dsh-builtin-browser 0.4.7 → 0.4.8
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 +28 -0
- package/README.en.md +62 -491
- package/README.md +44 -470
- package/client.js +10 -10
- package/cordis.patch.yml +4 -4
- package/desktop-bridge/plugin-browser-bridge.js +1 -1
- package/docs/README.md +8 -4
- package/docs/architecture.md +82 -11
- package/docs/development.md +38 -0
- package/docs/limitations.md +49 -0
- package/docs/tool-reference.md +42 -42
- package/docs/user-guide.md +93 -37
- package/docs/why-browser.md +3 -3
- package/lib/browser/types.d.ts +2 -2
- package/lib/browser-electron/entry.d.ts +2 -2
- package/lib/browser-electron/entry.js +1 -1
- package/lib/browser-electron/history-store.d.ts +1 -1
- package/lib/browser-electron/history-store.js +1 -1
- package/lib/browser-electron/host-main.js +3 -3
- package/lib/browser-electron/provider.d.ts +9 -9
- package/lib/browser-electron/provider.js +14 -14
- package/lib/browser-electron/remote-host.js +1 -1
- package/lib/browser-electron/settings-store.d.ts +4 -4
- package/lib/index.d.ts +1 -1
- package/lib/index.js +1 -1
- package/lib/tool-browser/element-fields.d.ts +4 -4
- package/lib/tool-browser/element-fields.js +6 -6
- package/lib/tool-browser/index.d.ts +6 -6
- package/lib/tool-browser/index.js +161 -139
- package/package.json +2 -2
- package/scripts/verify-host-compat.mjs +6 -6
- package/src/browser/types.ts +2 -2
- package/src/browser-electron/entry.ts +2 -2
- package/src/browser-electron/history-store.ts +1 -1
- package/src/browser-electron/host-main.ts +3 -3
- package/src/browser-electron/provider.ts +15 -15
- package/src/browser-electron/remote-host.ts +1 -1
- package/src/browser-electron/settings-store.ts +4 -4
- package/src/index.ts +1 -1
- package/src/tool-browser/element-fields.ts +6 -6
- package/src/tool-browser/index.ts +164 -140
|
@@ -1,14 +1,14 @@
|
|
|
1
1
|
/**
|
|
2
|
-
* Model-facing browser tools over `ctx.browser`: `
|
|
3
|
-
* `
|
|
4
|
-
* `
|
|
5
|
-
* `
|
|
2
|
+
* Model-facing browser tools over `ctx.browser`: `dsh_browser_open`,
|
|
3
|
+
* `dsh_browser_snapshot`, `dsh_browser_execute`, `dsh_browser_content`,
|
|
4
|
+
* `dsh_browser_screenshot`, and tab management (`dsh_browser_list_tabs`,
|
|
5
|
+
* `dsh_browser_switch_tab`, `dsh_browser_close_tab`, `dsh_browser_reset`).
|
|
6
6
|
*
|
|
7
7
|
* The tool layer owns only the model-facing schema, argument validation, and
|
|
8
8
|
* result formatting — never provider selection or page driving, which belong
|
|
9
9
|
* to the seam. Session lifecycle is owned here at the plugin level: each
|
|
10
10
|
* calling task (a DSH session) gets its own browser session — the first
|
|
11
|
-
* `
|
|
11
|
+
* `dsh_browser_open` (or any tool when no session exists) opens it, and later
|
|
12
12
|
* tools in the same task reuse it. Concurrent tasks therefore never fight
|
|
13
13
|
* over tabs, history, or navigation state.
|
|
14
14
|
* @module dsh-browser/tool-browser
|
|
@@ -57,33 +57,33 @@ const liveStates = new Set();
|
|
|
57
57
|
* Tools that only observe, plus the ones that undo a restriction.
|
|
58
58
|
*
|
|
59
59
|
* Never subject to an allow-list: restricting what the agent may *do* must not also
|
|
60
|
-
* blind it or trap it, and the tool descriptions promise as much.
|
|
60
|
+
* blind it or trap it, and the tool descriptions promise as much. dsh_browser_restrict
|
|
61
61
|
* is in here so a task that restricted everything can still lift it.
|
|
62
62
|
*/
|
|
63
63
|
export const READ_ONLY_TOOLS = new Set([
|
|
64
64
|
// Tools that only observe.
