@gethmy/mcp 2.4.7 → 2.5.0

package/src/__tests__/memory-audit.test.ts

@@ -1,528 +0,0 @@
-/**
- * Unit tests for memory quality audit.
- *
- * Run with: bun test packages/mcp-server/src/__tests__/memory-audit.test.ts
- */
-
-import { describe, expect, mock, test } from "bun:test";
-import { runMemoryAudit } from "../memory-audit.js";
-
-function daysAgo(days: number): string {
-  return new Date(Date.now() - days * 24 * 60 * 60 * 1000).toISOString();
-}
-
-function makeMockClient(
-  entities: unknown[],
-  relations?: Record<string, number>,
-) {
-  const deletedIds: string[] = [];
-  const updatedEntities: Array<{
-    id: string;
-    updates: Record<string, unknown>;
-  }> = [];
-
-  return {
-    client: {
-      listMemoryEntities: mock(async (opts: { offset?: number }) => {
-        const offset = opts.offset ?? 0;
-        if (offset > 0) return { entities: [], count: 0 };
-        return { entities, count: entities.length };
-      }),
-      getRelatedEntities: mock(async (id: string) => {
-        const n = relations?.[id] ?? 0;
-        return {
-          outgoing: Array(n).fill({}),
-          incoming: [],
-        };
-      }),
-      deleteMemoryEntity: mock(async (id: string) => {
-        deletedIds.push(id);
-        return { success: true };
-      }),
-      updateMemoryEntity: mock(
-        async (id: string, updates: Record<string, unknown>) => {
-          updatedEntities.push({ id, updates });
-          return { entity: { id, ...updates } };
-        },
-      ),
-    } as any,
-    deletedIds,
-    updatedEntities,
-  };
-}
-
-describe("runMemoryAudit", () => {
-  test("buckets a healthy modern entity into 'keep'", async () => {
-    const { client } = makeMockClient(
-      [
-        {
-          id: "healthy",
-          type: "pattern",
-          title: "Auth refresh token rotation pattern",
-          content:
-            "Rotate refresh tokens on every use. Keep a rolling window of two valid tokens to allow retry on network failures.",
-          confidence: 0.95,
-          memory_tier: "reference",
-          access_count: 25,
-          last_accessed_at: daysAgo(1),
-          created_at: daysAgo(90),
-          tags: ["auth", "security"],
-          embedding: [0.1, 0.2, 0.3],
-          promoted_from_id: "orig-1",
-          metadata: {},
-        },
-      ],
-      { healthy: 3 },
-    );
-
-    const report = await runMemoryAudit(client, "ws-1");
-    expect(report.summary.keep).toBe(1);
-    expect(report.summary.delete).toBe(0);
-    expect(report.lowest[0].score).toBeGreaterThanOrEqual(70);
-  });
-
-  test("buckets a legacy default-confidence entity into archive/delete", async () => {
-    const { client } = makeMockClient([
-      {
-        id: "legacy",
-        type: "context",
-        title: "x",
-        content: "",
-        confidence: 1.0,
-        memory_tier: "draft",
-        access_count: 0,
-        last_accessed_at: null,
-        created_at: daysAgo(120),
-        tags: [],
-        embedding: null,
-        promoted_from_id: null,
-        metadata: {},
-      },
-    ]);
-
-    const report = await runMemoryAudit(client, "ws-1");
-    expect(report.summary.legacyCount).toBe(1);
-    const a = report.lowest[0];
-    expect(a.legacy).toBe(true);
-    expect(a.bucket === "archive" || a.bucket === "delete").toBe(true);
-    expect(a.legacyReasons.length).toBeGreaterThan(1);
-  });
-
-  test("dryRun does not delete, archive, or flag", async () => {
-    const { client, deletedIds, updatedEntities } = makeMockClient([
-      {
-        id: "bad",
-        type: "context",
-        title: "x",
-        content: "",
-        confidence: 0.1,
-        memory_tier: "draft",
-        access_count: 0,
-        last_accessed_at: null,
-        created_at: daysAgo(100),
-        tags: [],
-        embedding: null,
-      },
-    ]);
-
-    await runMemoryAudit(client, "ws-1", undefined, { dryRun: true });
-    expect(deletedIds).toHaveLength(0);
-    expect(updatedEntities).toHaveLength(0);
-  });
-
-  test("executes delete + archive + flag in non-dryRun", async () => {
-    const { client, deletedIds, updatedEntities } = makeMockClient([
-      // delete (very low)
-      {
-        id: "trash",
-        type: "context",
-        title: "x",
-        content: "",
-        confidence: 0.05,
-        memory_tier: "draft",
-        access_count: 0,
-        last_accessed_at: null,
-        created_at: daysAgo(200),
-        tags: [],
-      },
-      // archive (middling-low)
-      {
-        id: "archive-me",
-        type: "pattern",
-        title: "Partial pattern that lacks context here",
