chrome-relay 0.5.18 → 0.5.20

package/dist/cli.js

@@ -383,10 +383,34 @@ function parseChromeReadPageArgs(input) {
 }
 function parseChromeClickArgs(input) {
   const obj = asObject(input, TOOL_NAMES.CLICK);
-  return {
-    selector: requireString(obj, "selector", TOOL_NAMES.CLICK),
-    ...parseTargetArgs(obj)
-  };
+  const target = parseTargetArgs(obj, TOOL_NAMES.CLICK);
+  const x = optNumber(obj, "x", TOOL_NAMES.CLICK);
+  const y = optNumber(obj, "y", TOOL_NAMES.CLICK);
+  if (x !== void 0 !== (y !== void 0)) {
+    throw new RelayError({
+      code: "invalid_arguments",
+      message: "chrome_click_element: pass BOTH x and y, or neither (selector mode).",
+      tool: TOOL_NAMES.CLICK,
+      phase: "parse_arguments",
+      details: { received: { x: obj.x, y: obj.y } },
+      retryable: false
+    });
+  }
+  if (x !== void 0 && y !== void 0) {
+    return { ...target, kind: "coords", x, y };
+  }
+  const selector = optString(obj, "selector", TOOL_NAMES.CLICK);
+  if (selector) {
+    return { ...target, kind: "selector", selector };
+  }
+  throw new RelayError({
+    code: "invalid_arguments",
+    message: "chrome_click_element requires either a selector or x AND y.",
+    tool: TOOL_NAMES.CLICK,
+    phase: "parse_arguments",
+    details: { received: { selector: obj.selector, x: obj.x, y: obj.y } },
+    retryable: false
+  });
 }
 function parseChromeFillArgs(input) {
   const obj = asObject(input, TOOL_NAMES.FILL);
@@ -1077,7 +1101,7 @@ var init_dist = __esm({
 import { Command } from "commander";
 
 // src/index.ts
-var CHROME_RELAY_VERSION = true ? "0.5.18" : "0.0.0-dev";
+var CHROME_RELAY_VERSION = true ? "0.5.20" : "0.0.0-dev";
 
 // src/commands/shared.ts
 init_dist();
@@ -1320,6 +1344,19 @@ async function runDoctor() {
 
 // src/release-notes.ts
 var RELEASE_NOTES = {
+  "0.5.20": [
+    "BREAKING \u2014 `chrome-relay navigate` no longer steals focus by default. Background is now the implicit behavior; agents pass `--active` when they actually want the user looking at the new tab. The whole product pitch is 'operate without stealing focus' \u2014 the default needed to match.",
+    "`--inactive` flag removed entirely. It was the opt-in for the previous (wrong) default; now it'd be a no-op, and no-op flags are dead code. Agents that were passing `--inactive` should drop it (the behavior is now the default). Commander will reject the unknown flag \u2014 that's the right signal to update.",
+    "Workspace `chrome.windows.create({focused: false})` was already correct from earlier work, so workspaces are unaffected.",
+    'Click now dispatches as a pointer event. Added `pointerType: "mouse"` to all three `Input.dispatchMouseEvent` calls in the CLICK handler. Without it, CDP only fires mouse events \u2014 modern UI libs (Radix, React-Aria, Headless UI) listen on `pointerdown` and silently ignore mouse-only clicks, so dropdowns / menu triggers / select widgets stayed closed even though the click registered. Discovered while dogfooding `chrome-relay click --x N --y N` against npm\'s token-creation form; verified by completing the full token-creation flow end-to-end (radio click, combobox open, submit) via chrome-relay tools.',
+    "Tests: navigate test updated to assert no `active` field in default request, plus a new test verifying `--active` opts in to focus."
+  ],
+  "0.5.19": [
+    "Coordinate click. `chrome-relay click --x N --y N --tab N` dispatches a trusted Input.dispatchMouseEvent at the given pixel coordinates \u2014 no selector required. The selector positional became optional; the protocol parser collapses click args into a discriminated union (`kind: 'selector'` | `kind: 'coords'`) and rejects partial coords (`--x` without `--y`) with `invalid_arguments`.",
+    "Intentionally NOT shipping `click-text`. Once coord-click exists, finding text + clicking is fully composable with `js` (TreeWalker \u2192 getBoundingClientRect) \u2192 `click --x/--y`. Per the CLI philosophy, that's a smart wrapper, not a primitive. The two-step recipe lives in docs/clicking-strategies.md as the documented pattern.",
+    "Updated docs: docs/clicking-strategies.md now lists the 4 click verbs (click selector/coords, click-ax, js) and the text-recipe; docs/cli-philosophy.md explains the hide-vs-expose calculus that led to this design.",
+    "Tests: +2 in program.test.ts (coord-click + partial-coords rejection), +1 in args-all.test.ts (discriminated-union parsing). Existing click tests updated to expect the parsed `kind: 'selector'` shape now that CLI-side parseToolArgs runs before send."
+  ],
   "0.5.18": [
     "Force-visible on attach \u2014 actually fixes Cloudflare-style SPAs now. 0.5.17 had two bugs: (1) missing `Page.enable` before `Page.addScriptToEvaluateOnNewDocument` (the script registration silently failed, so the shim only applied to the current doc \u2014 page reloads dropped it), (2) the JS shim used one try-block so a deprecated `Object.defineProperty(document, 'webkitVisibilityState')` could throw and skip the `document.hasFocus` patch.",
     "0.5.18 fixes both: enables Page domain first, then applies each patch in its own try/catch. New patches added: `document.hasFocus` (overridden on both instance and Document.prototype), `Emulation.setFocusEmulationEnabled` via CDP, `document.wasDiscarded`. End result: Cloudflare Web Analytics dashboard now fully renders on backgrounded tabs without focus theft (verified live).",
@@ -1567,14 +1604,19 @@ function registerNavigation(ctx) {
     await run("get_windows_and_tabs", {});
   });
   tabOpt(
-    program.command("navigate <url>").description("Navigate a tab to a URL. Use --tab <id> to target an existing tab.").option("--new", "open in a new tab").option("--inactive", "do not activate the tab").addHelpText(
+    program.command("navigate <url>").description("Navigate a tab to a URL. Use --tab <id> to target an existing tab.").option("--new", "open in a new tab").option("--active", "activate the tab after navigating (default: background \u2014 no focus theft)").addHelpText(
       "after",
       `
 