|
|
65
|
-
'
|
|
66
|
-
'
|
|
67
|
-
'
|
|
68
|
-
//
|
|
65
|
+
'dsh_browser_snapshot', 'dsh_browser_a11y', 'dsh_browser_content', 'dsh_browser_scrape',
|
|
66
|
+
'dsh_browser_screenshot', 'dsh_browser_get_value', 'dsh_browser_wait', 'dsh_browser_challenge',
|
|
67
|
+
'dsh_browser_list_tabs', 'dsh_browser_session', 'dsh_browser_history', 'dsh_browser_visited',
|
|
68
|
+
// dsh_browser_auth reads cookies out of the browser: an observation, and one the
|
|
69
69
|
// credentials switch gates on its own (see credentials.allowRead in the settings).
|
|
70
|
-
//
|
|
70
|
+
// dsh_browser_auth is deliberately NOT here. It reads cookies on "flush" but WRITES them on
|
|
71
71
|
// "restore", to arbitrary domains, which is an action like any other — exempting the tool
|
|
72
72
|
// would let a task that restricted its own actions rewrite the shared browser's logins.
|
|
73
73
|
//
|
|
74
74
|
// Getting out of a bad state, or lifting the restriction itself. Without these a task
|
|
75
75
|
// that restricted everything could neither recover nor be released.
|
|
76
|
-
'
|
|
76
|
+
'dsh_browser_restrict', 'dsh_browser_reset_session', 'dsh_browser_reset',
|
|
77
77
|
]);
|
|
78
78
|
function assertAllowed(state, toolName, task) {
|
|
79
|
-
//
|
|
79
|
+
// dsh_browser_restrict must stay usable, or a task that restricted everything
|
|
80
80
|
// could never lift it again.
|
|
81
81
|
// An allow-list is a statement about ACTIONS, so it must not take away the ability to
|
|
82
82
|
// look, or to get out of a bad state. The tool description and both READMEs promise
|
|
83
83
|
// that read-only tools are never blocked; before this they were, because only
|
|
84
|
-
//
|
|
85
|
-
//
|
|
86
|
-
if (READ_ONLY_TOOLS.has(toolName) || toolName === '
|
|
84
|
+
// dsh_browser_restrict itself was exempt — a narrow list then locked dsh_browser_history,
|
|
85
|
+
// dsh_browser_reset_session and dsh_browser_auth away with no way back.
|
|
86
|
+
if (READ_ONLY_TOOLS.has(toolName) || toolName === 'dsh_browser_restrict')
|
|
87
87
|
return;
|
|
88
88
|
const config = state.configRestrictedTo;
|
|
89
89
|
if (config !== undefined && !config.includes(toolName)) {
|
|
@@ -128,7 +128,7 @@ function assertNotBareDshWebUrl(rawUrl) {
|
|
|
128
128
|
// The authenticated URL printed by `dsh web` carries a token; the bare root does not.
|
|
129
129
|
if (target.searchParams.has('token'))
|
|
130
130
|
return;
|
|
131
|
-
throw new Error('
|
|
131
|
+
throw new Error('dsh_browser_open: refusing the bare DSH web root (' + dsh.origin + '/) because it carries no '
|
|
132
132
|
+ 'authentication token, so it answers with a 401 page that reads like a real site. Open the '
|
|
133
133
|
+ 'target site instead; to open DSH itself, use the authenticated URL printed by `dsh web`.');
|
|
134
134
|
}
|
|
@@ -262,17 +262,17 @@ function formatSnapshot(snapshot, options = {}) {
|
|
|
262
262
|
export function apply(ctx, config = {}) {
|
|
263
263
|
const timeoutMs = config.timeoutMs ?? 60_000;
|
|
264
264
|
/**
|
|
265
|
-
* Longest wait `
|
|
265
|
+
* Longest wait `dsh_browser_wait` may poll for, leaving room for the tool call itself to
|
|
266
266
|
* finish and report.
|
|
267
267
|
*
|
|
268
|
-
* `
|
|
268
|
+
* `dsh_browser_wait` promises a VERDICT — `{ready:false, reason}` — and the enclosing tool
|
|
269
269
|
* call aborts at `timeoutMs`. A caller-supplied budget longer than that aborted the call
|
|
270
270
|
* instead, so the documented verdict never arrived. Clamping to just inside the budget is
|
|
271
271
|
* what keeps the promise: the polling stops, the reason is returned.
|
|
272
272
|
*/
|
|
273
273
|
const maxWaitMs = Math.max(250, timeoutMs - 5_000);
|
|
274
274
|
/**
|
|
275
|
-
* Most fields `
|
|
275
|
+
* Most fields `dsh_browser_fill` accepts in one batch.