-        content: "Some content that is a bit more substantive than nothing.",
-        confidence: 0.3,
-        memory_tier: "draft",
-        access_count: 0,
-        last_accessed_at: daysAgo(40),
-        created_at: daysAgo(40),
-        tags: [],
-      },
-      // review (medium) — decent content but no tags, no relations, no embedding
-      {
-        id: "review-me",
-        type: "pattern",
-        title: "Reasonable pattern with decent content body here",
-        content:
-          "This entity has enough content to pass the length check. Confidence is moderate, access is limited.",
-        confidence: 0.5,
-        memory_tier: "episode",
-        access_count: 1,
-        last_accessed_at: daysAgo(25),
-        created_at: daysAgo(40),
-        tags: [],
-        embedding: null,
-      },
-    ]);
-
-    const report = await runMemoryAudit(client, "ws-1", undefined, {
-      dryRun: false,
-    });
-
-    expect(deletedIds).toContain("trash");
-    expect(
-      updatedEntities.find((u) => u.id === "archive-me")?.updates.confidence,
-    ).toBe(0.25);
-    expect(
-      (
-        updatedEntities.find((u) => u.id === "review-me")?.updates
-          .metadata as Record<string, unknown>
-      )?.needs_review,
-    ).toBe(true);
-
-    expect(report.actionsTaken.deleted).toBeGreaterThanOrEqual(1);
-    expect(report.actionsTaken.archived).toBeGreaterThanOrEqual(1);
-    expect(report.actionsTaken.flaggedReview).toBeGreaterThanOrEqual(1);
-  });
-
-  test("detects stuck-draft legacy signal", async () => {
-    const { client } = makeMockClient([
-      {
-        id: "stuck",
-        type: "context",
-        title: "Old draft that never made it",
-        content:
-          "Some content that is long enough to not count as thin content right here.",
-        confidence: 0.6,
-        memory_tier: "draft",
-        access_count: 1,
-        last_accessed_at: daysAgo(70),
-        created_at: daysAgo(75),
-        tags: ["x"],
-        embedding: [0.1],
-        promoted_from_id: null,
-      },
-    ]);
-
-    const report = await runMemoryAudit(client, "ws-1");
-    expect(report.legacyBreakdown.stuckDraft).toBe(1);
-    expect(report.lowest[0].reasons).toContain(
-      "stuck draft >60d never promoted",
-    );
-  });
-
-  test("distribution buckets sum to scanned count", async () => {
-    const entities = Array.from({ length: 10 }, (_, i) => ({
-      id: `e${i}`,
-      type: "context",
-      title: `Entity number ${i} with decent title length`,
-      content: "Some content string that is long enough to count properly.",
-      confidence: 0.1 * (i + 1),
-      memory_tier: i % 3 === 0 ? "reference" : "episode",
-      access_count: i,
-      last_accessed_at: daysAgo(i * 2),
-      created_at: daysAgo(i * 5 + 1),
-      tags: i % 2 === 0 ? ["tag"] : [],
-      embedding: i % 2 === 0 ? [0.1] : null,
-    }));
-    const { client } = makeMockClient(entities);
-
-    const report = await runMemoryAudit(client, "ws-1");
-    const total =
-      report.distribution["0-20"] +
-      report.distribution["20-40"] +
-      report.distribution["40-70"] +
-      report.distribution["70-100"];
-    expect(total).toBe(report.summary.scanned);
-    expect(report.summary.scanned).toBe(10);
-  });
-
-  test("empty workspace returns success with zeros", async () => {
-    const { client } = makeMockClient([]);
-    const report = await runMemoryAudit(client, "ws-1");
-    expect(report.success).toBe(true);
-    expect(report.summary.scanned).toBe(0);
-    expect(report.lowest).toHaveLength(0);
-  });
-
-  test("deleteBelow=0 disables deletion entirely", async () => {
-    const { client, deletedIds, updatedEntities } = makeMockClient([
-      {
-        id: "trash",
-        type: "context",
-        title: "x",
-        content: "",
-        confidence: 0.05,
-        memory_tier: "draft",
-        access_count: 0,
-        last_accessed_at: null,
-        created_at: daysAgo(200),
-        tags: [],
-      },
-    ]);
-
-    await runMemoryAudit(client, "ws-1", undefined, {
-      dryRun: false,
-      deleteBelow: 0,
-    });
-    expect(deletedIds).toHaveLength(0);
-    // Should land in archive bucket (score < 40 but >= 0)
-    expect(
-      updatedEntities.find((u) => u.id === "trash")?.updates.confidence,
-    ).toBe(0.25);
-  });
-
-  test("boilerplate override forces delete bucket regardless of confidence/access/tier", async () => {
-    // The exact failure mode that motivated this override: legacy task-transition
-    // entries promoted to reference tier with confidence=1.0 and high access_count
-    // were scoring ~80 and surviving in "keep". Verify the override demotes them.
