plugin, so the
+    // middleware onion order is unchanged
+    const names = getRegisteredPlugins().map((p) => p.manifest.name);
+    expect(names).toContain("default-compaction");
+    expect(names.indexOf("default-compaction")).toBeLessThan(
+      names.indexOf("user-after-defaults"),
+    );
   });
 
   // ── requiresFlag gating (G2.2) ──────────────────────────────────────────
@@ -335,11 +357,11 @@ describe("plugin bootstrap", () => {
   //   - no shutdown hook entry is installed (nothing to tear down later).
   // Plugins without `requiresFlag` are unaffected.
   //
-  // Uses `_setOverridesForTesting` to control the resolver deterministically
+  // Uses `setOverridesForTesting` to control the resolver deterministically
   // — no disk writes, no gateway IPC, no reliance on registry defaults.
 
   test("requiresFlag enabled: plugin inits normally", async () => {
-    _setOverridesForTesting({ "plugin-gated-enabled": true });
+    setOverridesForTesting({ "plugin-gated-enabled": true });
 
     let initFired = false;
     const plugin = buildPlugin(
@@ -353,13 +375,13 @@ describe("plugin bootstrap", () => {
     );
     registerPlugin(plugin);
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     expect(initFired).toBe(true);
   });
 
   test("requiresFlag disabled: init does not fire and no tools/routes/skills are registered", async () => {
-    _setOverridesForTesting({ "plugin-gated-disabled": false });
+    setOverridesForTesting({ "plugin-gated-disabled": false });
 
     let initFired = false;
     // Attach tool/route/skill contributions alongside init. If gating works,
@@ -374,6 +396,7 @@ describe("plugin bootstrap", () => {
           {
             name: "gated-off-tool",
             description: "should not be registered",
+            category: "test",
             defaultRiskLevel: RiskLevel.Low,
             executionTarget: "sandbox",
             input_schema: { type: "object", properties: {}, required: [] },
@@ -409,7 +432,7 @@ describe("plugin bootstrap", () => {
     const { matchSkillRoute } =
       await import("../runtime/skill-route-registry.js");
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     // init must not have fired.
     expect(initFired).toBe(false);
@@ -432,7 +455,7 @@ describe("plugin bootstrap", () => {
     });
     registerPlugin(plugin);
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     expect(initFired).toBe(true);
   });
@@ -440,7 +463,7 @@ describe("plugin bootstrap", () => {
   test("requiresFlag: one disabled flag out of several skips the plugin", async () => {
     // When ANY listed flag is disabled, the plugin is skipped wholesale —
     // this prevents sneaky partial activation on AND semantics.
-    _setOverridesForTesting({
+    setOverridesForTesting({
       "plugin-multi-a": true,
       "plugin-multi-b": false,
     });
@@ -457,56 +480,42 @@ describe("plugin bootstrap", () => {
     );
     registerPlugin(plugin);
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     expect(initFired).toBe(false);
   });
 
-  test("requiresFlag disabled: plugin middleware and injectors are dropped from the registry", async () => {
+  test("requiresFlag disabled: plugin middleware is dropped from the registry", async () => {
     // Regression: prior to the unregisterPlugin() call on the flag-gated skip
-    // path, `getMiddlewaresFor()` and `getInjectors()` iterated over every
-    // entry in `registeredPlugins` — so a gated-off plugin's middleware and
-    // injectors still ran on every pipeline invocation and system-prompt
-    // assembly even though `init()` had never fired to set up the state they
-    // depended on.
-    _setOverridesForTesting({ "plugin-middleware-disabled": false });
-
-    const gatedMiddleware: PipelineMiddlewareMap["llmCall"] = async (
+    // path, `getMiddlewaresFor()` iterated over every entry in
+    // `registeredPlugins` — so a gated-off plugin's middleware still ran on
+    // every pipeline invocation even though `init()` had never fired to set up
+    // the state it depended on.
+    setOverridesForTesting({ "plugin-middleware-disabled": false });
+
+    const gatedMiddleware: PipelineMiddlewareMap["compaction"] = async (
       args,
       next,
     ) => next(args);
     const plugin = buildPlugin(
       "gated-middleware",
       {
-        middleware: { llmCall: gatedMiddleware },
-        injectors: [
-          {
-            name: "gated-middleware-injector",
-            order: 100,
-            async produce() {
-              return null;
-            },
-          },
-        ],
+        middleware: { compaction: gatedMiddleware },
       },
       { requiresFlag: ["plugin-middleware-disabled"] },
     );
     registerPlugin(plugin);
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
-    // Neither the middleware slot nor the injector list should expose the
-    // flag-gated plugin's contributions. The default plugins also contribute
-    // llmCall middleware / injectors, so we key on identity rather than
-    // asserting empty lists.
-    expect(getMiddlewaresFor("llmCall")).not.toContain(gatedMiddleware);
-    expect(
-      getInjectors().some((i) => i.name === "gated-middleware-injector"),
-    ).toBe(false);
+    // The middleware slot must not expose the flag-gated plugin's contribution.
+    // The default plugins also contribute compaction middleware, so we key on
+    // identity rather than asserting an empty list.
+    expect(getMiddlewaresFor("compaction")).not.toContain(gatedMiddleware);
   });
 
