ravensight-playtest 0.2.8 → 0.3.0

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.
@@ -2,6 +2,7 @@ import { mkdir, rename, stat, readdir } from 'node:fs/promises';
2
2
  import path from 'node:path';
3
3
  import {
4
4
  DriverError,
5
+ STALE_REF_MESSAGE,
5
6
  THIN_OBSERVATION_CHARS,
6
7
  observationHash,
7
8
  screenshotName,
@@ -63,35 +64,337 @@ const INTERACTIVE_ROLES = new Set([
63
64
  'canvas'
64
65
  ]);
65
66
 
67
+ /**
68
+ * Landmark roles, and how a duplicate's hint names them when they carry no
69
+ * name of their own.
70
+ */
71
+ const LANDMARK_LABELS = Object.freeze({
72
+ banner: 'header',
73
+ navigation: 'navigation',
74
+ main: 'main',
75
+ contentinfo: 'footer',
76
+ complementary: 'sidebar',
77
+ region: 'region',
78
+ dialog: 'dialog',
79
+ alertdialog: 'dialog',
80
+ form: 'form',
81
+ search: 'search',
82
+ toolbar: 'toolbar',
83
+ tablist: 'tab list',
84
+ menu: 'menu',
85
+ menubar: 'menu bar',
86
+ group: 'group'
87
+ });
88
+
89
+ /** Attributes the snapshot prints that tell a persona nothing. */
90
+ const NOISE_ATTRIBUTE = / \[(?:ref=[^\]]*|box=[^\]]*|cursor=[^\]]*|active)\]/g;
91
+
92
+ /** The `[ref=e12]` Playwright's AI mode snapshot prints. */
93
+ const REF_ATTRIBUTE = /\[ref=(e\d+)\]/;
94
+
95
+ /** The `[box=x,y,width,height]` a snapshot taken with `boxes` prints. */
96
+ const BOX_ATTRIBUTE = /\[box=(-?[\d.]+),(-?[\d.]+),(-?[\d.]+),(-?[\d.]+)\]/;
97
+
66
98
  /**
67
99
  * Keep the nodes a player could act on or read, and drop the structure in
68
100
  * between. An aria snapshot prints every wrapper (`- generic:`, `- list:`, a
69
101
  * nameless `- img`) and the attribute lines under links (`- /url: ...`), none
70
102
  * of which tell a persona anything, and all of which cost tokens every turn.
71
103
  *
72
- * Indentation is dropped too: the persona targets nodes by role and name, not
73
- * by depth. Text that is not an aria snapshot (the innerText fallback for a
74
- * thin page) is returned unchanged.
104
+ * Indentation is dropped too: the persona targets nodes by ref, role and
105
+ * name, not by depth. Text that is not an aria snapshot (the innerText
106
+ * fallback for a thin page) is returned unchanged.
75
107
  *
76
108
  * @param {string} text
109
+ * @param {{viewport?: {width: number, height: number}}} [options]
77
110
  * @returns {string}
78
111
  */
79
- export function compactAriaSnapshot(text) {
112
+ export function compactAriaSnapshot(text, options = {}) {
113
+ const nodes = compactAriaNodes(text, options);
114
+ return nodes === null ? String(text == null ? '' : text) : nodes.map(node => node.line).join('\n');
115
+ }
116
+
117
+ /**
118
+ * The centre of a box, or null when the node is not visible in the viewport.
119
+ * Rounded, and bounded by the viewport, so a coordinate the persona copies is
120
+ * always one the mouse can reach.
121
+ *
122
+ * @param {Array<number>|null} box [x, y, width, height]
123
+ * @param {{width: number, height: number}} viewport
124
+ * @returns {{x: number, y: number}|null}
125
+ */
126
+ export function visibleCentre(box, viewport = DEFAULT_VIEWPORT) {
127
+ if (!box) return null;
128
+ const [x, y, width, height] = box;
129
+ if (!(width > 0 && height > 0)) return null;
130
+ const cx = Math.round(x + width / 2);
131
+ const cy = Math.round(y + height / 2);
132
+ if (cx < 0 || cy < 0 || cx > viewport.width || cy > viewport.height) return null;
133
+ return { x: cx, y: cy };
134
+ }
135
+
136
+ /**
137
+ * The kept nodes of an aria snapshot with the depth each one had, or null for
138
+ * text that is not an aria snapshot. The depth is not shown to the model; it
139
+ * is what tells repeated siblings apart from a panel that follows them.
140
+ *
141
+ * An interactive node from Playwright's AI mode snapshot carries its short ref
142
+ * (`[e12]`), which `click` and `type` accept as a target, and, when it is on
143
+ * screen, its centre `@(x,y)`. Two interactive nodes with the same role and
144
+ * name each get a hint naming the nearest landmark or named container around
145
+ * them ("in header", "in main"), so the persona can tell them apart.
146
+ *
147
+ * `hashLine` is the line without the centre: what "changed" and "new since
148
+ * the last screen" compare, so a scroll or an animation does not read as a
149
+ * new screen.