-    const { client, deletedIds } = makeMockClient(
-      [
-        {
-          id: "promoted-junk",
-          type: "context",
-          title: "Task transition: legacy auto-extracted noise",
-          content:
-            "Agent transitioned tasks. Previous: doing X. Current: doing Y. Progress: 100%.",
-          confidence: 1.0,
-          memory_tier: "reference",
-          access_count: 91,
-          last_accessed_at: daysAgo(0),
-          created_at: daysAgo(29),
-          tags: ["auto-extracted", "task-transition", "mid-session"],
-          embedding: [0.1],
-          promoted_from_id: "orig-junk",
-        },
-      ],
-      { "promoted-junk": 2 },
-    );
-
-    const report = await runMemoryAudit(client, "ws-1", undefined, {
-      dryRun: false,
-    });
-    expect(report.summary.delete).toBe(1);
-    expect(deletedIds).toContain("promoted-junk");
-    expect(report.lowest[0].reasons).toContain("boilerplate title override");
-  });
-
-  test("legitimate titles starting with boilerplate-prefix words are NOT deleted", async () => {
-    // Regression test for the over-broad regex bug. Pre-fix patterns matched
-    // any title starting with "Placeholder", "Untitled", "Note", etc. After
-    // tightening, only exact boilerplate forms (with optional digit suffix
-    // or colon) match — real titles survive.
-    const { client, deletedIds } = makeMockClient(
-      [
-        {
-          id: "legit-placeholder",
-          type: "pattern",
-          title: "Placeholder pattern in React Suspense",
-          content:
-            "Use React.Suspense with a fallback component as the placeholder pattern for streaming SSR.",
-          confidence: 0.9,
-          memory_tier: "reference",
-          access_count: 12,
-          last_accessed_at: daysAgo(1),
-          created_at: daysAgo(60),
-          tags: ["react", "ssr"],
-          embedding: [0.1],
-        },
-        {
-          id: "legit-untitled",
-          type: "context",
-          title: "UntitledMaster.fig — design source for the homepage",
-          content:
-            "Reference Figma file containing master components for landing page assets.",
-          confidence: 0.85,
-          memory_tier: "reference",
-          access_count: 8,
-          last_accessed_at: daysAgo(2),
-          created_at: daysAgo(45),
-          tags: ["design"],
-          embedding: [0.1],
-        },
-        {
-          id: "legit-note",
-          type: "context",
-          title: "Note: schema migration order matters",
-          content: "Always run 0042 before 0043 because of FK dependency.",
-          confidence: 0.8,
-          memory_tier: "reference",
-          access_count: 5,
-          last_accessed_at: daysAgo(3),
-          created_at: daysAgo(30),
-          tags: ["db"],
-          embedding: [0.1],
-        },
-      ],
-      { "legit-placeholder": 3, "legit-untitled": 2, "legit-note": 1 },
-    );
-
-    const report = await runMemoryAudit(client, "ws-1", undefined, {
-      dryRun: false,
-    });
-    expect(deletedIds).toHaveLength(0);
-    expect(report.summary.delete).toBe(0);
-  });
-
-  test("empty-content draft with real title is NOT delete-bucketed", async () => {
-    // Users sometimes save a draft with title only and fill content later.
-    // The override is title-only, so empty content alone must not delete.
-    const { client, deletedIds } = makeMockClient([
-      {
-        id: "draft-empty-body",
-        type: "decision",
-        title: "Decision: skip Q3 launch",
-        content: "",
-        confidence: 0.7,
-        memory_tier: "draft",
-        access_count: 1,
-        last_accessed_at: daysAgo(1),
-        created_at: daysAgo(2),
-        tags: ["q3"],
-        embedding: null,
-      },
-    ]);
-
-    await runMemoryAudit(client, "ws-1", undefined, { dryRun: false });
-    expect(deletedIds).not.toContain("draft-empty-body");
-  });
-
-  test("boilerplate override respects deleteBelow=0 escape hatch", async () => {
-    // deleteBelow=0 is a "no deletions, audit-only" knob. Boilerplate must
-    // honor it — operators should be able to inspect findings without losing
-    // data on the same call.
