Fix CI: update tests for renamed API methods and add press to Action type

- Align test method names with current API (execute→action, pressKeys→press,
  launchApp→openApp, press_keys→press)
- Add missing "press" to Action type union
- Add MCP server instructions for agent guidance

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: k4cper-g

src/mcp/server.ts

@@ -12,10 +12,37 @@ import { z } from "zod";
 import { Session } from "../index.js";
 import { formatLine } from "../format.js";
 
-export const server = new McpServer({
-  name: "cup",
-  version: "0.1.0",
-});
+export const server = new McpServer(
+  { name: "cup", version: "0.1.0" },
+  {
+    instructions:
+      "CUP (Computer Use Protocol) gives you access to the UI accessibility " +
+      "tree of the user's computer.\n\n" +
+      "WORKFLOW — follow this pattern:\n" +
+      "1. snapshot to capture the active window's UI\n" +
+      "2. find to locate specific elements (PREFERRED over re-capturing)\n" +
+      "3. action to interact (click, type, press, etc.)\n" +
+      "4. Re-capture ONLY after actions change the UI\n\n" +
+      "TOOLS:\n" +
+      "- snapshot() — active window tree + window list (most common)\n" +
+      "- snapshot_app(app) — specific app by title (when not in foreground)\n" +
+      "- overview() — just the window list, near-instant\n" +
+      "- snapshot_desktop() — desktop icons and widgets\n" +
+      "- find(role/name/state) — search last tree without re-capturing\n" +
+      "- action(action, ...) — interact with elements or press keys\n" +
+      "- open_app(name) — open an app by name with fuzzy matching\n" +
+      "- screenshot(region) — visual context when tree isn't enough\n\n" +
+      "IMPORTANT — minimize token usage:\n" +
+      "- Use find(name=...) to locate elements — NOT repeated tree captures\n" +
+      "- Use overview() to discover what apps are open\n" +
+      "- Use snapshot_app(app='...') to target a specific app\n" +
+      "- snapshot() is your default starting point\n\n" +
+      "Element IDs (e.g., 'e14') are ephemeral — only valid for the most " +
+      "recent tree snapshot. After any action, re-capture before using IDs.\n\n" +
+      "Use action(action='press', keys='ctrl+s') for keyboard shortcuts.\n\n" +
+      "Use screenshot when you need visual context (colors, images, layout).",
+  },
+);
 
 // ---------------------------------------------------------------------------
 // Session state

src/types.ts

@@ -45,7 +45,7 @@ export interface AppInfo {
 }
 
 // ---------------------------------------------------------------------------
-// Roles (54 canonical ARIA-derived roles)
+// Roles (59 canonical ARIA-derived roles)
 // ---------------------------------------------------------------------------
 
 export type Role =
@@ -133,7 +133,7 @@ export type State =
   | "selected";
 
 // ---------------------------------------------------------------------------
-// Actions (15 element-level canonical actions)
+// Actions (15 element-level canonical actions + press for global keys)
 // ---------------------------------------------------------------------------
 
 export type Action =
@@ -146,6 +146,7 @@ export type Action =
   | "focus"
   | "increment"
   | "longpress"
+  | "press"
   | "rightclick"
   | "scroll"
   | "select"

tests/actions.test.ts

@@ -78,7 +78,7 @@ describe("VALID_ACTIONS", () => {
   it("contains all canonical actions", () => {
     const expected = [
       "click", "collapse", "decrement", "dismiss", "doubleclick",
-      "expand", "focus", "increment", "longpress", "press_keys",
+      "expand", "focus", "increment", "longpress", "press",
       "rightclick", "scroll", "select", "setvalue", "toggle", "type",
     ];
     for (const action of expected) {
@@ -95,35 +95,35 @@ describe("MacosActionHandler", () => {
   it("returns error for unknown action", async () => {
     const { MacosActionHandler } = await import("../src/actions/macos.js");
     const handler = new MacosActionHandler();
-    const result = await handler.execute(null, "fly", {});
+    const result = await handler.action(null, "fly", {});
     expect(result.success).toBe(false);
     expect(result.error?.toLowerCase()).toContain("not implemented");
   });
 
-  it("pressKeys works on macOS", async () => {
+  it("press works on macOS", async () => {
     if (process.platform !== "darwin") return; // skip on non-macOS
 
     const { MacosActionHandler } = await import("../src/actions/macos.js");
     const handler = new MacosActionHandler();
-    const result = await handler.pressKeys("escape");
+    const result = await handler.press("escape");
     expect(result.success).toBe(true);
     expect(result.message).toContain("Pressed");
   }, 60000); // First run compiles the Swift helper (~30s)
 
-  it("launchApp rejects empty name", async () => {
+  it("openApp rejects empty name", async () => {
     const { MacosActionHandler } = await import("../src/actions/macos.js");
     const handler = new MacosActionHandler();
-    const result = await handler.launchApp("");
+    const result = await handler.openApp("");
     expect(result.success).toBe(false);
     expect(result.error?.toLowerCase()).toContain("empty");
   });
 
-  it("launchApp rejects nonexistent app", async () => {
+  it("openApp rejects nonexistent app", async () => {
     if (process.platform !== "darwin") return; // skip on non-macOS
 
     const { MacosActionHandler } = await import("../src/actions/macos.js");
     const handler = new MacosActionHandler();
-    const result = await handler.launchApp("zzzznonexistentapp99999");
+    const result = await handler.openApp("zzzznonexistentapp99999");
     expect(result.success).toBe(false);
     expect(result.error?.toLowerCase()).toContain("no installed app");
   });