@vellumai/assistant 0.4.41 → 0.4.43

package/src/__tests__/session-usage.test.ts

@@ -0,0 +1,180 @@
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+
+const testDir = mkdtempSync(join(tmpdir(), "session-usage-test-"));
+
+const updateConversationUsageCalls: Array<{
+  conversationId: string;
+  inputTokens: number;
+  outputTokens: number;
+  estimatedCost: number;
+}> = [];
+
+mock.module("../util/platform.js", () => ({
+  getDataDir: () => testDir,
+  isMacOS: () => process.platform === "darwin",
+  isLinux: () => process.platform === "linux",
+  isWindows: () => process.platform === "win32",
+  getSocketPath: () => join(testDir, "test.sock"),
+  getPidPath: () => join(testDir, "test.pid"),
+  getDbPath: () => join(testDir, "test.db"),
+  getLogPath: () => join(testDir, "test.log"),
+  ensureDataDir: () => {},
+}));
+
+mock.module("../util/logger.js", () => ({
+  getLogger: () =>
+    new Proxy({} as Record<string, unknown>, {
+      get: () => () => {},
+    }),
+}));
+
+mock.module("../config/loader.js", () => ({
+  getConfig: () => ({
+    pricingOverrides: [],
+  }),
+}));
+
+mock.module("../memory/conversation-crud.js", () => ({
+  updateConversationUsage: (
+    conversationId: string,
+    inputTokens: number,
+    outputTokens: number,
+    estimatedCost: number,
+  ) => {
+    updateConversationUsageCalls.push({
+      conversationId,
+      inputTokens,
+      outputTokens,
+      estimatedCost,
+    });
+  },
+}));
+
+import { recordUsage } from "../daemon/session-usage.js";
+import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { listUsageEvents } from "../memory/llm-usage-store.js";
+import type { PricingUsage } from "../usage/types.js";
+import { resolvePricingForUsageWithOverrides } from "../util/pricing.js";
+
+initializeDb();
+
+afterAll(() => {
+  resetDb();
+  try {
+    rmSync(testDir, { recursive: true });
+  } catch {
+    /* best effort */
+  }
+});
+
+describe("recordUsage", () => {
+  beforeEach(() => {
+    const db = getDb();
+    db.run(`DELETE FROM llm_usage_events`);
+    updateConversationUsageCalls.length = 0;
+  });
+
+  test("stores direct input separately from Anthropic cache usage while keeping live totals combined", () => {
+    const usageStats = {
+      inputTokens: 0,
+      outputTokens: 0,
+      estimatedCost: 0,
+    };
+    const onEventMessages: unknown[] = [];
+
+    const rawResponses = [
+      {
+        usage: {
+          cache_creation: {
+            ephemeral_5m_input_tokens: 173_619,
+          },
+        },
+      },
+      {
+        usage: {
+          cache_creation: {
+            ephemeral_1h_input_tokens: 200_000,
+          },
+        },
+      },
+    ];
+
+    recordUsage(
+      {
+        conversationId: "conv-usage-1",
+        providerName: "anthropic",
+        usageStats,
+      },
+      3_420_218,
+      11_768,
+      "claude-opus-4-6",
+      (msg) => onEventMessages.push(msg),
+      "main_agent",
+      "req-usage-1",
+      373_619,
+      3_046_461,
+      rawResponses,
+    );
+
+    const events = listUsageEvents();
+    expect(events).toHaveLength(1);
+
+    const expectedUsage: PricingUsage = {
+      directInputTokens: 138,
+      outputTokens: 11_768,
+      cacheCreationInputTokens: 373_619,
+      cacheReadInputTokens: 3_046_461,
+      anthropicCacheCreation: {
+        ephemeral_5m_input_tokens: 173_619,
+        ephemeral_1h_input_tokens: 200_000,
+      },
+    };
+    const expectedPricing = resolvePricingForUsageWithOverrides(
+      "anthropic",
+      "claude-opus-4-6",
+      expectedUsage,
+      [],
+    );
+
+    expect(events[0].conversationId).toBe("conv-usage-1");
+    expect(events[0].requestId).toBe("req-usage-1");
+    expect(events[0].inputTokens).toBe(138);
+    expect(events[0].outputTokens).toBe(11_768);
+    expect(events[0].cacheCreationInputTokens).toBe(373_619);
+    expect(events[0].cacheReadInputTokens).toBe(3_046_461);
+    expect(events[0].pricingStatus).toBe("priced");
+    expect(events[0].estimatedCostUsd).toBe(
+      expectedPricing.estimatedCostUsd ?? null,
+    );
+
+    expect(usageStats.inputTokens).toBe(3_420_218);
+    expect(usageStats.outputTokens).toBe(11_768);
+    expect(usageStats.estimatedCost).toBe(
+      expectedPricing.estimatedCostUsd ?? 0,
+    );
+
+    expect(updateConversationUsageCalls).toEqual([
+      {
+        conversationId: "conv-usage-1",
+        inputTokens: 3_420_218,
+        outputTokens: 11_768,
+        estimatedCost: expectedPricing.estimatedCostUsd ?? 0,
+      },
+    ]);
+
+    expect(onEventMessages).toEqual([
+      {
+        type: "usage_update",
+        inputTokens: 3_420_218,
+        outputTokens: 11_768,
+        totalInputTokens: 3_420_218,
+        totalOutputTokens: 11_768,
+        estimatedCost: expectedPricing.estimatedCostUsd ?? 0,
+        model: "claude-opus-4-6",
+      },
+    ]);
+  });
+});

