@phnx-labs/agents-cli 1.17.3 → 1.17.4

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,12 @@
1
1
  # Changelog
2
2
 
3
+ ## 1.17.4
4
+
5
+ **Browser**
6
+
7
+ - `agents browser type` now detects rich-text editor frameworks (Lexical, ProseMirror, Slate, Draft.js, Quill, CKEditor5, Trix) by walking up to 5 ancestor levels from each textbox and tagging refs with `[editor=<framework>]`. Editor-tagged refs route through the WHATWG `beforeinput` dispatch (`InputEvent('beforeinput', { inputType: 'insertText', ... })`) for Lexical/ProseMirror/Slate/Quill/CKEditor5/Draft and `el.editor.insertString()` for Trix. `agents browser refs --json` surfaces the new `editor` field, and `type --clear` prepends a select-all + `deleteContentBackward` dispatch before inserting.
8
+ - Plain-input reliability also improved: `typeText` now issues a single CDP `Input.insertText` instead of per-character `dispatchKeyEvent`, so framework-controlled inputs (React, Vue, Solid, MUI/Chakra/Mantine `TextField`, masked-number fields, Canva-style pickers) actually receive `beforeinput`/`input`/`textInput` events. `focusNode` falls back to the first focusable descendant when `DOM.focus` throws "Element is not focusable" — fixes wrapper-ref UIs like Slack composer, Linear comments, Notion blocks, and every MUI/Chakra/Mantine `TextField`. ([#12](https://github.com/phnx-labs/agents-cli/pull/12))
9
+
3
10
  ## 1.17.3
4
11
 
5
12
  **Browser**
@@ -754,6 +754,7 @@ function registerTaskCommands(browser) {
754
754
  .option('-t, --tab <tabId>', 'Tab ID (defaults to current)')
755
755
  .option('--all', 'Include non-interactive elements')
756
756
  .option('-l, --limit <n>', 'Max elements (default 500)', '500')
757
+ .option('--json', 'Output machine-readable JSON')
757
758
  .action(async (task, opts) => {
758
759
  const response = await sendIPCRequest({
759
760
  action: 'refs',
@@ -763,9 +764,18 @@ function registerTaskCommands(browser) {
763
764
  limit: parseInt(opts.limit, 10),
764
765
  });
765
766
  if (!response.ok) {
766
- console.error(response.error);
767
+ if (opts.json) {
768
+ console.log(JSON.stringify({ ok: false, error: response.error }));
769
+ }
770
+ else {
771
+ console.error(response.error);
772
+ }
767
773
  process.exit(1);
768
774
  }
775
+ if (opts.json) {
776
+ console.log(JSON.stringify(response.nodes ?? [], null, 2));
777
+ return;
778
+ }
769
779
  console.log(response.refs);
770
780
  });
771
781
  browser
@@ -789,6 +799,7 @@ function registerTaskCommands(browser) {
789
799
  .command('type <task> <ref> <text>')
790
800
  .description('Type text into an element by ref')
791
801
  .option('-t, --tab <tabId>', 'Tab ID (defaults to current)')
802
+ .option('--clear', 'Clear editor content before typing')
792
803
  .action(async (task, ref, text, opts) => {
793
804
  const response = await sendIPCRequest({
794
805
  action: 'type',
@@ -796,6 +807,7 @@ function registerTaskCommands(browser) {
796
807
  tabId: opts.tab,
797
808
  ref: parseInt(ref, 10),
798
809
  text,
810
+ clear: opts.clear,
799
811
  });
800
812
  if (!response.ok) {
801
813
  console.error(response.error);
@@ -0,0 +1,3 @@
1
+ import type { CDPClient } from './cdp.js';
2
+ import type { RefNode } from './refs.js';
3
+ export declare function typeEditorText(cdp: CDPClient, sessionId: string, node: RefNode, text: string, clear?: boolean): Promise<void>;
@@ -0,0 +1,50 @@
1
+ const BEFOREINPUT_INSERT_FN = `(function(text) {
2
+ this.focus();
3
+ var sel = window.getSelection();
4
+ var range = document.createRange();
5
+ range.selectNodeContents(this);
6
+ range.collapse(false);
7
+ sel.removeAllRanges();
8
+ sel.addRange(range);
9
+ this.dispatchEvent(new InputEvent('beforeinput', {
10
+ inputType: 'insertText',
11
