@vellumai/assistant 0.6.4 → 0.6.6

package/src/browser/operations.ts

@@ -0,0 +1,660 @@
+/**
+ * Shared browser operations contract.
+ *
+ * This module is the single execution entrypoint for all browser
+ * operations. Both the existing tool wrappers and the CLI command
+ * builder consume this contract. All metadata is defined inline —
+ * this module has no dependency on skill registration files.
+ *
+ * Responsibilities:
+ *   - Dispatch to existing browser-execution.ts implementations.
+ *   - Command-oriented metadata for CLI subcommand generation.
+ *   - `wait_for_download` mode-constraint enforcement.
+ *
+ * Identifier-only exports (BROWSER_TOOL_NAMES, name mapping helpers)
+ * live in `browser/identifiers.ts` to avoid pulling the browser
+ * execution stack into policy/classification modules. This module
+ * re-exports them for backwards compatibility.
+ */
+
+import {
+  executeBrowserAttach,
+  executeBrowserClick,
+  executeBrowserClose,
+  executeBrowserDetach,
+  executeBrowserExtract,
+  executeBrowserFillCredential,
+  executeBrowserHover,
+  executeBrowserNavigate,
+  executeBrowserPressKey,
+  executeBrowserScreenshot,
+  executeBrowserScroll,
+  executeBrowserSelectOption,
+  executeBrowserSnapshot,
+  executeBrowserStatus,
+  executeBrowserType,
+  executeBrowserWaitFor,
+} from "../tools/browser/browser-execution.js";
+import { browserManager } from "../tools/browser/browser-manager.js";
+import { normalizeBrowserMode } from "../tools/browser/browser-mode.js";
+import type { ToolContext, ToolExecutionResult } from "../tools/types.js";
+import {
+  BROWSER_OPERATIONS,
+  type BrowserOperation,
+  type BrowserOperationMeta,
+} from "./types.js";
+
+// ── Re-exports from identifiers.ts ───────────────────────────────────
+//
+// Identifier-only constants and helpers are defined in the lightweight
+// `browser/identifiers.ts` module so that policy/classification
+// consumers can import them without pulling in the browser execution
+// stack. Re-exported here for backwards compatibility with existing
+// callers that import from `browser/operations.js`.
+
+export {
+  BROWSER_TOOL_NAMES,
+  browserOperationToToolName,
+  browserToolNameToOperation,
+} from "./identifiers.js";
+
+/**
+ * All canonical browser operation identifiers (re-exported from types).
+ */
+export const BROWSER_OPERATION_NAMES: readonly BrowserOperation[] =
+  BROWSER_OPERATIONS;
+
+// ── Dispatch handlers ────────────────────────────────────────────────
+
+/**
+ * Handler signature for a browser operation dispatcher.
+ */
+type OperationHandler = (
+  input: Record<string, unknown>,
+  context: ToolContext,
+) => Promise<ToolExecutionResult>;
+
+/**
+ * Inline `wait_for_download` handler. Downloads are only supported
+ * on auto/local browser modes; the handler validates the mode and
+ * delegates to `browserManager.waitForDownload()`.
+ */
+async function executeWaitForDownload(
+  input: Record<string, unknown>,
+  context: ToolContext,
+): Promise<ToolExecutionResult> {
+  // Validate browser_mode: only auto/local are supported for downloads.
+  const modeResult = normalizeBrowserMode(input.browser_mode);
+  if ("error" in modeResult) {
+    return { content: `Error: ${modeResult.error}`, isError: true };
+  }
+  const { mode } = modeResult;
+  if (mode !== "auto" && mode !== "local") {
+    return {
+      content:
+        `Error: browser_wait_for_download does not support browser_mode "${mode}". ` +
+        `File downloads require the local Playwright backend. ` +
+        `Use browser_mode "auto" or "local" instead.`,
+      isError: true,
+    };
+  }
+
+  const timeout =
+    typeof input.timeout === "number"
+      ? Math.min(Math.max(input.timeout, 1000), 120_000)
+      : 30_000;
+
+  try {
+    const download = await browserManager.waitForDownload(
+      context.conversationId,
+      timeout,
+    );
+    return {
+      content: JSON.stringify({
+        filename: download.filename,
+        path: download.path,
+      }),
+      isError: false,
+    };
+  } catch (err) {
+    const msg = err instanceof Error ? err.message : String(err);
+    return { content: `Error: ${msg}`, isError: true };
+  }
+}
+
+/**
+ * Registry mapping each operation to its dispatch handler.
