Advertise flat object schemas for browser_handoff and browser_credentials in OMP

CHANGELOG.md

@@ -1,6 +1,7 @@
 # Changelog
 
 ## Unreleased
+- Fixed OMP requests to OpenAI-compatible providers failing with `Invalid schema for function 'browser_handoff'`. Codex and DeepSeek reject a tool whose root parameters schema is a union, so OMP runs that forward extension schemas unnormalized failed on every request. `browser_handoff` and `browser_credentials` now advertise one strict object, as `browser_open` and `browser_snapshot` already did; Core still enforces the fields each operation or action requires. Pi schemas are unchanged.
 - Fixed `browser_open` returning a page revision that was already stale. Chrome starts loading a created tab after the tab exists, and that load advanced the revision right after `browser_open` returned, so the first `browser_act` with the returned revision failed with `stale_page_revision`. `browser_open` now returns after the first load completes, up to 10 seconds, and reports the loaded page's revision and URL.
 - Made failed `browser_act` batches report their progress. The error `details` now carry `failed_action_index` and the `completed_actions` results, and in multi-action batches the message names the failing step and how many steps completed before it. A failure after a completed step now reports `outcome: "unknown"` instead of `not_started`, because the earlier steps already changed the page.
 - Fixed the OMP adapter's tool schemas lagging behind the protocol. OMP now offers `hover`, `double_click`, `context_click`, press `modifiers`, `frame_id`, `expect_download`, dialog `prompt_text`, snapshot `frame_id` and `match`, and the download-cursor, selector-state, and value waits that the Pi adapter already accepted, and rejects duplicate `keep_tab_ids` as Pi does. A parity test now fails when the OMP schemas accept less than the Pi schemas.

packages/omp/src/index.ts

@@ -283,45 +283,36 @@ const DEFINITIONS: ReadonlyArray<{
       label: "Browser Handoff",
       description: "Request user attention for credentials, MFA, CAPTCHA, or other human-only input. Non-blocking: post a notice, tell the user in chat, then verify the page yourself and resolve or dismiss it by notice_id.",
       approval: "write",
-      schema: (z) => z.union([
-        z.object({
-          operation: z.literal("request"),
-          tab_id: z.number().int().min(0),
-          expected_page_revision: z.number().int().min(0),
-          prompt: z.string().min(1).max(2000),
-          completion: z.object({
-            kind: z.enum(["url", "selector"]),
-            value: z.string().min(1).max(65_536),
-          }).strict().optional(),
-          timeout_ms: z.number().int().min(1000).max(900_000).optional(),
-        }).strict(),
-        z.object({
-          operation: z.enum(["status", "resolve", "dismiss"]),
-          notice_id: z.string().min(1).max(256),
-        }).strict(),
-      ]),
+      // One flat object covers every operation: OpenAI-compatible providers reject a root union.
+      // Core enforces which fields each operation requires.
+      schema: (z) => z.object({
+        operation: z.enum(["request", "status", "resolve", "dismiss"]),
+        tab_id: z.number().int().min(0).optional(),
+        expected_page_revision: z.number().int().min(0).optional(),
+        prompt: z.string().min(1).max(2000).optional(),
+        completion: z.object({
+          kind: z.enum(["url", "selector"]),
+          value: z.string().min(1).max(65_536),
+        }).strict().optional(),
+        timeout_ms: z.number().int().min(1000).max(900_000).optional(),
+        notice_id: z.string().min(1).max(256).optional(),
+      }).strict(),
     },
     {
       name: "browser_credentials",
       label: "Browser Credentials",
       description: "Prefer this before human handoff: use a URL-matching 1Password Login without exposing its values. Prepare first; more than three matches require the user. Fill only, then submit through Browser Act.",
       approval: "write",
-      schema: (z) => z.union([
-        z.object({
-          action: z.literal("prepare"),
-          tab_id: z.number().int().min(0),
-          expected_page_revision: z.number().int().min(0),
-        }).strict(),
-        z.object({
-          action: z.enum(["fill", "next"]),
-          tab_id: z.number().int().min(0),
-          expected_page_revision: z.number().int().min(0),
-          credential_token: z.string().min(32).max(256),
-          username_ref: z.string().min(1).max(256).optional(),
-          password_ref: z.string().min(1).max(256).optional(),
-          otp_ref: z.string().min(1).max(256).optional(),
-        }).strict(),
-      ]),
+      // One flat object covers every action; Core enforces credential_token for fill and next.
+      schema: (z) => z.object({
+        action: z.enum(["prepare", "fill", "next"]),
+        tab_id: z.number().int().min(0),
+        expected_page_revision: z.number().int().min(0),
+        credential_token: z.string().min(32).max(256).optional(),
+        username_ref: z.string().min(1).max(256).optional(),
+        password_ref: z.string().min(1).max(256).optional(),
+        otp_ref: z.string().min(1).max(256).optional(),
+      }).strict(),
     },
     {
       name: "browser_commit",

packages/omp/test/extension.test.ts

@@ -181,12 +181,12 @@ test("Standard OMP mode registers the complete Core RPC tool surface", () => {
 });
 
 
-test("Standard read and open tools expose provider-compatible object schemas", () => {
-  const registered = register(false);
+test("every OMP tool exposes a provider-compatible root object schema", () => {
+  const registered = register(true);
+  for (const tool of registered.tools) {
+    expect({ tool: tool.name, parameters: tool.parameters }).toMatchObject({ tool: tool.name, parameters: { kind: "object" } });
+  }
   const open = registered.tools.find((tool) => tool.name === "browser_open");
-  const snapshot = registered.tools.find((tool) => tool.name === "browser_snapshot");
-  expect(open?.parameters).toMatchObject({ kind: "object" });
-  expect(snapshot?.parameters).toMatchObject({ kind: "object" });
   expect(stubState(open?.parameters).vocabulary.values).toContain("new_window");
 });