+ data: text,
12
+ bubbles: true,
13
+ cancelable: true,
14
+ composed: true,
15
+ }));
16
+ })`;
17
+ const BEFOREINPUT_CLEAR_FN = `(function() {
18
+ this.focus();
19
+ var sel = window.getSelection();
20
+ var range = document.createRange();
21
+ range.selectNodeContents(this);
22
+ sel.removeAllRanges();
23
+ sel.addRange(range);
24
+ this.dispatchEvent(new InputEvent('beforeinput', {
25
+ inputType: 'deleteContentBackward',
26
+ bubbles: true,
27
+ cancelable: true,
28
+ composed: true,
29
+ }));
30
+ })`;
31
+ const TRIX_INSERT_FN = `(function(text) { this.editor.insertString(text); })`;
32
+ export async function typeEditorText(cdp, sessionId, node, text, clear = false) {
33
+ const { object } = await cdp.send('DOM.resolveNode', { backendNodeId: node.backendNodeId }, sessionId);
34
+ if (!object.objectId)
35
+ throw new Error(`Could not resolve DOM node for ref ${node.ref}`);
36
+ const objectId = object.objectId;
37
+ try {
38
+ if (node.editor === 'trix') {
39
+ await cdp.send('Runtime.callFunctionOn', { objectId, functionDeclaration: TRIX_INSERT_FN, arguments: [{ value: text }], returnByValue: true }, sessionId);
40
+ return;
41
+ }
42
+ if (clear) {
43
+ await cdp.send('Runtime.callFunctionOn', { objectId, functionDeclaration: BEFOREINPUT_CLEAR_FN, arguments: [], returnByValue: true }, sessionId);
44
+ }
45
+ await cdp.send('Runtime.callFunctionOn', { objectId, functionDeclaration: BEFOREINPUT_INSERT_FN, arguments: [{ value: text }], returnByValue: true }, sessionId);
46
+ }
47
+ finally {
48
+ await cdp.send('Runtime.releaseObject', { objectId }, sessionId);
49
+ }
50
+ }
@@ -8,11 +8,13 @@ export async function hoverAtCoords(cdp, sessionId, x, y) {
8
8
  export async function scrollAtCoords(cdp, sessionId, x, y, deltaX, deltaY) {
9
9
  await cdp.send('Input.dispatchMouseEvent', { type: 'mouseWheel', x, y, deltaX, deltaY }, sessionId);
10
10
  }
11
+ // `Input.insertText` is the CDP native text-insertion method. It dispatches a
12
+ // real `beforeinput`/`input`/`textInput` sequence on the focused element, which
13
+ // is what framework-controlled inputs (React, Vue, Solid, contenteditable
14
+ // editors) actually listen for. Per-character `dispatchKeyEvent` only fires
15
+ // `keydown`/`keyup` with no input event, so controlled inputs ignore it.
11
16
  export async function typeText(cdp, sessionId, text) {
12
- for (const char of text) {
13
- await cdp.send('Input.dispatchKeyEvent', { type: 'keyDown', text: char }, sessionId);
14
- await cdp.send('Input.dispatchKeyEvent', { type: 'keyUp', text: char }, sessionId);
15
- }
17
+ await cdp.send('Input.insertText', { text }, sessionId);
16
18
  }
17
19
  const KEY_CODES = {
18
20
  Enter: { key: 'Enter', code: 'Enter', keyCode: 13 },
@@ -50,6 +52,41 @@ export async function pressKey(cdp, sessionId, keyName) {
50
52
  nativeVirtualKeyCode: keyInfo.keyCode,
51
53
  }, sessionId);
52
54
  }
55
+ const FOCUS_DESCENDANT_FN = `(function() {
56
+ const selector = 'input:not([disabled]):not([type=hidden]),textarea:not([disabled]),select:not([disabled]),[contenteditable=""],[contenteditable=true],[tabindex]:not([tabindex="-1"])';
57
+ const candidates = this.querySelectorAll(selector);
58
+ for (const el of candidates) {
59
+ el.focus();
60
+ if (document.activeElement === el) return true;
61
+ }
62
+ return false;
63
+ })`;
64
+ // `DOM.focus` only works on natively focusable elements. UIs that wrap real
65
+ // inputs in styled containers (Slack composer, Linear comments, Notion blocks,
66
+ // Canva pickers, MUI/Chakra/Mantine TextField) often expose the wrapper as the
67
+ // accessible "ref" — focusing it throws "Element is not focusable". When that
68
+ // happens, walk the subtree for the first focusable descendant.