150
+ *
151
+ * @param {string} text
152
+ * @param {{viewport?: {width: number, height: number}}} [options]
153
+ * @returns {Array<{depth: number, line: string, hashLine: string, body: string, role: string, name: string|null, ref: string|null, centre: Object|null, hint: string|null, key: string|null, duplicate: boolean}>|null}
154
+ */
155
+ export function compactAriaNodes(text, { viewport = DEFAULT_VIEWPORT } = {}) {
80
156
  const value = String(text == null ? '' : text);
81
157
  const lines = value.split('\n');
82
- if (!lines.some(line => /^\s*- /.test(line))) return value;
158
+ if (!lines.some(line => /^\s*- /.test(line))) return null;
83
159
  const kept = [];
160
+ const ancestors = [];
84
161
  for (const line of lines) {
85
162
  const trimmed = line.trim();
86
163
  if (!trimmed.startsWith('- ')) continue;
87
- const body = unquoteKey(trimmed.slice(2));
88
- if (body.startsWith('/')) continue;
89
- const role = (body.match(/^([a-z]+)/) || [])[1] || '';
164
+ const depth = line.length - line.trimStart().length;
165
+ const raw = unquoteKey(trimmed.slice(2));
166
+ while (ancestors.length > 0 && ancestors[ancestors.length - 1].depth >= depth) ancestors.pop();
167
+ if (raw.startsWith('/')) continue;
168
+ const role = (raw.match(/^([a-z]+)/) || [])[1] || '';
169
+ const name = (raw.match(/^[a-z]+ "([^"]*)"/) || [])[1] ?? null;
170
+ const ref = (raw.match(REF_ATTRIBUTE) || [])[1] || null;
171
+ const boxMatch = raw.match(BOX_ATTRIBUTE);
172
+ const box = boxMatch ? boxMatch.slice(1, 5).map(Number) : null;
173
+ const body = raw.replace(NOISE_ATTRIBUTE, '');
174
+ ancestors.push({ depth, role, name });
90
175
  const named = /^[a-z]+ "[^"]*"/.test(body);
91
176
  const inlineText = /^[a-z]+(?: "[^"]*")?(?: \[[^\]]*\])*: \S/.test(body);
92
- if (INTERACTIVE_ROLES.has(role) || named || inlineText) kept.push(`- ${body}`);
177
+ const interactive = INTERACTIVE_ROLES.has(role);
178
+ if (!(interactive || named || inlineText)) continue;
179
+ const container = ancestors.slice(0, -1).reverse().find(entry => LANDMARK_LABELS[entry.role] || entry.name);
180
+ const hint = container
181
+ ? (container.name ? `${container.name}${LANDMARK_LABELS[container.role] && container.role !== 'region' ? ` ${LANDMARK_LABELS[container.role]}` : ''}` : LANDMARK_LABELS[container.role])
182
+ : null;
183
+ kept.push({
184
+ depth,
185
+ body,
186
+ role,
187
+ name,
188
+ ref: interactive ? ref : null,
189
+ centre: interactive ? visibleCentre(box, viewport) : null,
190
+ hint,
191
+ key: interactive && name ? `${role}|${name}` : null,
192
+ duplicate: false,
193
+ line: '',
194
+ hashLine: ''
195
+ });
93
196
  }
94
- return kept.join('\n');
197
+ const counts = new Map();
198
+ for (const node of kept) if (node.key) counts.set(node.key, (counts.get(node.key) || 0) + 1);
199
+ for (const node of kept) {
200
+ node.duplicate = Boolean(node.key) && counts.get(node.key) > 1;
201
+ const split = node.body.match(/^([a-z]+(?: "[^"]*")?(?: \[[^\]]*\])*)(.*)$/s) || [null, node.body, ''];
202
+ const head = split[1];
203
+ const tail = split[2];
204
+ const refPart = node.ref ? ` [${node.ref}]` : '';
205
+ const hintPart = node.duplicate && node.hint ? ` (in ${node.hint})` : '';
206
+ const centrePart = node.centre ? ` @(${node.centre.x},${node.centre.y})` : '';
207
+ node.hashLine = `- ${head}${refPart}${hintPart}${tail}`;
208
+ node.line = `- ${head}${refPart}${centrePart}${hintPart}${tail}`;
209
+ }
210
+ return kept;
211
+ }
212
+
213
+ /** Roles a player types into or sets, kept first when an observation has to be cut. */
214
+ const INPUT_ROLES = new Set(['textbox', 'searchbox', 'combobox', 'checkbox', 'radio', 'slider', 'spinbutton', 'switch', 'listbox']);
215
+
216
+ /** The share of interactive nodes, counted from the end of the page, kept ahead of the middle when a page is cut. */
217
+ export const TAIL_FRACTION = 0.1;
218
+
219
+ /** The tail window is never smaller than this many interactive nodes. */
220
+ const TAIL_MIN = 3;
221
+
222
+ /** A run of this many look-alike siblings or more is collapsed when the page is over the cap. */
223
+ export const COLLAPSE_MIN_RUN = 4;
224
+
225
+ /** How many of a collapsed run are still shown, so the model sees what the items look like. */
226
+ export const COLLAPSE_KEEP = 2;
227
+
228
+ /**
229
+ * The shape of one node, without the words that make it unique: its role,
230
+ * its state attributes, whether it has a name, whether the name carries
231
+ * digits and how many separators. Two school cards share a shape; the card
232
+ * list and a "TAKE THE JOB" button after it do not.