package/src/__tests__/session-workspace-cache-state.test.ts

@@ -47,7 +47,7 @@ mock.module("../config/loader.js", () => ({
   invalidateConfigCache: () => {},
 }));
 
-mock.module("../config/system-prompt.js", () => ({
+mock.module("../prompts/system-prompt.js", () => ({
   buildSystemPrompt: () => "system prompt",
 }));
 
@@ -78,7 +78,7 @@ mock.module("../security/secret-allowlist.js", () => ({
   resetAllowlist: () => {},
 }));
 
-mock.module("../memory/conversation-store.js", () => ({
+mock.module("../memory/conversation-crud.js", () => ({
   getConversationThreadType: () => "default",
   setConversationOriginChannelIfUnset: () => {},
   provenanceFromTrustContext: () => ({
@@ -103,6 +103,9 @@ mock.module("../memory/conversation-store.js", () => ({
   updateConversationContextWindow: () => {},
   deleteMessageById: () => ({ segmentIds: [], orphanedItemIds: [] }),
   deleteLastExchange: () => 0,
+}));
+
+mock.module("../memory/conversation-queries.js", () => ({
   isLastUserMessageToolResult: () => false,
 }));
 
@@ -129,6 +132,9 @@ mock.module("../memory/retrieval-budget.js", () => ({
 mock.module("../context/window-manager.js", () => ({
   ContextWindowManager: class {
     constructor() {}
+    shouldCompact() {
+      return { needed: false, estimatedTokens: 0 };
+    }
     async maybeCompact() {
       return { compacted: false };
     }

package/src/__tests__/session-workspace-injection.test.ts

@@ -75,7 +75,7 @@ mock.module("../config/loader.js", () => ({
   invalidateConfigCache: () => {},
 }));
 
-mock.module("../config/system-prompt.js", () => ({
+mock.module("../prompts/system-prompt.js", () => ({
   buildSystemPrompt: () => "system prompt",
 }));
 mock.module("../config/skills.js", () => ({
@@ -101,7 +101,7 @@ mock.module("../security/secret-allowlist.js", () => ({
   resetAllowlist: () => {},
 }));
 
-mock.module("../memory/conversation-store.js", () => ({
+mock.module("../memory/conversation-crud.js", () => ({
   getConversationThreadType: () => "default",
   setConversationOriginChannelIfUnset: () => {},
   provenanceFromTrustContext: () => ({
@@ -126,6 +126,9 @@ mock.module("../memory/conversation-store.js", () => ({
   updateConversationContextWindow: () => {},
   deleteMessageById: () => ({ segmentIds: [], orphanedItemIds: [] }),
   deleteLastExchange: () => 0,
+}));
+
+mock.module("../memory/conversation-queries.js", () => ({
   isLastUserMessageToolResult: () => false,
 }));
 