|
|
276
276
|
*
|
|
277
277
|
* The batch runs as ONE page evaluation, so hundreds of fields do not produce hundreds of
|
|
278
278
|
* per-field verdicts — they produce a single evaluate that exceeds the tool budget and
|
|
@@ -293,10 +293,32 @@ export function apply(ctx, config = {}) {
|
|
|
293
293
|
// Tool guidance band is 100-199; 150 keeps clear of the common 110/120
|
|
294
294
|
// tool sections so ordering does not depend on plugin load sequence.
|
|
295
295
|
order: 150,
|
|
296
|
-
text: 'Use the
|
|
296
|
+
text: 'Use the dsh_browser_* tools to operate the built-in browser. Each task gets its own browser session AND its own window (with a real toolbar: address bar, back/forward/reload, tab strip) — the human can see it, use it like any browser, and take over at any time; your tabs and history are isolated from other tasks, so do not assume another task\'s navigation state is visible to you. Understand a page with dsh_browser_a11y (semantic roles/names/states — the best structure map) or dsh_browser_snapshot (numbered interactive elements), then drive it: dsh_browser_click/dsh_browser_type accept a target {by: css|text|xpath, value} for semantic locating, or coordinates from dsh_browser_screenshot for visual targeting. For form filling prefer dsh_browser_fill (batch) or dsh_browser_set_value/dsh_browser_check/dsh_browser_select/dsh_browser_clear (single control, target-based); verify with dsh_browser_get_value. Use dsh_browser_scrape for structured extraction from list pages instead of hand-written dsh_browser_execute. After navigating on slow sites, dsh_browser_wait for the page. Keep the human informed of what you are doing on the page. If a snapshot or dsh_browser_challenge reports a human-verification challenge (CAPTCHA), stop retrying and ask the human to complete it in the browser window, then re-check.',
|
|
297
297
|
});
|
|
298
|
-
|
|
299
|
-
|
|
298
|
+
/**
|
|
299
|
+
* Register one tool, turning a name collision into a warning instead of a dead apply.
|
|
300
|
+
*
|
|
301
|
+
* Registering the tools in sequence without this is how ONE taken name becomes thirty-two
|
|
302
|
+
* missing ones: the throw aborts the rest of the apply, and what the user sees is a plugin that
|
|
303
|
+
* half-exists. That is precisely the shape of the failure reported against a sibling plugin,
|
|
304
|
+
* where the first spec collided and all six vanished behind a `console.warn` a desktop user
|
|
305
|
+
* never sees.
|
|
306
|
+
*
|
|
307
|
+
* A name this plugin cannot have is worth a sentence in the log, not the toolset.
|
|
308
|
+
* @param definition - the tool to register.
|
|
309
|
+
*/
|
|
310
|
+
const registerSafely = (definition) => {
|
|
311
|
+
try {
|
|
312
|
+
ctx.tools.register(definition);
|
|
313
|
+
}
|
|
314
|
+
catch (error) {
|
|
315
|
+
ctx.logger?.warn?.(`dsh-builtin-browser: could not register ${definition.name} (${String(error)}). Another `
|
|
316
|
+
+ 'plugin has probably taken that name — browser tool names are a shared namespace. The '
|
|
317
|
+
+ 'rest of this plugin\'s tools are unaffected.');
|
|
318
|
+
}
|
|
319
|
+
};
|
|
320
|
+
registerSafely(defineTool({
|
|
321
|
+
name: 'dsh_browser_open',
|
|
300
322
|
description: 'Open a URL in the shared browser window. Opens this task\'s browser session on first use; optionally opens in a new tab. Returns the resulting page snapshot. A fresh session starts on a blank, uncommitted tab: call this with the REAL target URL before reading, because reading that tab returns an empty page instead of waiting. Never treat the bare DSH host (e.g. http://127.0.0.1:3080/) as the target website — it requires the authenticated URL printed by `dsh web`, and the bare root is refused outright.',
|
|
301
323
|
parameters: {
|
|
302
324
|
url: { type: 'string', required: true, description: 'The URL to open (HTTP/HTTPS).' },
|
|
@@ -313,7 +335,7 @@ export function apply(ctx, config = {}) {
|
|
|
313
335
|
elements: {
|
|
314
336
|
type: 'array',
|
|
315
337
|
required: true,
|
|
316
|
-
// Same shared item schema as
|
|
338
|
+
// Same shared item schema as dsh_browser_open's, so the two cannot drift apart.