233
+ *
234
+ * @param {string} body
235
+ * @returns {string}
236
+ */
237
+ function nodeShape(body) {
238
+ const role = (body.match(/^([a-z]+)/) || [])[1] || '';
239
+ const name = (body.match(/^[a-z]+ "([^"]*)"/) || [])[1];
240
+ const attributes = [...body.matchAll(/\[([a-z-]+)(?:=[^\]]*)?\]/g)].map(match => match[1]).filter(key => key !== 'ref').join(',');
241
+ const nameShape = name === undefined ? '' : `n${/\d/.test(name) ? 'd' : ''}${(name.match(/[·•|/]/g) || []).length}`;
242
+ const inline = /: \S/.test(body.replace(/^[a-z]+(?: "[^"]*")?/, '')) ? 't' : '';
243
+ return `${role}|${nameShape}|${attributes}|${inline}`;
244
+ }
245
+
246
+ /**
247
+ * Collapse runs of look-alike siblings (a node and its kept descendants,
248
+ * compared by shape) into the first {@link COLLAPSE_KEEP} plus one line that
249
+ * says how many more there are.
250
+ *
251
+ * @param {Array<{depth: number, line: string, body: string, role: string}>} nodes
252
+ * @returns {{entries: Array<{line: string, role: string, marker?: boolean}>, collapsed: number}}
253
+ */
254
+ export function collapseRepeatedSiblings(nodes) {
255
+ const keyCounts = new Map();
256
+ for (const node of nodes) if (node.key) keyCounts.set(node.key, (keyCounts.get(node.key) || 0) + 1);
257
+ const unitEnd = start => {
258
+ let end = start + 1;
259
+ while (end < nodes.length && nodes[end].depth > nodes[start].depth) end += 1;
260
+ return end;
261
+ };
262
+ const signature = (start, end) => nodes.slice(start, end).map(node => `${node.depth - nodes[start].depth}:${nodeShape(node.body)}`).join(';');
263
+ const entries = [];
264
+ let collapsed = 0;
265
+ let i = 0;
266
+ while (i < nodes.length) {
267
+ const units = [[i, unitEnd(i)]];
268
+ const sig = signature(i, units[0][1]);
269
+ let next = units[0][1];
270
+ while (next < nodes.length && nodes[next].depth === nodes[i].depth) {
271
+ const end = unitEnd(next);
272
+ if (signature(next, end) !== sig) break;
273
+ units.push([next, end]);
274
+ next = end;
275
+ }
276
+ if (units.length >= COLLAPSE_MIN_RUN) {
277
+ // A node whose role and name also appear outside this run is never
278
+ // hidden in the count: two buttons called "Run the week", one in the
279
+ // header and one in the main panel, both stay visible with their refs.
280
+ const inRun = new Map();
281
+ for (const node of nodes.slice(i, next)) if (node.key) inRun.set(node.key, (inRun.get(node.key) || 0) + 1);
282
+ // Only the first in-run twin of a name that also appears outside the
283
+ // run is kept, beside the first two items. One "View" button in the
284
+ // header must not stop a list of 249 "View" cards from collapsing.
285
+ const outside = key => key && (keyCounts.get(key) || 0) > (inRun.get(key) || 0);
286
+ const shownKeys = new Set();
287
+ let more = 0;
288
+ units.forEach((unit, index) => {
289
+ const unitNodes = nodes.slice(unit[0], unit[1]);
290
+ const twins = unitNodes.map(node => node.key).filter(key => outside(key) && !shownKeys.has(key));
291
+ if (index < COLLAPSE_KEEP || twins.length > 0) {
292
+ for (const node of unitNodes) {
293
+ entries.push({ line: node.line, hashLine: node.hashLine, role: node.role, key: node.key });
294
+ if (node.key) shownKeys.add(node.key);
295
+ }
296
+ } else {
297
+ more += 1;
298
+ }
299
+ });
300
+ if (more > 0) {
301
+ collapsed += more;
302
+ entries.push({ line: `- ... and ${more} more like it (${nodes[i].role || 'items'}; target one by its name)`, role: '', marker: true });
303
+ }
304
+ i = next;
305
+ } else {
306
+ entries.push({ line: nodes[i].line, hashLine: nodes[i].hashLine, role: nodes[i].role, key: nodes[i].key });
307
+ i += 1;
308
+ }
309
+ }
310
+ return { entries, collapsed };
311
+ }
312
+
313
+ /**
314
+ * Fit a compacted aria snapshot under the web ceiling without losing the
315
+ * part of the page the persona has to act on.