@@ -169,6 +172,9 @@ mock.module("../memory/retrieval-budget.js", () => ({
 }));
 mock.module("../context/window-manager.js", () => ({
   ContextWindowManager: class {
+    shouldCompact() {
+      return { needed: false, estimatedTokens: 0 };
+    }
     async maybeCompact() {
       return { compacted: false };
     }

package/src/__tests__/session-workspace-tool-tracking.test.ts

@@ -73,7 +73,7 @@ mock.module("../config/loader.js", () => ({
   invalidateConfigCache: () => {},
 }));
 
-mock.module("../config/system-prompt.js", () => ({
+mock.module("../prompts/system-prompt.js", () => ({
   buildSystemPrompt: () => "system prompt",
 }));
 mock.module("../config/skills.js", () => ({
@@ -99,7 +99,7 @@ mock.module("../security/secret-allowlist.js", () => ({
   resetAllowlist: () => {},
 }));
 
-mock.module("../memory/conversation-store.js", () => ({
+mock.module("../memory/conversation-crud.js", () => ({
   getConversationThreadType: () => "default",
   setConversationOriginChannelIfUnset: () => {},
   provenanceFromTrustContext: () => ({
@@ -124,6 +124,9 @@ mock.module("../memory/conversation-store.js", () => ({
   updateConversationContextWindow: () => {},
   deleteMessageById: () => ({ segmentIds: [], orphanedItemIds: [] }),
   deleteLastExchange: () => 0,
+}));
+
+mock.module("../memory/conversation-queries.js", () => ({
   isLastUserMessageToolResult: () => false,
 }));
 
@@ -167,6 +170,9 @@ mock.module("../memory/retrieval-budget.js", () => ({
 }));
 mock.module("../context/window-manager.js", () => ({
   ContextWindowManager: class {
+    shouldCompact() {
+      return { needed: false, estimatedTokens: 0 };
+    }
     async maybeCompact() {
       return { compacted: false };
     }

package/src/__tests__/skill-feature-flags-integration.test.ts

@@ -24,7 +24,6 @@ const TEST_DIR = join(
 
 let currentConfig: Record<string, unknown> = {
   sandbox: { enabled: false, backend: "native" },
-  featureFlags: {},
 };
 
 const DECLARED_SKILL_ID = "hatch-new-assistant";
@@ -48,7 +47,7 @@ mock.module("../util/platform.js", () => ({
   getLogPath: () => join(TEST_DIR, "logs", "vellum.log"),
   getHistoryPath: () => join(TEST_DIR, "history"),
   getHooksDir: () => join(TEST_DIR, "hooks"),
-  getIpcBlobDir: () => join(TEST_DIR, "ipc-blobs"),
+
   getSandboxRootDir: () => join(TEST_DIR, "sandbox"),
   getSandboxWorkingDir: () => TEST_DIR,
   getInterfacesDir: () => join(TEST_DIR, "interfaces"),
@@ -58,10 +57,6 @@ mock.module("../util/platform.js", () => ({
   getPlatformName: () => "linux",
   getClipboardCommand: () => null,
   readSessionToken: () => null,
-  removeSocketFile: () => {},
-  migratePath: () => {},
-  migrateToWorkspaceLayout: () => {},
-  migrateToDataLayout: () => {},
 }));
 
 const noopLogger = new Proxy({} as Record<string, unknown>, {
@@ -93,8 +88,8 @@ mock.module("../config/loader.js", () => ({
 }));
 
 // eslint-disable-next-line @typescript-eslint/no-require-imports
-const realUserReference = require("../config/user-reference.js");
-mock.module("../config/user-reference.js", () => ({
+const realUserReference = require("../prompts/user-reference.js");
+mock.module("../prompts/user-reference.js", () => ({
   ...realUserReference,
   resolveUserReference: () => "TestUser",
   resolveUserPronouns: () => null,
@@ -107,7 +102,7 @@ mock.module("../tools/credentials/metadata-store.js", () => ({
   listCredentialMetadata: () => [],
 }));
 
-const { buildSystemPrompt } = await import("../config/system-prompt.js");
+const { buildSystemPrompt } = await import("../prompts/system-prompt.js");
 
 // ---------------------------------------------------------------------------
 // Setup / Teardown
@@ -118,7 +113,6 @@ beforeEach(() => {
   // Reset config to defaults before each test
   currentConfig = {
     sandbox: { enabled: false, backend: "native" },
-    featureFlags: {},
   };
 });
 