53
69
  export async function focusNode(cdp, sessionId, backendNodeId) {
54
- await cdp.send('DOM.focus', { backendNodeId }, sessionId);
70
+ try {
71
+ await cdp.send('DOM.focus', { backendNodeId }, sessionId);
72
+ return;
73
+ }
74
+ catch (err) {
75
+ const focused = await focusFirstFocusableDescendant(cdp, sessionId, backendNodeId);
76
+ if (!focused)
77
+ throw err;
78
+ }
79
+ }
80
+ async function focusFirstFocusableDescendant(cdp, sessionId, backendNodeId) {
81
+ const { object } = await cdp.send('DOM.resolveNode', { backendNodeId }, sessionId);
82
+ if (!object.objectId)
83
+ return false;
84
+ const objectId = object.objectId;
85
+ try {
86
+ const { result } = await cdp.send('Runtime.callFunctionOn', { objectId, functionDeclaration: FOCUS_DESCENDANT_FN, returnByValue: true }, sessionId);
87
+ return result.value === true;
88
+ }
89
+ finally {
90
+ await cdp.send('Runtime.releaseObject', { objectId }, sessionId);
91
+ }
55
92
  }
@@ -164,11 +164,17 @@ export class BrowserIPCServer {
164
164
  if (!request.task) {
165
165
  return { ok: false, error: 'Task required' };
166
166
  }
167
- const { refs } = await this.service.refs(request.task, request.tabId, {
167
+ const { refs, nodeMap } = await this.service.refs(request.task, request.tabId, {
168
168
  interactive: request.interactive ?? true,
169
169
  limit: request.limit ?? 500,
170
170
  });
171
- return { ok: true, refs };
171
+ const nodes = Array.from(nodeMap.values()).map(n => {
172
+ const entry = { ref: n.ref, role: n.role, name: n.name, attrs: n.attrs };
173
+ if (n.editor !== undefined)
174
+ entry.editor = n.editor;
175
+ return entry;
176
+ });
177
+ return { ok: true, refs, nodes };
172
178
  }
173
179
  case 'click': {
174
180
  if (!request.task || request.ref === undefined) {
@@ -181,7 +187,7 @@ export class BrowserIPCServer {
181
187
  if (!request.task || request.ref === undefined || !request.text) {
182
188
  return { ok: false, error: 'Task, ref, and text required' };
183
189
  }
184
- await this.service.type(request.task, request.ref, request.text, request.tabId);
190
+ await this.service.type(request.task, request.ref, request.text, request.tabId, request.clear);
185
191
  return { ok: true };
186
192
  }
187
193
  case 'press': {
@@ -10,6 +10,7 @@ export interface RefNode {
10
10
  name: string;
11
11
  attrs: string[];
12
12
  backendNodeId?: number;
13
+ editor?: string;
13
14
  }
14
15
  export declare function getRefs(cdp: CDPClient, sessionId: string, opts?: RefOpts): Promise<{
15
16
  refs: string;
@@ -1,3 +1,31 @@
1
+ const EDITOR_DETECT_FN = `(function() {
2
+ let el = this;
3
+ for (let i = 0; i < 5; i++) {
4
+ if (!el || el === document.documentElement) break;
5
+ if (el.hasAttribute && el.hasAttribute('data-lexical-editor')) return 'lexical';
6
+ if (el.classList && el.classList.contains('ProseMirror')) return 'prosemirror';
7
+ if (el.hasAttribute && el.hasAttribute('data-slate-editor')) return 'slate';
8
+ if (el.classList && Array.from(el.classList).some(function(c) { return /^DraftEditor-/.test(c); })) return 'draft';
9
+ if (el.classList && el.classList.contains('ql-editor')) return 'quill';
10
+ if (el.classList && el.classList.contains('ck-editor__editable')) return 'ckeditor5';
11
+ if (el.tagName === 'TRIX-EDITOR') return 'trix';
12
+ el = el.parentElement;
13
+ }
14
+ return null;
15
+ })`;
16
+ async function detectEditorForNode(cdp, sessionId, backendNodeId) {
17
+ const { object } = await cdp.send('DOM.resolveNode', { backendNodeId }, sessionId);
18
+ if (!object.objectId)
19
+ return undefined;
20
+ const objectId = object.objectId;
21
+ try {
22
+ const { result } = await cdp.send('Runtime.callFunctionOn', { objectId, functionDeclaration: EDITOR_DETECT_FN, returnByValue: true }, sessionId);
23
+ return result.value ?? undefined;
24
+ }
25
+ finally {
26
+ await cdp.send('Runtime.releaseObject', { objectId }, sessionId);
27
+ }
28
+ }
1
29
  const INTERACTIVE_ROLES = new Set([
2
30
  'button',
3
31
  'link',
@@ -59,12 +87,18 @@ export async function getRefs(cdp, sessionId, opts = {}) {
59
87
  attrs,
60
88
  backendNodeId: node.backendDOMNodeId,
61
89
  };
90
+ if (role === 'textbox' && node.backendDOMNodeId) {
91
+ const editor = await detectEditorForNode(cdp, sessionId, node.backendDOMNodeId);