-    const { client, deletedIds } = makeMockClient([
-      {
-        id: "boilerplate-protected",
-        type: "context",
-        title: "Task transition: would normally delete",
-        content: "noise content",
-        confidence: 1.0,
-        memory_tier: "reference",
-        access_count: 50,
-        last_accessed_at: daysAgo(0),
-        created_at: daysAgo(20),
-        tags: ["x"],
-        embedding: [0.1],
-      },
-    ]);
-
-    await runMemoryAudit(client, "ws-1", undefined, {
-      dryRun: false,
-      deleteBelow: 0,
-    });
-    expect(deletedIds).toHaveLength(0);
-  });
-
-  test("stale-draft filter flags draft+0access+age>threshold separately from bucket", async () => {
-    const { client } = makeMockClient(
-      [
-        // Stale draft — should be flagged by the filter, but otherwise healthy
-        // enough to bucket as "review" (not archive).
-        {
-          id: "stale-draft",
-          type: "context",
-          title:
-            "Task transition: feature work started but never touched again",
-          content:
-            "This draft has enough content and tags to score reasonably, " +
-            "but nobody ever accessed it after creation — classic promote-or-drop candidate.",
-          confidence: 0.4,
-          memory_tier: "draft",
-          access_count: 0,
-          last_accessed_at: null,
-          created_at: daysAgo(10),
-          tags: ["task"],
-          embedding: [0.1],
-        },
-        // Fresh draft — same shape but under the age threshold, must NOT flag.
-        {
-          id: "fresh-draft",
-          type: "context",
-          title: "Task transition: a fresh draft still within the grace window",
-          content:
-            "Content long enough to not be thin at all, really properly sized.",
-          confidence: 0.4,
-          memory_tier: "draft",
-          access_count: 0,
-          last_accessed_at: null,
-          created_at: daysAgo(3),
-          tags: ["task"],
-          embedding: [0.1],
-        },
-        // Non-draft old zero-access — must NOT flag (filter is draft-only).
-        {
-          id: "old-episode",
-          type: "pattern",
-          title: "Episode entity that is old and unaccessed but not a draft",
-          content:
-            "Sometimes reference/episode tier entities sit unaccessed; " +
-            "they're not draft-promotion candidates so the filter should skip them.",
-          confidence: 0.8,
-          memory_tier: "episode",
-          access_count: 0,
-          last_accessed_at: null,
-          created_at: daysAgo(30),
-          tags: ["pat"],
-          embedding: [0.1],
-        },
-      ],
-      { "stale-draft": 1, "fresh-draft": 1, "old-episode": 2 },
-    );
-
-    const report = await runMemoryAudit(client, "ws-1");
-    expect(report.summary.staleDraftCount).toBe(1);
-    expect(report.staleDrafts).toHaveLength(1);
-    expect(report.staleDrafts[0].id).toBe("stale-draft");
-    expect(report.healthReport).toContain("Stale Drafts");
-  });
-
-  test("staleDraftAgeDays option tunes the filter threshold", async () => {
-    const { client } = makeMockClient([
-      {
-        id: "draft-5d",
-        type: "context",
-        title: "Five-day-old untouched draft",
-        content:
-          "Content long enough to pass the thin-content check, absolutely.",
-        confidence: 0.4,
-        memory_tier: "draft",
-        access_count: 0,
-        last_accessed_at: null,
-        created_at: daysAgo(5),
-        tags: ["x"],
-        embedding: [0.1],
-      },
-    ]);
-
-    const defaultRun = await runMemoryAudit(client, "ws-1");
-    expect(defaultRun.summary.staleDraftCount).toBe(0);
-
-    const tightRun = await runMemoryAudit(client, "ws-1", undefined, {
-      staleDraftAgeDays: 3,
-    });
-    expect(tightRun.summary.staleDraftCount).toBe(1);
-  });
-
-  test("fetch error surfaces as report.success=false", async () => {
-    const client = {
-      listMemoryEntities: mock(async () => {
-        throw new Error("API down");
-      }),
-    } as any;
-
-    const report = await runMemoryAudit(client, "ws-1");
-    expect(report.success).toBe(false);
-    expect(report.errors.length).toBeGreaterThan(0);
-  });
-});