+ * Every entry in BROWSER_OPERATIONS must have a corresponding handler.
+ */
+const DISPATCH_HANDLERS: Record<BrowserOperation, OperationHandler> = {
+  navigate: executeBrowserNavigate,
+  snapshot: executeBrowserSnapshot,
+  screenshot: executeBrowserScreenshot,
+  close: executeBrowserClose,
+  attach: executeBrowserAttach,
+  detach: executeBrowserDetach,
+  click: executeBrowserClick,
+  type: executeBrowserType,
+  press_key: executeBrowserPressKey,
+  scroll: executeBrowserScroll,
+  select_option: executeBrowserSelectOption,
+  hover: executeBrowserHover,
+  wait_for: executeBrowserWaitFor,
+  extract: executeBrowserExtract,
+  wait_for_download: executeWaitForDownload,
+  fill_credential: executeBrowserFillCredential,
+  status: executeBrowserStatus,
+};
+
+// ── Execute ──────────────────────────────────────────────────────────
+
+/**
+ * Execute a browser operation by its canonical identifier.
+ *
+ * This is the single execution entrypoint. Callers pass the operation
+ * name (e.g. `"navigate"`), a flat input object, and a {@link ToolContext}.
+ * The function looks up the handler in the dispatch registry and
+ * delegates to the existing browser-execution.ts implementation.
+ *
+ * @param operation - Canonical operation identifier (e.g. `"navigate"`).
+ * @param input     - Flat input object matching the operation's field schema.
+ * @param context   - Tool execution context (conversation ID, signal, etc.).
+ * @returns The tool execution result from the underlying handler.
+ *   If the operation identifier is not recognized, returns an error
+ *   result (`isError: true`) rather than throwing.
+ */
+export async function executeBrowserOperation(
+  operation: BrowserOperation,
+  input: Record<string, unknown>,
+  context: ToolContext,
+): Promise<ToolExecutionResult> {
+  const handler = DISPATCH_HANDLERS[operation];
+  if (!handler) {
+    return {
+      content: `Error: Unknown browser operation "${operation}".`,
+      isError: true,
+    };
+  }
+  return handler(input, context);
+}
+
+// ── Command-oriented metadata ────────────────────────────────────────
+
+/**
+ * Metadata for every browser operation, describing fields, types, and
+ * constraints. Used by the CLI command builder to generate subcommands.
+ *
+ * The `browser_mode` field is handled as a shared parent-level option
+ * on the `assistant browser` command (--browser-mode), not as a
+ * per-operation field. The `activity` field is omitted because it is
+ * an internal execution concern, not a user-facing parameter.
+ */
+export const BROWSER_OPERATION_META: readonly BrowserOperationMeta[] = [
+  {
+    operation: "navigate",
+    description: "Navigate the browser to a URL and return the page title.",
+    fields: [
+      {
+        name: "url",
+        type: "string",
+        description: "The URL to navigate to.",
+        required: true,
+      },
+      {
+        name: "allow_private_network",
+        type: "boolean",
+        description: "Allow navigation to localhost/private-network hosts.",
+        required: false,
+      },
+    ],
+    helpText: `Loads the given URL and waits for the page to reach a stable state.
+Returns the page title on success.
+
+Examples:
+  $ assistant browser navigate --url https://example.com
+  $ assistant browser navigate --url http://localhost:3000 --allow-private-network
+  $ assistant browser --session s1 navigate --url https://github.com`,
+  },
+  {
+    operation: "snapshot",
+    description:
+      "List interactive elements on the current page with unique IDs.",
+    fields: [],
+    helpText: `Returns a structured list of interactive elements (buttons, links,
+inputs, etc.) with stable element IDs that can be passed to click,
+type, and other element-targeting commands.