|
|
317
339
|
items: ELEMENT_ITEM_SCHEMA,
|
|
318
340
|
},
|
|
319
341
|
challenge: {
|
|
@@ -327,15 +349,15 @@ export function apply(ctx, config = {}) {
|
|
|
327
349
|
},
|
|
328
350
|
},
|
|
329
351
|
},
|
|
330
|
-
//
|
|
331
|
-
// `coords` parameter, so coordinates stay off here (
|
|
352
|
+
// dsh_browser_open is an overview of what just loaded, not a locator: it has no
|
|
353
|
+
// `coords` parameter, so coordinates stay off here (dsh_browser_snapshot owns
|
|
332
354
|
// that switch for callers that actually target pixels).
|
|
333
355
|
render: (args, value) => [{ type: 'text', text: formatSnapshot(value, {}) }],
|
|
334
356
|
},
|
|
335
357
|
timeoutMs,
|
|
336
358
|
isConcurrencySafe: () => false, // opens tabs / navigates; exclusive within a task
|
|
337
359
|
async execute(args, exec) {
|
|
338
|
-
assertAllowed(state, '
|
|
360
|
+
assertAllowed(state, 'dsh_browser_open', taskKey(exec));
|
|
339
361
|
assertNotBareDshWebUrl(args.url);
|
|
340
362
|
const browser = ctx.get('browser');
|
|
341
363
|
if (browser === undefined)
|
|
@@ -355,9 +377,9 @@ export function apply(ctx, config = {}) {
|
|
|
355
377
|
};
|
|
356
378
|
},
|
|
357
379
|
}));
|
|
358
|
-
|
|
359
|
-
name: '
|
|
360
|
-
description: 'Return an AI-friendly snapshot of the current shared-browser page: numbered interactive elements (inputs, buttons, links) the model can cite. This is the primary way to understand a page, and it needs no image input — every listed element can afterwards be addressed by a semantic target (
|
|
380
|
+
registerSafely(defineTool({
|
|
381
|
+
name: 'dsh_browser_snapshot',
|
|
382
|
+
description: 'Return an AI-friendly snapshot of the current shared-browser page: numbered interactive elements (inputs, buttons, links) the model can cite. This is the primary way to understand a page, and it needs no image input — every listed element can afterwards be addressed by a semantic target (dsh_browser_click, dsh_browser_type, dsh_browser_check, …) rather than by coordinates.',
|
|
361
383
|
parameters: {
|
|
362
384
|
coords: { type: 'boolean', description: 'Also print each element\'s viewport coordinates. Off by default; only useful when a vision pass is going to click a pixel position.' },
|
|
363
385
|
},
|
|
@@ -421,9 +443,9 @@ export function apply(ctx, config = {}) {
|
|
|
421
443
|
};
|
|
422
444
|
},
|
|
423
445
|
}));
|
|
424
|
-
|
|
425
|
-
name: '
|
|
426
|
-
description: 'Read the page\'s accessibility tree: every interactive node with its semantic role (button/link/textbox/checkbox/…), accessible name, current value, and states (enabled/disabled/checked/expanded/…), (coordinates only when coords: true). Prefer this over
|
|
446
|
+
registerSafely(defineTool({
|
|
447
|
+
name: 'dsh_browser_a11y',
|
|
448
|
+
description: 'Read the page\'s accessibility tree: every interactive node with its semantic role (button/link/textbox/checkbox/…), accessible name, current value, and states (enabled/disabled/checked/expanded/…), (coordinates only when coords: true). Prefer this over dsh_browser_snapshot to understand a page\'s structure and find the right element: roles and names tell you WHAT each node is, and the name is what dsh_browser_click matches when you target by text. Penetrates same-origin iframes and shadow roots.',
|
|
427
449
|
parameters: {
|
|
428
450
|
includeHidden: { type: 'boolean', description: 'Include hidden elements (default false).' },
|
|
429
451
|
maxNodes: { type: 'number', description: 'Maximum nodes (default 150, range 10-5000).' },
|
|
@@ -506,9 +528,9 @@ export function apply(ctx, config = {}) {
|
|
|
506
528
|
};
|
|
507
529
|
},
|
|
508
530
|
}));
|
|
509
|
-
|
|
510
|
-
name: '
|
|
511
|
-
description: 'Check whether a human-verification challenge (CAPTCHA / bot detection: Cloudflare "Just a moment", reCAPTCHA, hCaptcha, Turnstile) is blocking the current page. When blocked, do NOT keep retrying automated steps — ask the human to complete the verification in the shared browser window, then re-check with
|
|
531
|
+
registerSafely(defineTool({
|
|
532
|
+
name: 'dsh_browser_challenge',
|
|
533
|
+
description: 'Check whether a human-verification challenge (CAPTCHA / bot detection: Cloudflare "Just a moment", reCAPTCHA, hCaptcha, Turnstile) is blocking the current page. When blocked, do NOT keep retrying automated steps — ask the human to complete the verification in the shared browser window, then re-check with dsh_browser_snapshot.',
|
|
512
534
|
parameters: {},
|
|
513
535
|
output: {
|
|
514
536
|
schema: {
|
|
@@ -541,14 +563,14 @@ export function apply(ctx, config = {}) {
|
|
|
541
563
|
...challenge.kind !== undefined ? { kind: challenge.kind } : {},
|
|
542
564
|
...challenge.reason !== undefined ? { reason: challenge.reason } : {},
|
|
543
565
|
hint: challenge.blocked
|
|
544
|
-
? 'Ask the human to complete the verification in the shared browser window (the page is visible to them), then re-check with
|
|
566
|
+
? 'Ask the human to complete the verification in the shared browser window (the page is visible to them), then re-check with dsh_browser_snapshot.'