316
+ *
317
+ * Under the cap nothing changes. Over it, in this order: interactive nodes
318
+ * are what is kept (the last tenth of them on the page first, since that is
319
+ * where a form after a long list sits, then inputs, then buttons and links,
320
+ * then any role with a name); runs of look-alike siblings collapse to their
321
+ * first two and a count; nodes that are new since the previous observation
322
+ * come next; then
323
+ * the rest, and only what still does not fit is trimmed. Kept lines stay in
324
+ * page order, and a first line says how many items were collapsed and how
325
+ * many lines were left out, so the model knows the screen was cut.
326
+ *
327
+ * @param {string} text the raw aria snapshot
328
+ * @param {{max?: number, previous?: Set<string>|null}} [options]
329
+ * @returns {{text: string, lines: Array<string>}|null} null when the text is not an aria snapshot
330
+ */
331
+ export function fitAriaSnapshot(text, { max = WEB_OBSERVATION_MAX_CHARS, previous = null, viewport = DEFAULT_VIEWPORT } = {}) {
332
+ const nodes = compactAriaNodes(text, { viewport });
333
+ if (nodes === null) return null;
334
+ const lines = nodes.map(node => node.hashLine);
335
+ const full = nodes.map(node => node.line).join('\n');
336
+ if (full.length <= max) return { text: full, lines, nodes };
337
+
338
+ const { entries, collapsed } = collapseRepeatedSiblings(nodes);
339
+ const header = (left, out) => {
340
+ const parts = [];
341
+ if (left > 0) parts.push(`${left} similar items collapsed`);
342
+ if (out > 0) parts.push(`${out} lines left out to fit`);
343
+ return parts.length > 0 ? `[${parts.join('; ')}]` : '';
344
+ };
345
+ const collapsedText = entries.map(entry => entry.line).join('\n');
346
+ const withHeader = `${header(collapsed, 0)}\n${collapsedText}`;
347
+ if (withHeader.length <= max) return { text: withHeader, lines, nodes };
348
+
349
+ const rank = entry => {
350
+ if (entry.marker) return 0;
351
+ if (INPUT_ROLES.has(entry.role)) return 1;
352
+ if (entry.role === 'button' || entry.role === 'link') return 2;
353
+ if (INTERACTIVE_ROLES.has(entry.role) || /^- [a-z]+ "/.test(entry.line)) return 3;
354
+ return 5;
355
+ };
356
+ const isNew = entry => Boolean(previous) && !entry.marker && !previous.has(entry.hashLine || entry.line);
357
+ // The last interactive nodes on the page are usually its form or its
358
+ // confirm button (a panel after a list). When the list could not be
359
+ // collapsed they would otherwise lose to the list on page order, so the
360
+ // tail window, the final tenth of interactive nodes, ranks above the middle.
361
+ const interactive = entries.map((entry, index) => ({ entry, index })).filter(({ entry }) => !entry.marker && rank(entry) < 5);
362
+ const tailSize = Math.max(TAIL_MIN, Math.ceil(interactive.length * TAIL_FRACTION));
363
+ const tail = new Set(interactive.slice(-tailSize).map(({ index }) => index));
364
+ const tierOf = (entry, index) => {
365
+ if (entry.marker) return 0;
366
+ if (tail.has(index)) return 1 + (isNew(entry) ? 0 : 0.5);
367
+ if (rank(entry) < 5) return 2 + rank(entry) * 2 + (isNew(entry) ? 0 : 1);
368
+ return isNew(entry) ? 20 : 21;
369
+ };
370
+ const order = entries
371
+ .map((entry, index) => ({ entry, index, tier: tierOf(entry, index) }))
372
+ .sort((a, b) => a.tier - b.tier || a.index - b.index);
373
+ // Room for the header line, whose figures are not known until the end.
374
+ const budget = max - 80;
375
+ const chosen = new Set();
376
+ let used = 0;
377
+ for (const candidate of order) {
378
+ const cost = candidate.entry.line.length + 1;
379
+ if (used + cost > budget) continue;
380
+ chosen.add(candidate.index);
381
+ used += cost;
382
+ }
383
+ // Duplicates by name travel together: showing one "Run the week" and not
384
+ // the other is how a persona ends up clicking a name that matches twice.
385
+ const chosenKeys = new Set(entries.filter((entry, index) => chosen.has(index) && entry.key).map(entry => entry.key));
386
+ // Bounded by the cap's own slack: a page whose every card shares a name
387
+ // is the trim's job, not this one's.