+
+Examples:
+  $ assistant browser snapshot
+  $ assistant browser --json snapshot`,
+  },
+  {
+    operation: "screenshot",
+    description: "Take a visual screenshot of the current page.",
+    fields: [
+      {
+        name: "full_page",
+        type: "boolean",
+        description:
+          "Capture the full scrollable page instead of just the viewport.",
+        required: false,
+      },
+    ],
+    helpText: `Captures a JPEG screenshot. Use --output to save to a file, or
+--json to receive base64-encoded image data in the output.
+
+Examples:
+  $ assistant browser screenshot --output page.jpg
+  $ assistant browser screenshot --full-page --output full.jpg
+  $ assistant browser --json screenshot`,
+  },
+  {
+    operation: "close",
+    description: "Close the browser page for the current conversation.",
+    fields: [
+      {
+        name: "close_all_pages",
+        type: "boolean",
+        description: "Close all browser pages and the browser context.",
+        required: false,
+      },
+    ],
+    helpText: `Closes the browser page for the current session. Use --close-all-pages
+to tear down the entire browser context including all pages.
+
+Examples:
+  $ assistant browser close
+  $ assistant browser close --close-all-pages
+  $ assistant browser --session s1 close`,
+  },
+  {
+    operation: "attach",
+    description: "Attach the Chrome debugger to the active browser tab.",
+    fields: [],
+    helpText: `Connects the assistant to a running Chrome instance via the Chrome
+DevTools Protocol. Required before interacting with Chrome-attached tabs.
+
+Examples:
+  $ assistant browser attach
+  $ assistant browser --session s1 attach`,
+  },
+  {
+    operation: "detach",
+    description: "Detach the Chrome debugger from the active browser tab.",
+    fields: [],
+    helpText: `Disconnects the assistant from the Chrome DevTools Protocol session.
+The browser tab continues running but is no longer controlled.
+
+Examples:
+  $ assistant browser detach
+  $ assistant browser --session s1 detach`,
+  },
+  {
+    operation: "click",
+    description: "Click an element on the page.",
+    fields: [
+      {
+        name: "element_id",
+        type: "string",
+        description: "Element ID from a previous browser snapshot.",
+        required: false,
+      },
+      {
+        name: "selector",
+        type: "string",
+        description: "CSS selector to target.",
+        required: false,
+      },
+    ],
+    helpText: `Clicks an element identified by element ID (from snapshot) or CSS
+selector. Provide at least one of --element-id or --selector.
+
+Examples:
+  $ assistant browser click --element-id e14
+  $ assistant browser click --selector "#login-button"
+  $ assistant browser click --selector "a.nav-link"`,
+  },
+  {
+    operation: "type",
+    description: "Type text into an input element.",
+    fields: [
+      {
+        name: "text",
+        type: "string",
+        description: "The text to type into the element.",
+        required: true,
+      },
+      {
+        name: "element_id",
+        type: "string",
+        description: "Element ID from a previous browser snapshot.",
+        required: false,
+      },
+      {
+        name: "selector",
+        type: "string",
+        description: "CSS selector to target.",
+        required: false,
+      },
+      {
+        name: "clear_first",
+        type: "boolean",
+        description: "Clear existing content before typing. Default: true.",
+        required: false,
+      },
+      {
+        name: "press_enter",
+        type: "boolean",
+        description: "Press Enter after typing the text.",
+        required: false,
+      },
+    ],
+    helpText: `Types text into the focused or targeted element. By default, existing
+content is cleared first (--clear-first). Use --no-clear-first to
+append to existing content. Use --press-enter to submit after typing.