|
|
545
567
|
: '',
|
|
546
568
|
};
|
|
547
569
|
},
|
|
548
570
|
}));
|
|
549
|
-
|
|
550
|
-
name: '
|
|
551
|
-
description: 'Wait until the shared-browser page is ready: load complete, and optionally the expected URL and/or a CSS selector present (top document or same-origin iframes). Use after
|
|
571
|
+
registerSafely(defineTool({
|
|
572
|
+
name: 'dsh_browser_wait',
|
|
573
|
+
description: 'Wait until the shared-browser page is ready: load complete, and optionally the expected URL and/or a CSS selector present (top document or same-origin iframes). Use after dsh_browser_open on slow sites instead of snapshotting a white page — wait for the URL you navigated to first. Returns ready=true/false with a short reason; a miss is not an error.',
|
|
552
574
|
parameters: {
|
|
553
575
|
timeoutMs: { type: 'number', description: 'Maximum wait in ms (default 30000).' },
|
|
554
576
|
url: { type: 'string', description: 'Expected page URL (exact or prefix), e.g. the URL you opened.' },
|
|
@@ -582,8 +604,8 @@ export function apply(ctx, config = {}) {
|
|
|
582
604
|
return { ready: result.ready, reason: result.reason };
|
|
583
605
|
},
|
|
584
606
|
}));
|
|
585
|
-
|
|
586
|
-
name: '
|
|
607
|
+
registerSafely(defineTool({
|
|
608
|
+
name: 'dsh_browser_execute',
|
|
587
609
|
description: 'Execute JavaScript in the shared-browser page context. This is the primary way to interact with page elements: focus, fill inputs (use the native value setter for framework-controlled inputs, then dispatch an input event), click buttons (element.click() or a constructed MouseEvent). Returns the evaluation result by value, or the exception text.',
|
|
588
610
|
parameters: {
|
|
589
611
|
script: { type: 'string', required: true, description: 'The JavaScript expression to evaluate in the page context.' },
|
|
@@ -608,7 +630,7 @@ export function apply(ctx, config = {}) {
|
|
|
608
630
|
const rendered = String(value.value ?? '');
|
|
609
631
|
const cap = 50_000;
|
|
610
632
|
const text = rendered.length > cap
|
|
611
|
-
? `Result: ${rendered.slice(0, cap)}\n(result truncated at ${cap} of ${rendered.length} characters — return a smaller value, or write it with
|
|
633
|
+
? `Result: ${rendered.slice(0, cap)}\n(result truncated at ${cap} of ${rendered.length} characters — return a smaller value, or write it with dsh_browser_execute + a variable)`
|
|
612
634
|
: `Result: ${rendered}`;
|
|
613
635
|
return [{ type: 'text', text }];
|
|
614
636
|
},
|
|
@@ -616,7 +638,7 @@ export function apply(ctx, config = {}) {
|
|
|
616
638
|
timeoutMs,
|
|
617
639
|
isConcurrencySafe: () => false, // page JS can be stateful
|
|
618
640
|
async execute(args, exec) {
|
|
619
|
-
assertAllowed(state, '
|
|
641
|
+
assertAllowed(state, 'dsh_browser_execute', taskKey(exec));
|
|
620
642
|
const browser = ctx.get('browser');
|
|
621
643
|
if (browser === undefined)
|
|
622
644
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -633,8 +655,8 @@ export function apply(ctx, config = {}) {
|
|
|
633
655
|
return { ok: false, exception: result.exception };
|
|
634
656
|
},
|
|
635
657
|
}));
|
|
636
|
-
|
|
637
|
-
name: '
|
|
658
|
+
registerSafely(defineTool({
|
|
659
|
+
name: 'dsh_browser_content',
|
|
638
660
|
description: 'Fetch the current shared-browser page content in a chosen format: html (raw DOM), markdown (structured reading), txt (plain text), or json. Optionally scope to a CSS selector and cap the length. Use this to read page content, not to interact.',
|
|
639
661
|
parameters: {
|
|
640
662
|
format: { type: 'string', required: true, enum: ['html', 'markdown', 'txt', 'json'], description: 'Output format.' },
|
|
@@ -669,9 +691,9 @@ export function apply(ctx, config = {}) {
|
|
|
669
691
|