388
+ entries.forEach((entry, index) => {
389
+ if (!entry.key || !chosenKeys.has(entry.key) || chosen.has(index)) return;
390
+ const cost = entry.line.length + 1;
391
+ if (used + cost > budget) return;
392
+ chosen.add(index);
393
+ used += cost;
394
+ });
395
+ const kept = entries.filter((entry, index) => chosen.has(index)).map(entry => entry.line);
396
+ const out = entries.length - kept.length;
397
+ return { text: truncateObservation(`${header(collapsed, out)}\n${kept.join('\n')}`, max), lines, nodes };
95
398
  }
96
399
 
97
400
  /**
@@ -253,6 +556,78 @@ export function isLoopback(hostname) {
253
556
  return LOOPBACK_HOSTS.includes(hostname) || /^127\.\d{1,3}\.\d{1,3}\.\d{1,3}$/.test(hostname);
254
557
  }
255
558
 
559
+ /**
560
+ * A target written as locator code rather than as something on screen. A
561
+ * persona that writes `getByTestId('strip-advance')` means a selector, and
562
+ * searching the page for that literal text waits the whole action timeout
563
+ * for nothing.
564
+ */
565
+ export const LOCATOR_CODE = /^\s*(?:getBy|page\.|locator\(|css=|xpath=|\/\/)/;
566
+
567
+ /** What the persona is told when it writes locator code as a target. */
568
+ export const LOCATOR_CODE_MESSAGE = 'targets are refs or visible text, not locator code';
569
+
570
+ export { STALE_REF_MESSAGE };
571
+
572
+ /** A ref as the snapshot prints it. */
573
+ export const REF_PATTERN = /^e\d{1,6}$/;
574
+
575
+ /**
576
+ * A failure on the driver's side, not the game's: a target that names two
577
+ * elements, a target that is not a target, a ref that is not on screen. The
578
+ * transcript records these as `driver_error`, and the report step is told
579
+ * they are never evidence of a game bug.
580
+ */
581
+ export class TargetError extends Error {
582
+ constructor(message) {
583
+ super(message);
584
+ this.name = 'TargetError';
585
+ this.driverSide = true;
586
+ }
587
+ }
588
+
589
+ /**
590
+ * The candidates a target could mean, from the last snapshot: the
591
+ * interactive nodes whose name is the target (or whose role and name are,
592
+ * for `button "Start"`), with their refs and hints.
593
+ *
594
+ * @param {Array<Object>|null} nodes
595
+ * @param {string} target
596
+ * @returns {Array<Object>}
597
+ */
598
+ export function candidatesFor(nodes, target) {
599
+ if (!Array.isArray(nodes)) return [];
600
+ const value = String(target).trim();
601
+ const roleAndName = value.match(/^([a-z]+)\s+"([^"]*)"$/);
602
+ return nodes.filter(node => node.key && node.ref && (roleAndName
603
+ ? node.role === roleAndName[1] && node.name === roleAndName[2]
604
+ : String(node.name).trim().toLowerCase() === value.toLowerCase()));
605
+ }
606
+
607
+ /** The most candidates an ambiguity answer lists before "and N more". */
608
+ export const AMBIGUITY_LIST_MAX = 5;
609
+
610
+ /**
611
+ * "2 elements match "Run the week"; use ref e3 (in header) or e5 (in main)".
612
+ *
613
+ * @param {string} target
614
+ * @param {Array<Object>} candidates
615
+ * @param {number} [count] how many Playwright said matched, when known
616
+ * @returns {string}
617
+ */
618
+ export function ambiguityMessage(target, candidates, count = null) {
619
+ const total = Math.max(count || 0, candidates.length);
620
+ const shown = candidates.slice(0, AMBIGUITY_LIST_MAX);
621
+ const listed = shown.map(node => `${node.ref}${node.hint ? ` (in ${node.hint})` : ''}`);
622
+ const more = candidates.length - shown.length;
623
+ const choice = more > 0
624
+ ? `${listed.join(', ')} and ${more} more`
625
+ : listed.length > 1 ? `${listed.slice(0, -1).join(', ')} or ${listed[listed.length - 1]}` : listed[0];
626
+ return listed.length > 0
627
+ ? `${total} elements match ${JSON.stringify(String(target))}; use ref ${choice}`
628
+ : `${total} elements match ${JSON.stringify(String(target))}; take a snapshot and use the ref of the one you mean`;
629
+ }
630
+
256
631
  /**
257
632
  * Turn what the model wrote into a Playwright locator.
258
633
  *
@@ -268,6 +643,7 @@ export function isLoopback(hostname) {
268
643
  */
269
644
  export function resolveLocator(page, target) {
270
645
  const value = String(target).trim();
646
+ if (LOCATOR_CODE.test(value)) throw new TargetError(LOCATOR_CODE_MESSAGE);
271
647
  if (KNOWN_SELECTOR_ENGINES.test(value)) return page.locator(value);
272
648
 
273
649
  // The name group cannot contain a quote (`[^"]*`), so there is nothing to
@@ -281,6 +657,16 @@ export function resolveLocator(page, target) {
281
657
  return page.getByText(value, { exact: false }).first();
282
658
  }
283
659
 
660
+ /**
661
+ * Give the document a random token. Runs in the browser at the start of every
662
+ * new document, so a reload or a navigation gets a new one and a single page
663
+ * app's route change keeps it. Serialised by Playwright, so it must not close
664
+ * over anything.