@@ -179,7 +173,7 @@ describe("buildSystemPrompt feature flag filtering", () => {
     expect(result).not.toContain(`id="${DECLARED_SKILL_ID}"`);
   });
 
-  test("declared skills hidden when featureFlags is empty (registry defaults to false)", () => {
+  test("declared skills hidden when no overrides set (registry defaults to false)", () => {
     createSkillOnDisk(
       DECLARED_SKILL_ID,
       "Hatch New Assistant",

package/src/__tests__/skill-feature-flags.test.ts

@@ -41,6 +41,7 @@ function makeSkill(
   return {
     id,
     name: `${id} skill`,
+    displayName: `${id} skill`,
     description: `Description for ${id}`,
     directoryPath: `/fake/skills/${id}`,
     skillFilePath: `/fake/skills/${id}/SKILL.md`,

package/src/__tests__/skill-include-graph.test.ts

@@ -12,6 +12,7 @@ function makeSkill(id: string, includes?: string[]): SkillSummary {
   return {
     id,
     name: id,
+    displayName: id,
     description: `Skill ${id}`,
     directoryPath: `/skills/${id}`,
     skillFilePath: `/skills/${id}/SKILL.md`,

package/src/__tests__/skill-load-feature-flag.test.ts

@@ -12,9 +12,7 @@ const TEST_DIR = join(
   `vellum-skill-load-flag-test-${crypto.randomUUID()}`,
 );
 
-let currentConfig: Record<string, unknown> = {
-  featureFlags: {},
-};
+let currentConfig: Record<string, unknown> = {};
 
 const DECLARED_SKILL_ID = "hatch-new-assistant";
 const DECLARED_FLAG_KEY = "feature_flags.hatch-new-assistant.enabled";
@@ -35,7 +33,7 @@ const platformOverrides: Record<string, (...args: unknown[]) => unknown> = {
     join(TEST_DIR, "workspace", String(f)),
   getInterfacesDir: () => join(TEST_DIR, "interfaces"),
   getHooksDir: () => join(TEST_DIR, "hooks"),
-  getIpcBlobDir: () => join(TEST_DIR, "blobs"),
+
   getSandboxRootDir: () => join(TEST_DIR, "sandbox"),
   getSandboxWorkingDir: () => join(TEST_DIR, "sandbox", "work"),
   getHistoryPath: () => join(TEST_DIR, "history"),
@@ -46,10 +44,6 @@ const platformOverrides: Record<string, (...args: unknown[]) => unknown> = {
   isLinux: () => process.platform === "linux",
   isWindows: () => process.platform === "win32",
   getPlatformName: () => process.platform,
-  migratePath: () => {},
-  migrateToWorkspaceLayout: () => {},
-  migrateToDataLayout: () => {},
-  removeSocketFile: () => {},
 };
 // eslint-disable-next-line @typescript-eslint/no-require-imports
 const realPlatform = require("../util/platform.js");
@@ -121,7 +115,7 @@ async function executeSkillLoad(
 describe("skill_load feature flag enforcement", () => {
   beforeEach(() => {
     mkdirSync(join(TEST_DIR, "skills"), { recursive: true });
-    currentConfig = { featureFlags: {} };
+    currentConfig = {};
   });
 