return { content: result.content, truncated: result.truncated };
|
|
670
692
|
},
|
|
671
693
|
}));
|
|
672
|
-
|
|
673
|
-
name: '
|
|
674
|
-
description: 'Extract structured data from a page: give the container CSS selector (item) and a field map (name -> selector, optionally selector@attr to take an attribute; a@href yields the absolute URL). Returns one object per item. Static CSS queries only — no arbitrary code runs — so it is safe on any page. Use for list pages (search results, cards, tables) instead of hand-writing
|
|
694
|
+
registerSafely(defineTool({
|
|
695
|
+
name: 'dsh_browser_scrape',
|
|
696
|
+
description: 'Extract structured data from a page: give the container CSS selector (item) and a field map (name -> selector, optionally selector@attr to take an attribute; a@href yields the absolute URL). Returns one object per item. Static CSS queries only — no arbitrary code runs — so it is safe on any page. Use for list pages (search results, cards, tables) instead of hand-writing dsh_browser_execute.',
|
|
675
697
|
parameters: {
|
|
676
698
|
item: { type: 'string', required: true, description: 'CSS selector of each result container (e.g. "div.card").' },
|
|
677
699
|
fields: {
|
|
@@ -706,7 +728,7 @@ export function apply(ctx, config = {}) {
|
|
|
706
728
|
timeoutMs,
|
|
707
729
|
isConcurrencySafe: () => true,
|
|
708
730
|
async execute(args, exec) {
|
|
709
|
-
assertAllowed(state, '
|
|
731
|
+
assertAllowed(state, 'dsh_browser_scrape', taskKey(exec));
|
|
710
732
|
const browser = ctx.get('browser');
|
|
711
733
|
if (browser === undefined)
|
|
712
734
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -720,9 +742,9 @@ export function apply(ctx, config = {}) {
|
|
|
720
742
|
return { count: result.count, items: result.items.map(it => ({ ...it })) };
|
|
721
743
|
},
|
|
722
744
|
}));
|
|
723
|
-
|
|
724
|
-
name: '
|
|
725
|
-
description: 'Click in the shared browser. Prefer the semantic target: pass target {by: css|text|xpath, value, index?} and the element is located from the DOM, scrolled into view and clicked at its centre — this needs no image input and is the only form that works without vision. The alternative is viewport coordinates (x/y), which are meaningful only after a vision model has located the element on a
|
|
745
|
+
registerSafely(defineTool({
|
|
746
|
+
name: 'dsh_browser_click',
|
|
747
|
+
description: 'Click in the shared browser. Prefer the semantic target: pass target {by: css|text|xpath, value, index?} and the element is located from the DOM, scrolled into view and clicked at its centre — this needs no image input and is the only form that works without vision. The alternative is viewport coordinates (x/y), which are meaningful only after a vision model has located the element on a dsh_browser_screenshot; coordinate clicks are refused outright under the non-visual strategy. Provide exactly one of target or x/y.',
|
|
726
748
|
parameters: {
|
|
727
749
|
target: {
|
|
728
750
|
type: 'object',
|
|
@@ -734,7 +756,7 @@ export function apply(ctx, config = {}) {
|
|
|
734
756
|
},
|
|
735
757
|
description: 'Locate the element semantically and click it (css/text/xpath).',
|
|
736
758
|
},
|
|
737
|
-
x: { type: 'number', description: 'Viewport x coordinate (CSS px), taken from a
|
|
759
|
+
x: { type: 'number', description: 'Viewport x coordinate (CSS px), taken from a dsh_browser_screenshot read by a vision model. Prefer a semantic target — it works without image input.' },
|
|
738
760
|
y: { type: 'number', description: 'Viewport y coordinate (CSS px), taken from the same screenshot as x.' },
|
|
739
761
|
},
|
|
740
762
|
output: {
|
|
@@ -744,7 +766,7 @@ export function apply(ctx, config = {}) {
|
|
|
744
766
|
timeoutMs,
|
|
745
767
|
isConcurrencySafe: () => false,
