feat(tools): add reusable execution contexts for shell, filesystem, and browser tools

2026-05-15 05:38:35 +00:00
parent f183bf19ac
commit f138495edf
10 changed files with 604 additions and 103 deletions
@@ -0,0 +1,94 @@
+import * as plugins from './plugins.js';
+import {
+  formatToolOutput,
+  type IBrowserToolInput,
+  type IToolExecutionContext,
+  type TBrowserToolAction,
+} from './tool.context.js';
+
+export interface ICreateBrowserToolsOptions {
+  /** Maximum output lines before truncating. */
+  maxLines?: number;
+  /** Maximum output bytes before truncating. */
+  maxBytes?: number;
+}
+
+const browserActions = new Set<TBrowserToolAction>(['navigate', 'snapshot', 'screenshot', 'click', 'fill', 'press', 'evaluate', 'close']);
+
+export function createBrowserTools(context: IToolExecutionContext, options: ICreateBrowserToolsOptions = {}): plugins.ToolSet {
+  return {
+    browser: plugins.tool({
+      description: [
+        'Control a browser supplied by the host execution context for web UI inspection and interaction.',
+        'Actions: navigate, snapshot, screenshot, click, fill, press, evaluate, close.',
+        'Use snapshot after navigation or interaction to inspect page text and interactive selectors before choosing the next action.',
+        'Actions that navigate or modify page state require host permission when configured.',
+      ].join(' '),
+      inputSchema: plugins.z.object({
+        action: plugins.z.string().default('snapshot').describe('Action: navigate, snapshot, screenshot, click, fill, press, evaluate, or close'),
+        url: plugins.z.string().optional().describe('URL for navigate'),
+        selector: plugins.z.string().optional().describe('CSS or Playwright selector for click/fill'),
+        text: plugins.z.string().optional().describe('Text for fill, key name for press, or screenshot mode/full-page hint'),
+        script: plugins.z.string().optional().describe('JavaScript expression or function body for evaluate'),
+        timeoutMs: plugins.z.number().optional().describe('Optional action timeout in milliseconds'),
+      }),
+      execute: async (input: IBrowserToolInput) => {
+        if (!context.browser) {
+          throw new Error('Browser tool is not available in this execution context.');
+        }
+        const action = normalizeBrowserAction(input.action);
+        await requestBrowserPermission(context, { ...input, action });
+        const result = await context.browser.execute({ ...input, action }, {
+          timeoutMs: input.timeoutMs,
+          abortSignal: context.abortSignal,
+        });
+        return plugins.truncateOutput(formatToolOutput(result), {
+          maxLines: options.maxLines,
+          maxBytes: options.maxBytes,
+        }).content;
+      },
+    }),
+  };
+}
+
+const normalizeBrowserAction = (input: unknown): TBrowserToolAction => {
+  const action = typeof input === 'string' && input.trim()
+    ? input.trim().toLowerCase()
+    : 'snapshot';
+  if (browserActions.has(action as TBrowserToolAction)) {
+    return action as TBrowserToolAction;
+  }
+  throw new Error(`Unsupported browser action: ${String(input)}. Use one of: ${[...browserActions].join(', ')}.`);
+};
+
+const requestBrowserPermission = async (context: IToolExecutionContext, input: IBrowserToolInput & { action: TBrowserToolAction }): Promise<void> => {
+  if (!context.requestPermission) return;
+  if (input.action === 'snapshot' || input.action === 'screenshot') return;
+  const titleByAction: Record<TBrowserToolAction, string> = {
+    navigate: 'Navigate browser',
+    snapshot: 'Inspect browser',
+    screenshot: 'Capture browser screenshot',
+    click: 'Click browser element',
+    fill: 'Fill browser element',
+    press: 'Press browser key',
+    evaluate: 'Evaluate browser JavaScript',
+    close: 'Close browser session',
+  };
+  await context.requestPermission({
+    type: 'browser',
+    title: titleByAction[input.action],
+    metadata: {
+      action: input.action,
+      url: input.url,
+      selector: input.selector,
+      key: input.action === 'press' ? input.text : undefined,
+      textLength: input.action === 'fill' ? input.text?.length ?? 0 : undefined,
+      scriptPreview: input.action === 'evaluate' && input.script ? compactMetadataText(input.script) : undefined,
+    },
+  });
+};
+
+const compactMetadataText = (text: string): string => {
+  const compacted = text.replace(/\s+/g, ' ').trim();
+  return compacted.length > 160 ? `${compacted.slice(0, 157)}...` : compacted;
+};