diff --git a/unit-testing/SKILL.md b/unit-testing/SKILL.md
new file mode 100644
index 0000000..c7d639a
--- /dev/null
+++ b/unit-testing/SKILL.md
@@ -0,0 +1,188 @@
+# Unit Testing (Osherove 3rd Edition)
+
+Write and review unit tests following Roy Osherove's "The Art of Unit Testing" (3rd Edition) principles, including entry points, exit points, and trust boundaries.
+
+## When to Use
+
+Activate this skill when:
+- Writing new unit tests
+- Reviewing existing test quality
+- Refactoring tests for readability/maintainability
+- Deciding what to test and how to structure tests
+- Identifying missing test coverage via exit points
+
+## Core Concepts
+
+### Entry Points
+The public API surface through which the unit under test is invoked:
+- Public methods/functions
+- Constructors
+- Event subscriptions
+- CLI arguments
+- IPC messages / HTTP endpoints
+
+### Exit Points (what to verify)
+Every unit has 1+ exit points. Each test should verify exactly **one exit point**:
+
+1. **Return value** — function returns a result
+2. **State change** — observable state mutation (property, file, DB)
+3. **Dependency call** — calls to a dependency (3rd party, network, I/O)
+
+### Trust Boundaries
+- **Don't mock what you own** — test real code paths unless I/O is involved
+- **Mock at the boundary** — mock only external dependencies (network, filesystem, time)
+- **Don't test internals** — test behavior through entry points, not implementation
+
+## Naming Convention
+
+```
+[UnitOfWork]_[Scenario]_[ExpectedBehavior]
+```
+
+Examples:
+```typescript
+describe("IpcServer", () => {
+  it("start_whenSocketAlreadyInUse_throwsAlreadyRunningError")
+  it("start_withStaleSocket_removesAndListens")
+  it("handler_withInvalidJson_returnsErrorResponse")
+  it("handler_whenPayloadExceeds64KB_destroysConnection")
+});
+```
+
+Or the shorter readable form (acceptable in vitest/jest):
+```typescript
+it("returns error response for invalid JSON")
+it("destroys connection when payload exceeds 64KB")
+```
+
+## Test Structure: AAA (Arrange-Act-Assert)
+
+```typescript
+it("returns sum of two numbers", () => {
+  // Arrange
+  const calc = new Calculator();
+
+  // Act
+  const result = calc.add(2, 3);
+
+  // Assert
+  expect(result).toBe(5);
+});
+```
+
+**Rules:**
+- One logical assertion per test (multiple `expect` calls OK if verifying one exit point)
+- No logic in tests (no `if`, `switch`, `for`, `try/catch`)
+- No shared mutable state between tests
+- Tests must be independent and run in any order
+
+## Exit Point Categories & Patterns
+
+### 1. Return Value Tests (simplest)
+```typescript
+it("parses duration string to milliseconds", () => {
+  expect(parseDuration("6h")).toBe(21_600_000);
+});
+```
+
+### 2. State Change Tests
+```typescript
+it("marks job as paused after pause()", async () => {
+  const store = new CronStore(tmpDir);
+  await store.writeJob(makeJob({ id: "x", enabled: true }));
+
+  await store.pauseJob("x");
+
+  const job = await store.readJob("x");
+  expect(job.enabled).toBe(false);
+});
+```
+
+### 3. Dependency Call Tests (use fakes/stubs)
+```typescript
+it("sends notification to transport on notify request", async () => {
+  const sent: string[] = [];
+  const fakeTransport = { notify: async (_ids: number[], text: string) => { sent.push(text); } };
+
+  await handler({ type: "notify", text: "hello" }, fakeTransport);
+
+  expect(sent).toEqual(["hello"]);
+});
+```
+
+## Stub vs Mock vs Fake
+
+| Type | Purpose | Assertion target? |
+|------|---------|-------------------|
+| **Stub** | Returns canned data | No — just enables the test |
+| **Mock** | Records calls for verification | Yes — assert it was called correctly |
+| **Fake** | Working implementation (in-memory) | No — used as real substitute |
+
+**Rule:** Prefer fakes > stubs > mocks. Only mock when verifying a dependency call exit point.
+
+## Anti-Patterns to Avoid
+
+1. **Testing implementation** — Don't assert private methods, internal state, or call order unless it's the exit point
+2. **Overspecified tests** — Don't verify every parameter; verify what matters
+3. **Test-per-method** — Don't mirror 1:1 with methods; test behaviors/scenarios
+4. **Shared state** — Don't use `beforeAll` to set up mutable state shared across tests
+5. **Magic values** — Use named constants or builder functions for test data
+6. **Flaky async** — Don't use `setTimeout` to "wait" for async; use proper awaits/events
+7. **Testing the framework** — Don't test that vitest/jest assertions work
+
+## Isolation Strategies
+
+### For I/O (filesystem, network):
+```typescript
+// Use temp directories
+const tmpDir = resolve(tmpdir(), `test-${process.pid}`);
+afterEach(() => rmSync(tmpDir, { recursive: true, force: true }));
+
+// Use injectable paths/sockets
+const server = new IpcServer(handler, TEST_SOCKET_PATH);
+```
+
+### For time:
+```typescript
+// Use vi.useFakeTimers() or inject a clock
+vi.useFakeTimers();
+vi.advanceTimersByTime(60_000);
+vi.useRealTimers();
+```
+
+### For external services:
+```typescript
+// Inject fakes via constructor or parameter
+const fakeNotifier = { notify: vi.fn().mockResolvedValue(undefined) };
+const runner = new CronRunner(store, fakeNotifier);
+```
+
+## Coverage Checklist
+
+For each unit, ask:
+1. What are the **entry points**? (How is it invoked?)
+2. What are the **exit points**? (What observable things happen?)
+3. For each exit point, what are the **scenarios**?
+   - Happy path
+   - Edge cases (empty input, boundary values)
+   - Error paths (invalid input, dependency failure)
+   - Concurrency (if applicable)
+
+## Readability Rules
+
+- Test file mirrors source: `src/ipc/server.ts` → `test/ipc-server.test.ts`
+- Group by unit: `describe("IpcServer", () => { ... })`
+- Helper functions at bottom of file or in `test/helpers/`
+- Builder pattern for complex test objects:
+  ```typescript
+  function makeJob(overrides: Partial<CronJobConfig> = {}): CronJobConfig {
+    return { id: "test", enabled: true, prompt: "do stuff", ...overrides };
+  }
+  ```
+
+## When to Stop Testing
+
+- Don't test trivial code (getters, simple delegation)
+- Don't test third-party libraries
+- Don't test generated/derived code
+- Stop at the trust boundary — if a dependency is tested elsewhere, stub it