   test("requiresFlag disabled: no shutdown hook entry installed for the skipped plugin", async () => {
-    _setOverridesForTesting({ "plugin-shutdown-flag": false });
+    setOverridesForTesting({ "plugin-shutdown-flag": false });
 
     let shutdownFired = false;
     const plugin = buildPlugin(
@@ -521,7 +530,7 @@ describe("plugin bootstrap", () => {
     );
     registerPlugin(plugin);
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
     await runShutdownHooks("test-shutdown");
 
     // The shutdown hook is a single registered callback that walks a

package/src/__tests__/plugin-registry.test.ts

@@ -10,7 +10,6 @@ import { beforeEach, describe, expect, test } from "bun:test";
 
 import {
   closeRegistration,
-  getInjectors,
   getMiddlewaresFor,
   getRegisteredPlugins,
   registerPlugin,
@@ -19,7 +18,6 @@ import {
 import {
   type CompactionArgs,
   type CompactionResult,
-  type Injector,
   type Middleware,
   type Plugin,
   PluginExecutionError,
@@ -149,31 +147,6 @@ describe("plugin registry", () => {
     expect(() => registerPlugin(bad)).toThrow(/manifest\.version is required/);
   });
 
-  test("getInjectors returns injectors sorted by order ascending", () => {
-    const high: Injector = {
-      name: "high-order",
-      order: 20,
-      async produce() {
-        return null;
-      },
-    };
-    const low: Injector = {
-      name: "low-order",
-      order: 10,
-      async produce() {
-        return null;
-      },
-    };
-
-    // Register the higher-order plugin first so registration order alone
-    // would produce the wrong sequence — the test proves sort-by-order wins.
-    registerPlugin(buildPlugin("high", { injectors: [high] }));
-    registerPlugin(buildPlugin("low", { injectors: [low] }));
-
-    const injectors = getInjectors();
-    expect(injectors.map((i) => i.name)).toEqual(["low-order", "high-order"]);
-  });
-
   test("getMiddlewaresFor returns middleware in registration order", () => {
     const firstMw: Middleware<CompactionArgs, CompactionResult> = async (
       args,

package/src/__tests__/plugin-route-contribution.test.ts

@@ -43,11 +43,7 @@ mock.module("../security/secure-keys.js", () => ({
   getSecureKeyAsync: getSecureKeyAsyncMock,
 }));
 
-import type { AssistantConfig } from "../config/schema.js";
-import {
-  bootstrapPlugins,
-  type DaemonContext,
-} from "../daemon/external-plugins-bootstrap.js";
+import { bootstrapPlugins } from "../daemon/external-plugins-bootstrap.js";
 import { runShutdownHooks } from "../daemon/shutdown-registry.js";
 import {
   registerPlugin,
@@ -69,12 +65,6 @@ const TEST_WORKSPACE_DIR = join(
 );
 process.env.VELLUM_WORKSPACE_DIR = TEST_WORKSPACE_DIR;
 