665
+ */
666
+ function markDocument() {
667
+ globalThis.__ravensightDoc = `${Math.random().toString(36).slice(2)}${Date.now().toString(36)}`;
668
+ }
669
+
284
670
  /**
285
671
  * Hand the run's playtest identity to the page. Runs in the browser, before
286
672
  * any of the game's own scripts, so the Ravensight JS SDK finds it at
@@ -323,6 +709,17 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
323
709
  let shots = 0;
324
710
  let lastHash = null;
325
711
  let lastObservedHash = null;
712
+ /** The compacted lines of the last observation, so a cut can prefer what is new. */
713
+ let lastObservedLines = null;
714
+ /** The kept nodes of the last observation: refs, names and hints, for a target that matches twice. */
715
+ let lastNodes = null;
716
+ let viewportSize = DEFAULT_VIEWPORT;
717
+ // Refs are numbered per document: after a navigation or a reload, e12 can
718
+ // be a different element. Each document gets a random token from an init
719
+ // script; a pushState or hash route change keeps it, a new document does
720
+ // not. The token the last observation (and so its refs) came from is
721
+ // compared with the current one before any ref is used.
722
+ let lastNodesDocument = null;
326
723
  const consoleBuffer = [];
327
724
 
328
725
  /** @returns {any} */
@@ -341,7 +738,17 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
341
738
  let text = '';
342
739
  let aria = true;
343
740
  const body = current.locator('body');
344
- if (typeof body.ariaSnapshot === 'function') {
741
+ // The page level AI mode snapshot carries a ref per element, which
742
+ // `aria-ref=e12` resolves, and each element's box. Older Playwright
743
+ // builds have neither, and the body snapshot is the fallback.
744
+ if (typeof current.ariaSnapshot === 'function') {
745
+ try {
746
+ text = await current.ariaSnapshot({ mode: 'ai', boxes: true, timeout: ACTION_TIMEOUT_MS });
747
+ } catch {
748
+ text = '';
749
+ }
750
+ }
751
+ if (!text && typeof body.ariaSnapshot === 'function') {
345
752
  try {
346
753
  text = await body.ariaSnapshot({ timeout: ACTION_TIMEOUT_MS });
347
754
  } catch {
@@ -367,13 +774,33 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
367
774
  }
368
775
 
369
776
  /**
370
- * What the model is shown for a raw snapshot. `observe` and `act` both use
371
- * this, so the two hashes compare like with like.
777
+ * What the model is shown for a raw snapshot, and the lines it was built
778
+ * from. A page over the web ceiling is fitted rather than cut at the
779
+ * character limit (see fitAriaSnapshot), so a panel after a long list
780
+ * still reaches the model.
372
781
  * @param {{text: string, aria: boolean}} snap
373
- * @returns {string}
782
+ * @returns {{text: string, lines: Array<string>|null}}
374
783
  */
375
784
  function shownText(snap) {
376
- return truncateObservation(snap.aria ? compactAriaSnapshot(snap.text) : snap.text, WEB_OBSERVATION_MAX_CHARS);
785
+ if (snap.aria) {
786
+ const fitted = fitAriaSnapshot(snap.text, { max: WEB_OBSERVATION_MAX_CHARS, previous: lastObservedLines ? new Set(lastObservedLines) : null, viewport: viewportSize });
787
+ if (fitted) return fitted;
788
+ }
789
+ return { text: truncateObservation(snap.text, WEB_OBSERVATION_MAX_CHARS), lines: null, nodes: null };
790
+ }
791
+
792
+ /**
793
+ * What `changed` and `unchanged` compare. For an aria snapshot that is the
794
+ * whole compacted tree, not the fitted text: the fit depends on the
795
+ * previous observation, so hashing it would read the same screen as
796
+ * changed. For the innerText fallback it is the truncated text, so a
797
+ * difference the model never saw is not a change.
798
+ * @param {{text: string, aria: boolean}} snap
799
+ * @param {{text: string, lines: Array<string>|null}} shown
800
+ * @returns {string}
801
+ */
802
+ function hashText(snap, shown) {
803
+ return shown.lines ? shown.lines.join('\n') : shown.text;
377
804
  }
378
805
 
379
806
  /**
@@ -391,7 +818,7 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
391
818
  */
392
819
  async function typeText(current, action) {
393
820
  const text = String(action.text == null ? '' : action.text);
394
- const field = action.target ? resolveLocator(current, action.target) : null;
821
+ const field = action.target || action.ref ? await targetFor(current, action) : null;
395
822
  if (action.replace) {
396
823
  try {
397
824
  await (field || current.locator('*:focus')).fill(text, { timeout: field ? ACTION_TIMEOUT_MS : FOCUSED_FILL_TIMEOUT_MS });
@@ -409,8 +836,75 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
409
836
  if (text) await current.keyboard.type(text, { delay: 20 });
410
837
  }
411
838
 
839
+ /**
840
+ * The locator a click or a typing target means. A ref is preferred and
841
+ * checked against the page; locator code is refused at once, with no wait;
842
+ * a name that matches more than one interactive node on the last screen is
843
+ * answered with the candidates and their refs rather than guessed.