   afterEach(() => {

package/src/__tests__/skill-load-tool.test.ts

@@ -26,21 +26,27 @@ const platformOverrides: Record<string, (...args: unknown[]) => unknown> = {
     join(TEST_DIR, "workspace", String(f)),
   getInterfacesDir: () => join(TEST_DIR, "interfaces"),
   getHooksDir: () => join(TEST_DIR, "hooks"),
-  getIpcBlobDir: () => join(TEST_DIR, "blobs"),
+
   getSandboxRootDir: () => join(TEST_DIR, "sandbox"),
   getSandboxWorkingDir: () => join(TEST_DIR, "sandbox", "work"),
   getHistoryPath: () => join(TEST_DIR, "history"),
   getSessionTokenPath: () => join(TEST_DIR, "session-token"),
   readSessionToken: () => null,
   getClipboardCommand: () => null,
+  readLockfile: () => null,
+  normalizeAssistantId: (id: unknown) => String(id),
+  writeLockfile: () => {},
+  getEmbeddingModelsDir: () => join(TEST_DIR, "embedding-models"),
+  getTCPPort: () => 8765,
+  isTCPEnabled: () => false,
+  getTCPHost: () => "127.0.0.1",
+  isIOSPairingEnabled: () => false,
+  getPlatformTokenPath: () => join(TEST_DIR, "platform-token"),
+  readPlatformToken: () => null,
   isMacOS: () => process.platform === "darwin",
   isLinux: () => process.platform === "linux",
   isWindows: () => process.platform === "win32",
   getPlatformName: () => process.platform,
-  migratePath: () => {},
-  migrateToWorkspaceLayout: () => {},
-  migrateToDataLayout: () => {},
-  removeSocketFile: () => {},
 };
 mock.module("../util/platform.js", () => platformOverrides);
 
@@ -49,6 +55,8 @@ mock.module("../util/logger.js", () => ({
     new Proxy({} as Record<string, unknown>, {
       get: () => () => {},
     }),
+  isDebug: () => false,
+  truncateForLog: (s: unknown) => String(s),
 }));
 
 await import("../tools/skills/load.js");
@@ -79,8 +87,8 @@ function writeSkillWithIncludes(
   mkdirSync(skillDir, { recursive: true });
   writeFileSync(
     join(skillDir, "SKILL.md"),
-    `---\nname: "${name}"\ndescription: "${description}"\nincludes: ${JSON.stringify(
-      includes,
+    `---\nname: "${name}"\ndescription: "${description}"\nmetadata: ${JSON.stringify(
+      { vellum: { includes } },
     )}\n---\n\n${body}\n`,
   );
 }
@@ -325,7 +333,7 @@ describe("skill_load tool", () => {
     mkdirSync(skillDir, { recursive: true });
     writeFileSync(
       join(skillDir, "SKILL.md"),
-      '---\nname: "Marker Missing"\ndescription: "test"\nincludes: ["nonexistent"]\n---\n\nBody.\n',
+      '---\nname: "Marker Missing"\ndescription: "test"\nmetadata: {"vellum":{"includes":["nonexistent"]}}\n---\n\nBody.\n',
     );
     writeFileSync(join(TEST_DIR, "skills", "SKILLS.md"), "- marker-missing\n");
 
@@ -342,11 +350,11 @@ describe("skill_load tool", () => {
     mkdirSync(dirB, { recursive: true });
     writeFileSync(
       join(dirA, "SKILL.md"),
-      '---\nname: "Cycle A"\ndescription: "test"\nincludes: ["cycle-b"]\n---\n\nBody A.\n',
+      '---\nname: "Cycle A"\ndescription: "test"\nmetadata: {"vellum":{"includes":["cycle-b"]}}\n---\n\nBody A.\n',
     );
     writeFileSync(
       join(dirB, "SKILL.md"),
-      '---\nname: "Cycle B"\ndescription: "test"\nincludes: ["cycle-a"]\n---\n\nBody B.\n',
+      '---\nname: "Cycle B"\ndescription: "test"\nmetadata: {"vellum":{"includes":["cycle-a"]}}\n---\n\nBody B.\n',
     );
     writeFileSync(
       join(TEST_DIR, "skills", "SKILLS.md"),
@@ -374,7 +382,7 @@ describe("skill_load tool", () => {
     mkdirSync(parentDir, { recursive: true });
     writeFileSync(
       join(parentDir, "SKILL.md"),
-      '---\nname: "Parent"\ndescription: "Has children"\nincludes: ["child-skill"]\n---\n\nParent body.\n',
+      '---\nname: "Parent"\ndescription: "Has children"\nmetadata: {"vellum":{"includes":["child-skill"]}}\n---\n\nParent body.\n',
     );
     writeFileSync(
       join(TEST_DIR, "skills", "SKILLS.md"),
@@ -561,6 +569,76 @@ describe("skill_load tool", () => {
     expect(result.content).not.toContain("<loaded_skill");
   });
 