|
|
746
768
|
async execute(args, exec) {
|
|
747
|
-
assertAllowed(state, '
|
|
769
|
+
assertAllowed(state, 'dsh_browser_click', taskKey(exec));
|
|
748
770
|
const browser = ctx.get('browser');
|
|
749
771
|
if (browser === undefined)
|
|
750
772
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -768,16 +790,16 @@ export function apply(ctx, config = {}) {
|
|
|
768
790
|
}
|
|
769
791
|
else {
|
|
770
792
|
if (typeof args.x !== 'number' || typeof args.y !== 'number') {
|
|
771
|
-
throw new Error('
|
|
793
|
+
throw new Error('dsh_browser_click: provide a target (css/text/xpath) or x/y coordinates');
|
|
772
794
|
}
|
|
773
795
|
await browser.click(session, { x: args.x, y: args.y }, exec.signal);
|
|
774
796
|
}
|
|
775
797
|
return { clicked: true };
|
|
776
798
|
},
|
|
777
799
|
}));
|
|
778
|
-
|
|
779
|
-
name: '
|
|
780
|
-
description: 'Type text into the focused element of the shared browser, or into a located element (pass target {by: css|text|xpath, value, index?} to focus it first). Text is inserted at the focus via CDP Input.insertText. For setting whole field values (and React-controlled inputs) prefer
|
|
800
|
+
registerSafely(defineTool({
|
|
801
|
+
name: 'dsh_browser_type',
|
|
802
|
+
description: 'Type text into the focused element of the shared browser, or into a located element (pass target {by: css|text|xpath, value, index?} to focus it first). Text is inserted at the focus via CDP Input.insertText. For setting whole field values (and React-controlled inputs) prefer dsh_browser_set_value; use dsh_browser_key for Enter/Tab/arrows.',
|
|
781
803
|
parameters: {
|
|
782
804
|
text: { type: 'string', required: true, description: 'The text to insert.' },
|
|
783
805
|
target: {
|
|
@@ -798,7 +820,7 @@ export function apply(ctx, config = {}) {
|
|
|
798
820
|
timeoutMs,
|
|
799
821
|
isConcurrencySafe: () => false,
|
|
800
822
|
async execute(args, exec) {
|
|
801
|
-
assertAllowed(state, '
|
|
823
|
+
assertAllowed(state, 'dsh_browser_type', taskKey(exec));
|
|
802
824
|
const browser = ctx.get('browser');
|
|
803
825
|
if (browser === undefined)
|
|
804
826
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -827,8 +849,8 @@ export function apply(ctx, config = {}) {
|
|
|
827
849
|
return { typed: true };
|
|
828
850
|
},
|
|
829
851
|
}));
|
|
830
|
-
|
|
831
|
-
name: '
|
|
852
|
+
registerSafely(defineTool({
|
|
853
|
+
name: 'dsh_browser_scroll',
|
|
832
854
|
description: 'Scroll the shared-browser page: by pixel deltas (deltaX/deltaY), to a CSS selector\'s element, or to the top/bottom. Use to reveal below-the-fold content before snapshotting or clicking.',
|
|
833
855
|
parameters: {
|
|
834
856
|
deltaX: { type: 'number', description: 'Horizontal scroll delta in CSS pixels.' },
|
|
@@ -844,7 +866,7 @@ export function apply(ctx, config = {}) {
|
|
|
844
866
|
timeoutMs,
|
|
845
867
|
isConcurrencySafe: () => false, // mutates page scroll state; exclusive within a task
|
|
846
868
|
async execute(args, exec) {
|
|
847
|
-
assertAllowed(state, '
|
|
869
|
+
assertAllowed(state, 'dsh_browser_scroll', taskKey(exec));
|
|
848
870
|
const browser = ctx.get('browser');
|
|
849
871
|
if (browser === undefined)
|
|
850
872
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -859,8 +881,8 @@ export function apply(ctx, config = {}) {
|
|
|
859
881
|
return { scrolled: true };
|
|
860
882
|
},
|
|
861
883
|
}));
|
|
862
|
-
|
|
863
|
-
name: '
|
|
884
|
+
registerSafely(defineTool({
|
|
885
|
+
name: 'dsh_browser_back',
|
|
864
886
|
description: 'Go back one step in the shared-browser page history. A no-op when there is no previous entry.',
|
|
865
887
|
parameters: {},
|
|
866
888
|
output: {
|
|
@@ -870,7 +892,7 @@ export function apply(ctx, config = {}) {
|
|
|
870
892