92
+ if (editor)
93
+ refNode.editor = editor;
94
+ }
62
95
  nodeMap.set(ref, refNode);
63
96
  const attrStr = attrs.length > 0 ? ` [${attrs.join('] [')}]` : '';
97
+ const editorStr = refNode.editor ? ` [editor=${refNode.editor}]` : '';
64
98
  const nameStr = name ? ` "${truncate(name, 50)}"` : '';
65
99
  const line = compact
66
- ? `${role}${nameStr} [ref=${ref}]${attrStr}`
67
- : `- ${role}${nameStr} [ref=${ref}]${attrStr}`;
100
+ ? `${role}${nameStr} [ref=${ref}]${attrStr}${editorStr}`
101
+ : `- ${role}${nameStr} [ref=${ref}]${attrStr}${editorStr}`;
68
102
  lines.push(line);
69
103
  }
70
104
  return { refs: lines.join('\n'), nodeMap };
@@ -99,7 +99,7 @@ export declare class BrowserService {
99
99
  nodeMap: Map<number, RefNode>;
100
100
  }>;
101
101
  click(taskId: string, ref: number, tabHint?: string): Promise<void>;
102
- type(taskId: string, ref: number, text: string, tabHint?: string): Promise<void>;
102
+ type(taskId: string, ref: number, text: string, tabHint?: string, clear?: boolean): Promise<void>;
103
103
  press(taskId: string, key: string, tabHint?: string): Promise<void>;
104
104
  hover(taskId: string, ref: number, tabHint?: string): Promise<void>;
105
105
  scroll(taskId: string, deltaX: number, deltaY: number, atX?: number, atY?: number, tabHint?: string): Promise<void>;
@@ -8,6 +8,7 @@ import { connectSSH } from './drivers/ssh.js';
8
8
  import { generateTaskId, generateShortId, generateFunName, } from './types.js';
9
9
  import { getRefs, resolveRefToCoords } from './refs.js';
10
10
  import { clickAtCoords, hoverAtCoords, scrollAtCoords, typeText, pressKey, focusNode } from './input.js';
11
+ import { typeEditorText } from './editor.js';
11
12
  import { emit } from '../events.js';
12
13
  /**
13
14
  * Parse a `targetFilter` string into its kind + value, or return `null`
@@ -502,7 +503,7 @@ export class BrowserService {
502
503
  const { x, y } = await resolveRefToCoords(conn.cdp, sessionId, nodeMap, ref);
503
504
  await clickAtCoords(conn.cdp, sessionId, x, y);
504
505
  }
505
- async type(taskId, ref, text, tabHint) {
506
+ async type(taskId, ref, text, tabHint, clear) {
506
507
  const { conn, task } = await this.findTask(taskId);
507
508
  const shortId = tabHint ? await this.resolveTabHint(conn, task, tabHint) : this.resolveCurrentTab(task);
508
509
  const cdpTargetId = this.getCdpTargetId(task, shortId);
@@ -514,10 +515,15 @@ export class BrowserService {
514
515
  const node = nodeMap.get(ref);
515
516
  if (!node)
516
517
  throw new Error(`Ref ${ref} not found`);
517
- if (node.backendNodeId) {
518
- await focusNode(conn.cdp, sessionId, node.backendNodeId);
518
+ if (node.editor) {
519
+ await typeEditorText(conn.cdp, sessionId, node, text, clear);
520
+ }
521
+ else {
522
+ if (node.backendNodeId) {
523
+ await focusNode(conn.cdp, sessionId, node.backendNodeId);
524
+ }
525
+ await typeText(conn.cdp, sessionId, text);
519
526
  }
520
- await typeText(conn.cdp, sessionId, text);
521
527
  }
522
528
  async press(taskId, key, tabHint) {
523
529
  const { conn, task } = await this.findTask(taskId);
@@ -129,6 +129,7 @@ export interface IPCResponse {
129
129
  result?: unknown;
130
130
  path?: string;
131
131
  refs?: string;
132
+ nodes?: RefNodeJson[];
132
133
  port?: number;
133
134
  pid?: number;
134
135
  logs?: ConsoleEntry[];
@@ -160,6 +161,13 @@ export interface NetworkRequest {
160
161
  mimeType?: string;
161
162
  timestamp: number;
162
163
  }
164
+ export interface RefNodeJson {
165
+ ref: number;
166
+ role: string;
167
+ name: string;
168
+ attrs: string[];
169
+ editor?: string;
170
+ }
163
171
  export interface DeviceDescriptor {
164
172
  width: number;
165
173
  height: number;
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@phnx-labs/agents-cli",
3
- "version": "1.17.3",
3
+ "version": "1.17.4",
4
4
  "description": "One CLI for all your AI coding agents - versions, config, cloud dispatch, sessions, and teams",
5
5
  "type": "module",
6
6
  "main": "dist/index.js",
@@ -87,6 +87,7 @@
87
87
  "@types/diff": "^6.0.0",
88
88
  "@types/marked-terminal": "^6.1.1",
89
89
  "@types/node": "^22.0.0",
90
+ "playwright": "^1.44.0",
90
91
  "tsx": "^4.19.0",
91
92
  "typescript": "^5.5.0",
92
93
  "vitest": "^2.0.0"