+  test("skill with references/ directory appends reference file contents", async () => {
+    // Create a skill with a references/ subdirectory
+    const skillDir = join(TEST_DIR, "skills", "with-refs");
+    mkdirSync(skillDir, { recursive: true });
+    mkdirSync(join(skillDir, "references"), { recursive: true });
+    writeFileSync(
+      join(skillDir, "SKILL.md"),
+      '---\nname: "With Refs"\ndescription: "Has references"\n---\n\nMain body.\n',
+    );
+    writeFileSync(
+      join(skillDir, "references", "GUIDE.md"),
+      "# Guide\n\nDetailed guide content.",
+    );
+    writeFileSync(
+      join(skillDir, "references", "TROUBLESHOOTING.md"),
+      "# Troubleshooting\n\nFix things here.",
+    );
+    writeFileSync(join(TEST_DIR, "skills", "SKILLS.md"), "- with-refs\n");
+
+    const result = await executeSkillLoad({ skill: "with-refs" });
+    expect(result.isError).toBe(false);
+    expect(result.content).toContain("Main body.");
+    // Reference files should be appended with headers
+    expect(result.content).toContain("--- Reference: Guide ---");
+    expect(result.content).toContain("Detailed guide content.");
+    expect(result.content).toContain("--- Reference: Troubleshooting ---");
+    expect(result.content).toContain("Fix things here.");
+    // References must be appended in alphabetical order (Guide before Troubleshooting)
+    const guideIdx = result.content.indexOf("--- Reference: Guide ---");
+    const troubleshootIdx = result.content.indexOf(
+      "--- Reference: Troubleshooting ---",
+    );
+    expect(guideIdx).toBeLessThan(troubleshootIdx);
+  });
+
+  test("skill without references/ directory loads normally", async () => {
+    writeSkill("no-refs", "No Refs", "No references dir", "Just body.");
+    writeFileSync(join(TEST_DIR, "skills", "SKILLS.md"), "- no-refs\n");
+
+    const result = await executeSkillLoad({ skill: "no-refs" });
+    expect(result.isError).toBe(false);
+    expect(result.content).toContain("Just body.");
+    expect(result.content).not.toContain("--- Reference:");
+  });
+
+  test("references/ directory ignores non-markdown files", async () => {
+    const skillDir = join(TEST_DIR, "skills", "refs-filter");
+    mkdirSync(skillDir, { recursive: true });
+    mkdirSync(join(skillDir, "references"), { recursive: true });
+    writeFileSync(
+      join(skillDir, "SKILL.md"),
+      '---\nname: "Refs Filter"\ndescription: "Filters non-md"\n---\n\nBody.\n',
+    );
+    writeFileSync(
+      join(skillDir, "references", "GUIDE.md"),
+      "# Guide\n\nGuide content.",
+    );
+    writeFileSync(
+      join(skillDir, "references", "data.json"),
+      '{"key": "value"}',
+    );
+    writeFileSync(join(TEST_DIR, "skills", "SKILLS.md"), "- refs-filter\n");
+
+    const result = await executeSkillLoad({ skill: "refs-filter" });
+    expect(result.isError).toBe(false);
+    expect(result.content).toContain("--- Reference: Guide ---");
+    expect(result.content).not.toContain("data.json");
+    expect(result.content).not.toContain('"key"');
+  });
+
   test('skill with empty includes array loads successfully as "none"', async () => {
     // Write a skill with `includes: []` directly in frontmatter.
     // The parser normalizes this to undefined, so it should behave identically
@@ -569,7 +647,7 @@ describe("skill_load tool", () => {
     mkdirSync(skillDir, { recursive: true });
     writeFileSync(
       join(skillDir, "SKILL.md"),
-      '---\nname: "Empty Includes"\ndescription: "Has empty array"\nincludes: []\n---\n\nBody.\n',
+      '---\nname: "Empty Includes"\ndescription: "Has empty array"\nmetadata: {"vellum":{"includes":[]}}\n---\n\nBody.\n',
     );
     writeFileSync(join(TEST_DIR, "skills", "SKILLS.md"), "- empty-includes\n");
 

package/src/__tests__/skill-projection-feature-flag.test.ts

@@ -20,9 +20,9 @@ let mockRegisteredTools: Map<string, Tool[]> = new Map();
 let mockUnregisteredSkillIds: string[] = [];
 let mockSkillRefCount: Map<string, number> = new Map();
 