+
+Examples:
+  $ assistant browser type --text "hello world" --element-id e14
+  $ assistant browser type --text "search query" --selector "#search" --press-enter
+  $ assistant browser type --text "append this" --no-clear-first`,
+  },
+  {
+    operation: "press_key",
+    description: "Press a keyboard key, optionally targeting an element.",
+    fields: [
+      {
+        name: "key",
+        type: "string",
+        description:
+          'The key to press (e.g. "Enter", "Escape", "Tab", "ArrowDown").',
+        required: true,
+      },
+      {
+        name: "element_id",
+        type: "string",
+        description: "Optional element ID from browser snapshot.",
+        required: false,
+      },
+      {
+        name: "selector",
+        type: "string",
+        description: "Optional CSS selector to target.",
+        required: false,
+      },
+    ],
+    helpText: `Sends a keyboard key press. If --element-id or --selector is given,
+the key is dispatched to that element; otherwise to the focused element.
+
+Examples:
+  $ assistant browser press-key --key Enter
+  $ assistant browser press-key --key Tab --element-id e5
+  $ assistant browser press-key --key Escape`,
+  },
+  {
+    operation: "scroll",
+    description: "Scroll the page or a specific element.",
+    fields: [
+      {
+        name: "direction",
+        type: "string",
+        description: "The direction to scroll.",
+        required: true,
+        enum: ["up", "down", "left", "right"],
+      },
+      {
+        name: "amount",
+        type: "number",
+        description: "The number of pixels to scroll. Default: 500.",
+        required: false,
+      },
+      {
+        name: "element_id",
+        type: "string",
+        description: "Optional element ID to scroll within.",
+        required: false,
+      },
+      {
+        name: "selector",
+        type: "string",
+        description: "Optional CSS selector of element to scroll within.",
+        required: false,
+      },
+    ],
+    helpText: `Scrolls the page or a specific scrollable element. Direction is
+required; amount defaults to 500 pixels.
+
+Examples:
+  $ assistant browser scroll --direction down
+  $ assistant browser scroll --direction up --amount 1000
+  $ assistant browser scroll --direction down --element-id e8`,
+  },
+  {
+    operation: "select_option",
+    description: "Select an option from a native <select> element.",
+    fields: [
+      {
+        name: "element_id",
+        type: "string",
+        description: "Element ID of the <select> from browser snapshot.",
+        required: false,
+      },
+      {
+        name: "selector",
+        type: "string",
+        description: "CSS selector for the <select> element.",
+        required: false,
+      },
+      {
+        name: "value",
+        type: "string",
+        description: "The value attribute of the <option> to select.",
+        required: false,
+      },
+      {
+        name: "label",
+        type: "string",
+        description: "The visible text of the <option> to select.",
+        required: false,
+      },
+      {
+        name: "index",
+        type: "number",
+        description: "The zero-based index of the <option> to select.",
+        required: false,
+      },
+    ],
+    helpText: `Selects an option in a <select> element by value, label, or index.
+Provide at least one of --value, --label, or --index to identify
+the option, and --element-id or --selector to identify the <select>.
+
+Examples:
+  $ assistant browser select-option --element-id e12 --label "United States"
+  $ assistant browser select-option --selector "#country" --value "us"
+  $ assistant browser select-option --element-id e12 --index 3`,
+  },
+  {
+    operation: "hover",
+    description: "Hover over an element on the page.",
+    fields: [
+      {
+        name: "element_id",
+        type: "string",
+        description: "Element ID from a previous browser snapshot.",
+        required: false,
+      },
+      {
+        name: "selector",
+        type: "string",
+        description: "CSS selector to target.",
+        required: false,
+      },
+    ],
+    helpText: `Moves the mouse cursor over an element to trigger hover effects
+(tooltips, dropdowns, etc.). Provide --element-id or --selector.
+
+Examples:
+  $ assistant browser hover --element-id e7
+  $ assistant browser hover --selector ".dropdown-trigger"`,
+  },
+  {
+    operation: "wait_for",
+    description:
+      "Wait for a condition: a CSS selector, text, or fixed duration.",
+    fields: [
+      {
+        name: "selector",
+        type: "string",
+        description: "Wait for an element matching this CSS selector.",
+        required: false,
+      },
+      {
+        name: "text",
+        type: "string",
+        description: "Wait for this text to appear on the page.",
+        required: false,
+      },
+      {
+        name: "duration",
+        type: "number",
+        description: "Wait for this many milliseconds.",
+        required: false,
+      },
+      {
+        name: "timeout",
+        type: "number",
+        description:
+          "Maximum wait time in milliseconds. Default and max: 30000.",
+        required: false,
+      },
+    ],
+    helpText: `Blocks until a condition is met: an element appears (--selector),
+text appears (--text), or a fixed duration elapses (--duration).