844
+ *
845
+ * @param {any} current
846
+ * @param {{ref?: string, target?: string}} action
847
+ */
848
+ async function targetFor(current, action) {
849
+ if (action.ref !== undefined && action.ref !== null && action.ref !== '') {
850
+ const ref = String(action.ref).trim();
851
+ if (!REF_PATTERN.test(ref)) throw new TargetError(`a ref looks like e12, as the snapshot prints it; ${JSON.stringify(ref)} is not one`);
852
+ await assertSameDocument(current);
853
+ const locator = current.locator(`aria-ref=${ref}`);
854
+ if (typeof locator.count === 'function' && (await locator.count()) === 0) {
855
+ throw new TargetError(`ref ${ref} is not on the current screen; take a snapshot and use a ref from it`);
856
+ }
857
+ return locator;
858
+ }
859
+ const target = String(action.target == null ? '' : action.target);
860
+ const candidates = candidatesFor(lastNodes, target);
861
+ if (candidates.length > 1) throw new TargetError(ambiguityMessage(target, candidates));
862
+ if (LOCATOR_CODE.test(target)) {
863
+ // Text that only looks like code: a node on screen really is named
864
+ // that, so it is clicked by its ref rather than refused.
865
+ if (candidates.length === 1) {
866
+ await assertSameDocument(current);
867
+ return current.locator(`aria-ref=${candidates[0].ref}`);
868
+ }
869
+ throw new TargetError(LOCATOR_CODE_MESSAGE);
870
+ }
871
+ return resolveLocator(current, target);
872
+ }
873
+
874
+ /**
875
+ * The current document's token, read from the main frame only (a child
876
+ * frame has its own document and its own token), or null when the page
877
+ * cannot answer.
878
+ */
879
+ async function documentToken(current) {
880
+ try {
881
+ const token = await current.evaluate(() => globalThis.__ravensightDoc || null);
882
+ return typeof token === 'string' && token ? token : null;
883
+ } catch {
884
+ return null;
885
+ }
886
+ }
887
+
888
+ /** Refuse a ref when the document changed since the observation it came from. */
889
+ async function assertSameDocument(current) {
890
+ const now = await documentToken(current);
891
+ if (lastNodesDocument && now !== lastNodesDocument) throw new TargetError(STALE_REF_MESSAGE);
892
+ }
893
+
894
+ /** The hash of the screen as it is now, on observe()'s basis. */
895
+ async function currentHash(current) {
896
+ const snap = await snapshotText();
897
+ const after = { text: hashText(snap, shownText(snap)) };
898
+ if (String(snap.text).trim().length < THIN_OBSERVATION_CHARS) {
899
+ const frame = await current.screenshot({ type: 'png', timeout: ACTION_TIMEOUT_MS });
900
+ after.screenshotBase64 = frame.toString('base64');
901
+ }
902
+ return observationHash(after);
903
+ }
904
+
412
905
  return {
413
906
  name: 'playwright_web',
907
+ supportsRefs: true,
414
908
  editModifier,
415
909
  keepObservations: WEB_KEEP_OBSERVATIONS,
416
910
 
@@ -424,6 +918,7 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
424
918
  async launch({ target: url, runDir, recordVideo = false, viewport = DEFAULT_VIEWPORT }) {
425
919
  if (!url) throw new DriverError('playwright_web needs a target URL', { fatal: true });
426
920
  target = url;
921
+ viewportSize = viewport || DEFAULT_VIEWPORT;
427
922
  runDirectory = runDir;
428
923
  videoRequested = Boolean(recordVideo);
429
924
 
@@ -447,6 +942,9 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
447
942
  persona: playtest.persona || null
448
943
  });
449
944
  }
945
+ // One token per document: set again by every new document, kept by
946
+ // pushState and hash route changes, which do not run init scripts.
947
+ if (typeof context.addInitScript === 'function') await context.addInitScript(markDocument);
450
948
  page = await context.newPage();
451
949
  page.on('console', message => {
452
950
  if (message.type() === 'error') consoleBuffer.push({ type: 'console', text: message.text() });
@@ -468,9 +966,10 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
468
966
  step += 1;
469
967
  const snap = await snapshotText();
470
968
  const text = snap.text;
969
+ const shown = shownText(snap);
471
970
  const observation = {
472
971
  step,
473
- text: shownText(snap),
972
+ text: shown.text,
474
973
  url: current.url(),
475
974
  elapsedMs: Date.now() - started
476
975
  };
@@ -497,11 +996,13 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
497
996
  // read it as a blank screen, which is what it is.