-let currentConfig: Record<string, unknown> = { featureFlags: {} };
+let currentConfig: Record<string, unknown> = {};
 const DECLARED_SKILL_ID = "hatch-new-assistant";
-const DECLARED_LEGACY_KEY = "skills.hatch-new-assistant.enabled";
+const DECLARED_FLAG_KEY = "feature_flags.hatch-new-assistant.enabled";
 
 // ---------------------------------------------------------------------------
 // Mocks
@@ -50,10 +50,6 @@ mock.module("../config/assistant-feature-flags.js", () => ({
       }
     ).assistantFeatureFlagValues;
     if (vals && typeof vals[key] === "boolean") return vals[key];
-    // Check legacy featureFlags too
-    const legacy = (config as { featureFlags?: Record<string, boolean> })
-      .featureFlags;
-    if (legacy && typeof legacy[key] === "boolean") return legacy[key];
     return true; // default enabled
   },
   loadDefaultsRegistry: () => ({}),
@@ -62,7 +58,7 @@ mock.module("../config/assistant-feature-flags.js", () => ({
 }));
 
 mock.module("../config/skill-state.js", () => ({
-  skillFlagKey: (skillId: string) => `skills.${skillId}.enabled`,
+  skillFlagKey: (skillId: string) => `feature_flags.${skillId}.enabled`,
 }));
 
 mock.module("../skills/active-skill-tools.js", () => {
@@ -99,8 +95,6 @@ mock.module("../skills/active-skill-tools.js", () => {
 
   return {
     deriveActiveSkills: (messages: Message[]) => parseMarkers(messages),
-    deriveActiveSkillIds: (messages: Message[]) =>
-      parseMarkers(messages).map((e) => e.id),
   };
 });
 
@@ -231,6 +225,7 @@ function makeSkill(id: string): SkillSummary {
   return {
     id,
     name: id,
+    displayName: id,
     description: `Skill ${id}`,
     directoryPath: `/skills/${id}`,
     skillFilePath: `/skills/${id}/SKILL.md`,
@@ -293,7 +288,7 @@ describe("projectSkillTools feature flag enforcement", () => {
     mockRegisteredTools = new Map();
     mockUnregisteredSkillIds = [];
     mockSkillRefCount = new Map();
-    currentConfig = { featureFlags: {} };
+    currentConfig = {};
     resetSkillToolProjection();
   });
 
@@ -308,7 +303,9 @@ describe("projectSkillTools feature flag enforcement", () => {
     const prevActive = new Map<string, string>();
 
     // Feature flag is OFF
-    currentConfig = { featureFlags: { [DECLARED_LEGACY_KEY]: false } };
+    currentConfig = {
+      assistantFeatureFlagValues: { [DECLARED_FLAG_KEY]: false },
+    };
 
     const result = projectSkillTools(history, {
       previouslyActiveSkillIds: prevActive,
@@ -329,7 +326,9 @@ describe("projectSkillTools feature flag enforcement", () => {
     const prevActive = new Map<string, string>();
 
     // Feature flag is ON
-    currentConfig = { featureFlags: { [DECLARED_LEGACY_KEY]: true } };
+    currentConfig = {
+      assistantFeatureFlagValues: { [DECLARED_FLAG_KEY]: true },
+    };
 
     const result = projectSkillTools(history, {
       previouslyActiveSkillIds: prevActive,
@@ -347,8 +346,8 @@ describe("projectSkillTools feature flag enforcement", () => {
     const history = buildHistoryWithMarker(DECLARED_SKILL_ID);
     const prevActive = new Map<string, string>();
 
-    // featureFlags is empty — should default to enabled
-    currentConfig = { featureFlags: {} };
+    // No overrides — should default to enabled
+    currentConfig = {};
 
     const result = projectSkillTools(history, {
       previouslyActiveSkillIds: prevActive,
@@ -414,7 +413,7 @@ describe("projectSkillTools feature flag enforcement", () => {
 
     // Declared skill is OFF, twitter is undeclared with no persisted override so remains ON.
     currentConfig = {
-      featureFlags: { [DECLARED_LEGACY_KEY]: false },
+      assistantFeatureFlagValues: { [DECLARED_FLAG_KEY]: false },
     };
 
     const result = projectSkillTools(history, {