+Provide exactly one condition. --timeout caps the overall wait.
+
+Examples:
+  $ assistant browser wait-for --selector ".results-loaded"
+  $ assistant browser wait-for --text "Success"
+  $ assistant browser wait-for --duration 2000`,
+  },
+  {
+    operation: "extract",
+    description: "Extract the text content of the current page.",
+    fields: [
+      {
+        name: "include_links",
+        type: "boolean",
+        description: "Include a list of links found on the page.",
+        required: false,
+      },
+    ],
+    helpText: `Extracts the visible text content of the current page. Optionally
+includes a list of all links found on the page.
+
+Examples:
+  $ assistant browser extract
+  $ assistant browser extract --include-links
+  $ assistant browser --json extract`,
+  },
+  {
+    operation: "wait_for_download",
+    description: "Wait for a file download to complete on the current page.",
+    allowedModes: ["auto", "local"],
+    fields: [
+      {
+        name: "timeout",
+        type: "number",
+        description:
+          "Maximum wait time in milliseconds. Default: 30000, max: 120000.",
+        required: false,
+      },
+    ],
+    helpText: `Waits for an in-progress file download to complete and returns the
+filename and path. Only supported in "auto" and "local" browser modes.
+
+Examples:
+  $ assistant browser wait-for-download
+  $ assistant browser wait-for-download --timeout 60000`,
+  },
+  {
+    operation: "fill_credential",
+    description:
+      "Fill a stored credential into a form field without exposing the value.",
+    fields: [
+      {
+        name: "service",
+        type: "string",
+        description: "Credential vault service name.",
+        required: true,
+      },
+      {
+        name: "field",
+        type: "string",
+        description: "Credential vault field name.",
+        required: true,
+      },
+      {
+        name: "element_id",
+        type: "string",
+        description: "Element ID from browser snapshot.",
+        required: false,
+      },
+      {
+        name: "selector",
+        type: "string",
+        description: "CSS selector for target element.",
+        required: false,
+      },
+      {
+        name: "press_enter",
+        type: "boolean",
+        description: "Press Enter after filling.",
+        required: false,
+      },
+    ],
+    helpText: `Fills a credential from the assistant's credential vault into a form
+field. The credential value is never exposed in CLI output. Use
+'assistant credentials list' to see available service:field pairs.
+
+Examples:
+  $ assistant browser fill-credential --service github --field token --element-id e9
+  $ assistant browser fill-credential --service github --field password --selector "#password" --press-enter`,
+  },
+  {
+    operation: "status",
+    description: "Check browser backend readiness and remediation guidance.",
+    fields: [
+      {
+        name: "check_local_launch",
+        type: "boolean",
+        description:
+          "Run an active local Playwright launch probe. Default: false.",
+        required: false,
+      },
+    ],
+    helpText: `Reports the readiness of available browser backends (local Playwright,
+Chrome extension). Includes remediation steps if a backend is not ready.
+Use --check-local-launch for an active Playwright launch probe (slower).
+
+Examples:
+  $ assistant browser status
+  $ assistant browser status --check-local-launch
+  $ assistant browser --json status`,
+  },
+];
+
+// ── Lookup helper ────────────────────────────────────────────────────
+
+/** Index for O(1) metadata lookups by operation. */
+const META_BY_OPERATION = new Map<BrowserOperation, BrowserOperationMeta>(
+  BROWSER_OPERATION_META.map((m) => [m.operation, m]),
+);
+
+/**
+ * Get metadata for a specific operation. Returns `undefined` if the
+ * operation is not recognized.
+ */
+export function getBrowserOperationMeta(
+  operation: BrowserOperation,
+): BrowserOperationMeta | undefined {
+  return META_BY_OPERATION.get(operation);
+}