 Examples:
-  chrome-relay navigate "https://example.com"
-  chrome-relay navigate --tab 123456789 "https://example.com"
-  chrome-relay navigate "https://example.com" --new --inactive
+  chrome-relay navigate "https://example.com"                    # navigate current tab
+  chrome-relay navigate --tab 123 "https://example.com"          # navigate an existing tab
+  chrome-relay navigate "https://example.com" --new              # open in a new background tab
+  chrome-relay navigate "https://example.com" --new --active     # open new tab AND show it to the user
+
+By default chrome-relay never steals focus \u2014 navigated tabs (new or
+existing) stay in whatever state they're in. Pass --active when you
+actually want the user looking at the page.
 `
     )
   ).action(async (url, opts) => {
@@ -1588,7 +1630,7 @@ Use "chrome-relay switch ${url}" to activate that tab, or "chrome-relay navigate
     }
     const extras = { url };
     if (opts.new) extras.newTab = true;
-    if (opts.inactive) extras.active = false;
+    if (opts.active) extras.active = true;
     await run("chrome_navigate", withBase(opts, extras));
   });
   program.command("switch <tabId>").description("Activate a tab by ID.").action(async (tabId) => {
@@ -1607,9 +1649,26 @@ Use "chrome-relay switch ${url}" to activate that tab, or "chrome-relay navigate
 function registerInput(ctx) {
   const { program, withBase, run } = ctx;
   tabOpt(
-    program.command("click <selector>").description("Click an element by CSS selector.")
+    program.command("click [selector]").description("Click an element. Pass a CSS selector OR --x and --y for a coordinate click.").option("--x <px>", "explicit x coordinate (CSS pixels); requires --y", (v) => Number(v)).option("--y <px>", "explicit y coordinate (CSS pixels); requires --x", (v) => Number(v)).addHelpText(
+      "after",
+      `
+
+Examples:
+  chrome-relay click 'button[aria-label="Save"]'
+  chrome-relay click --tab 123 --x 1327 --y 771
+
+Pick selector mode when the element has a stable CSS query. Pick
+coordinate mode when the page wraps content in unmarked divs (Cloudflare
+dashboard, Vercel dashboard, etc.) and you got the rect from a prior
+\`chrome-relay js\` call or a screenshot. See docs/clicking-strategies.md.
+`
+    )
   ).action(async (selector, opts) => {
-    await run("chrome_click_element", withBase(opts, { selector }));
+    const extras = {};
+    if (selector) extras.selector = selector;
+    if (typeof opts.x === "number") extras.x = opts.x;
+    if (typeof opts.y === "number") extras.y = opts.y;
+    await run("chrome_click_element", withBase(opts, extras));
   });
   tabOpt(
     program.command("fill <selector> <value>").description("Fill an input or textarea.")

package/dist/index.js

@@ -1,5 +1,5 @@
 // src/index.ts
-var CHROME_RELAY_VERSION = true ? "0.5.18" : "0.0.0-dev";
+var CHROME_RELAY_VERSION = true ? "0.5.20" : "0.0.0-dev";
 export {
   CHROME_RELAY_VERSION
 };

package/dist/native-host.js

@@ -56,7 +56,7 @@ function toBridgeError(unknownErr, fallbackTool) {
 }
 
 // src/index.ts
-var CHROME_RELAY_VERSION = true ? "0.5.18" : "0.0.0-dev";
+var CHROME_RELAY_VERSION = true ? "0.5.20" : "0.0.0-dev";
 
 // src/release-notes.ts
 function compareSemver(a, b) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "chrome-relay",
-  "version": "0.5.18",
+  "version": "0.5.20",
   "type": "module",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",