-const fakeConfig = {} as unknown as AssistantConfig;
-const fakeCtx: DaemonContext = {
-  config: fakeConfig,
-  assistantVersion: "9.9.9-test",
-};
-
 /** Build a minimal valid plugin with optional route contributions. */
 /**
  * Test helper. Accepts the new `hooks` bag and ALSO legacy top-level
@@ -149,7 +139,7 @@ describe("plugin route contributions", () => {
       }),
     );
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     // init() must have run — route registration is gated on init success.
     expect(initFired).toBe(true);
@@ -179,7 +169,7 @@ describe("plugin route contributions", () => {
 
     registerPlugin(buildPlugin("echo-plugin", { routes: [route] }));
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     // Sanity: route is live after bootstrap.
     expect(matchSkillRoute("/_plugin/echo", "GET")).not.toBeNull();
@@ -197,7 +187,7 @@ describe("plugin route contributions", () => {
     // route handling entirely (the guard is `if plugin.routes && length > 0`).
     registerPlugin(buildPlugin("no-routes-plugin", { async init() {} }));
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
     await runShutdownHooks("test-shutdown");
 
     // Nothing to verify beyond "neither throws" — an empty `routes` must not
@@ -228,7 +218,7 @@ describe("plugin route contributions", () => {
       }),
     );
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     // Simulate an external wipe before the shutdown hook runs — e.g. a
     // different subsystem calling `resetSkillRoutesForTests` or a hot-reload
@@ -301,7 +291,7 @@ describe("plugin route contributions", () => {
       }),
     );
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     // Both plugins' routes landed in the registry; matching returns one of
     // them (order defined by registration, but we only care that *some*

package/src/__tests__/plugin-skill-contribution.test.ts

@@ -31,12 +31,8 @@ mock.module("../security/secure-keys.js", () => ({
   getSecureKeyAsync: getSecureKeyAsyncMock,
 }));
 
-import type { AssistantConfig } from "../config/schema.js";
 import { loadSkillBySelector, loadSkillCatalog } from "../config/skills.js";
-import {
-  bootstrapPlugins,
-  type DaemonContext,
-} from "../daemon/external-plugins-bootstrap.js";
+import { bootstrapPlugins } from "../daemon/external-plugins-bootstrap.js";
 import { runShutdownHooks } from "../daemon/shutdown-registry.js";
 import {
   getPluginContributedSkillDefinition,
@@ -64,12 +60,6 @@ const TEST_WORKSPACE_DIR = join(
 );
 process.env.VELLUM_WORKSPACE_DIR = TEST_WORKSPACE_DIR;
 
-const fakeConfig = {} as unknown as AssistantConfig;
-const fakeCtx: DaemonContext = {
-  config: fakeConfig,
-  assistantVersion: "9.9.9-test",
-};
-
 /** Build a plugin that contributes one or more skills. */
 function buildSkillPlugin(
   name: string,
@@ -106,7 +96,7 @@ describe("plugin skill contributions", () => {
     };
 
     registerPlugin(buildSkillPlugin("demo-plugin", [skill]));
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     // Ref count bumped to exactly 1 so we can tell register and unregister
     // are balanced downstream.
@@ -138,7 +128,7 @@ describe("plugin skill contributions", () => {
     };
 
     registerPlugin(buildSkillPlugin("catalog-plugin", [skill]));
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     // loadSkillCatalog is the exact entry point `skill_load` consults via
     // `loadSkillBySelector` -> `resolveSkillSelector`.
@@ -171,7 +161,7 @@ describe("plugin skill contributions", () => {
     };
 
     registerPlugin(buildSkillPlugin("ephemeral-plugin", [skill]));
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     // Sanity: present before shutdown.
     expect(loadSkillCatalog().some((s) => s.id === "ephemeral-skill")).toBe(
@@ -206,7 +196,7 @@ describe("plugin skill contributions", () => {
       },
     });
 
-    await bootstrapPlugins(fakeCtx);
+    await bootstrapPlugins();
 
     expect(getPluginSkillRefCount("no-skills-plugin")).toBe(0);
     expect(getPluginContributedSkillSummaries()).toEqual([]);
@@ -231,7 +221,7 @@ describe("plugin skill contributions", () => {
 
     let caught: unknown;
     try {
-      await bootstrapPlugins(fakeCtx);
+      await bootstrapPlugins();
     } catch (err) {
       caught = err;
     }
@@ -347,7 +337,7 @@ describe("plugin skill contributions", () => {
     registerPlugin(buildSkillPlugin("shadow-plugin", [pluginSkill]));
 
     try {
-      await bootstrapPlugins(fakeCtx);
+      await bootstrapPlugins();
 
       const catalog = loadSkillCatalog();
       const entry = catalog.find((s) => s.id === "shared-id");