498
997
  }
499
998
  }
500
- // Hash what was REPORTED, not the untruncated snapshot: `act` compares
501
- // against this, so two screens that differ only past the truncation
502
- // point would otherwise read as changed to `act` and identical to
503
- // anything reading the observation.
504
- lastHash = observationHash(observation);
999
+ // Hash what hashText says, which `act` hashes the same way: the whole
1000
+ // compacted tree for an aria snapshot (a change anywhere in it is shown,
1001
+ // because a fit keeps new nodes first), the truncated text otherwise.
1002
+ lastHash = observationHash({ text: hashText(snap, shown), screenshotBase64: observation.screenshotBase64 });
1003
+ lastObservedLines = shown.lines;
1004
+ lastNodes = shown.nodes || null;
1005
+ lastNodesDocument = await documentToken(current);
505
1006
  // The same screen twice is sent as a marker rather than in full
506
1007
  // (./tools.js), which is most of what a stuck persona would otherwise
507
1008
  // pay for.
@@ -527,11 +1028,11 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
527
1028
  if (action.submit) await current.keyboard.press('Enter');
528
1029
  break;
529
1030
  case 'click': {
530
- if (action.target && typeof action.target === 'object') {
1031
+ if (!action.ref && action.target && typeof action.target === 'object') {
531
1032
  await current.mouse.click(Number(action.target.x), Number(action.target.y));
532
1033
  break;
533
1034
  }
534
- await resolveLocator(current, action.target).click({ timeout: ACTION_TIMEOUT_MS });
1035
+ await (await targetFor(current, action)).click({ timeout: ACTION_TIMEOUT_MS });
535
1036
  break;
536
1037
  }
537
1038
  case 'navigate': {
@@ -549,20 +1050,44 @@ export function createWebDriver({ importer, headless = true, playtest = null, pl
549
1050
  return { ok: false, changed: false, error: `the playwright_web driver cannot do "${action.kind}"` };
550
1051
  }
551
1052
  } catch (error) {
552
- return { ok: false, changed: false, error: String(error && error.message ? error.message : error) };
1053
+ // Playwright colours its call log; the colour codes are noise here.
1054
+ const text = String(error && error.message ? error.message : error).replace(/\u001b\[[0-9;]*m/g, '');
1055
+ // Two elements under one name: Playwright's strict mode. Answered
1056
+ // with the candidates and their refs, so the next try can work.
1057
+ if (/strict mode violation/i.test(text)) {
1058
+ const count = Number((text.match(/resolved to (\d+) elements/) || [])[1]) || null;
1059
+ const candidates = action.target && typeof action.target === 'string' ? candidatesFor(lastNodes, action.target) : [];
1060
+ return { ok: false, changed: false, error: ambiguityMessage(action.target || action.ref || '', candidates, count), driverError: true };
1061
+ }
1062
+ if (error && error.driverSide) return { ok: false, changed: false, error: text, driverError: true };
1063
+ // A click the page kept intercepting is the game's: something sat on
1064
+ // top of the target and never went away. Named, never a driver error.
1065
+ const interceptor = text.match(/^\s*-?\s*(.+?)\s+(?:from .+? subtree\s+)?intercepts pointer events/m);
1066
+ const intercepted = interceptor ? interceptor[1] : null;
1067
+ if (/Timeout \d+ms exceeded/i.test(text) && intercepted) {
1068
+ return { ok: false, changed: false, error: `timed out: ${intercepted.slice(0, 160)} intercepts pointer events over the target`, driverError: false };
1069
+ }
1070
+ // A timeout with the page unchanged while resolving a click or typing
1071
+ // target means the target was never found or never actionable: the
1072
+ // driver's failure, not the game's. Never for navigate: a game that
1073
+ // never loads is a real blocker.
1074
+ if (/Timeout \d+ms exceeded/i.test(text) && (action.kind === 'click' || action.kind === 'type')) {
1075
+ let unchanged = false;
1076
+ try {
1077
+ unchanged = (await currentHash(current)) === before;
1078
+ } catch {
1079
+ unchanged = false;
1080
+ }
1081
+ return { ok: false, changed: false, error: text, driverError: unchanged };
1082
+ }
1083
+ return { ok: false, changed: false, error: text };
553
1084
  }
554
1085
 
555
1086
  let changed = false;
556
1087
  try {
557
- // The same truncation observe() applies, so the two hashes are
558
- // comparable. Without it a long screen is always "changed".
559
- const snap = await snapshotText();
560
- const after = { text: shownText(snap) };
561
- if (String(snap.text).trim().length < THIN_OBSERVATION_CHARS) {
562
- const frame = await current.screenshot({ type: 'png', timeout: ACTION_TIMEOUT_MS });
563
- after.screenshotBase64 = frame.toString('base64');
564
- }
565
- changed = observationHash(after) !== before;
1088
+ // The same basis observe() hashes, so the two are comparable.
1089
+ // Without it a long screen is always "changed".
1090
+ changed = (await currentHash(current)) !== before;
566
1091
  } catch {
567
1092
  changed = true;
568
1093
  }