|
timeoutMs,
|
|
871
893
|
isConcurrencySafe: () => false, // mutates page state; exclusive within a task
|
|
872
894
|
async execute(_args, exec) {
|
|
873
|
-
assertAllowed(state, '
|
|
895
|
+
assertAllowed(state, 'dsh_browser_back', taskKey(exec));
|
|
874
896
|
const browser = ctx.get('browser');
|
|
875
897
|
if (browser === undefined)
|
|
876
898
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -879,8 +901,8 @@ export function apply(ctx, config = {}) {
|
|
|
879
901
|
return { back: true };
|
|
880
902
|
},
|
|
881
903
|
}));
|
|
882
|
-
|
|
883
|
-
name: '
|
|
904
|
+
registerSafely(defineTool({
|
|
905
|
+
name: 'dsh_browser_forward',
|
|
884
906
|
description: 'Go forward one step in the shared-browser page history. A no-op when there is no next entry.',
|
|
885
907
|
parameters: {},
|
|
886
908
|
output: {
|
|
@@ -890,7 +912,7 @@ export function apply(ctx, config = {}) {
|
|
|
890
912
|
timeoutMs,
|
|
891
913
|
isConcurrencySafe: () => false, // mutates page state; exclusive within a task
|
|
892
914
|
async execute(_args, exec) {
|
|
893
|
-
assertAllowed(state, '
|
|
915
|
+
assertAllowed(state, 'dsh_browser_forward', taskKey(exec));
|
|
894
916
|
const browser = ctx.get('browser');
|
|
895
917
|
if (browser === undefined)
|
|
896
918
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -899,8 +921,8 @@ export function apply(ctx, config = {}) {
|
|
|
899
921
|
return { forward: true };
|
|
900
922
|
},
|
|
901
923
|
}));
|
|
902
|
-
|
|
903
|
-
name: '
|
|
924
|
+
registerSafely(defineTool({
|
|
925
|
+
name: 'dsh_browser_refresh',
|
|
904
926
|
description: 'Reload the current page in the shared browser (like a browser\'s refresh button). Use after a page got stuck, to apply script changes, or to re-fetch a page.',
|
|
905
927
|
parameters: {},
|
|
906
928
|
output: {
|
|
@@ -910,7 +932,7 @@ export function apply(ctx, config = {}) {
|
|
|
910
932
|
timeoutMs,
|
|
911
933
|
isConcurrencySafe: () => false, // reloads the page; exclusive within a task
|
|
912
934
|
async execute(_args, exec) {
|
|
913
|
-
assertAllowed(state, '
|
|
935
|
+
assertAllowed(state, 'dsh_browser_refresh', taskKey(exec));
|
|
914
936
|
const browser = ctx.get('browser');
|
|
915
937
|
if (browser === undefined)
|
|
916
938
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -919,8 +941,8 @@ export function apply(ctx, config = {}) {
|
|
|
919
941
|
return { refreshed: true };
|
|
920
942
|
},
|
|
921
943
|
}));
|
|
922
|
-
|
|
923
|
-
name: '
|
|
944
|
+
registerSafely(defineTool({
|
|
945
|
+
name: 'dsh_browser_key',
|
|
924
946
|
description: 'Press one named key in the shared-browser page (Enter, Tab, Escape, Backspace, Delete, ArrowUp/Down/Left/Right, Home, End, PageUp, PageDown, Space). Use after focusing an input to submit a chat box (Enter), move focus (Tab), or navigate a list (arrows).',
|
|
925
947
|
parameters: {
|
|
926
948
|
key: { type: 'string', required: true, enum: ['Enter', 'Tab', 'Escape', 'Backspace', 'Delete', 'ArrowUp', 'ArrowDown', 'ArrowLeft', 'ArrowRight', 'Home', 'End', 'PageUp', 'PageDown', 'Space'], description: 'The key to press.' },
|
|
@@ -932,7 +954,7 @@ export function apply(ctx, config = {}) {
|
|
|
932
954
|
timeoutMs,
|
|
933
955
|
isConcurrencySafe: () => false, // mutates page state; exclusive within a task
|
|
934
956
|
async execute(args, exec) {
|
|
935
|
-
assertAllowed(state, '
|
|
957
|
+
assertAllowed(state, 'dsh_browser_key', taskKey(exec));
|
|
936
958
|
const browser = ctx.get('browser');
|
|
937
959
|
if (browser === undefined)
|
|
938
960
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -941,9 +963,9 @@ export function apply(ctx, config = {}) {
|
|
|
941
963
|
return { pressed: true };
|
|
942
964
|
},
|
|
943
965
|
}));
|
|
944
|
-
|
|
945
|
-
name: '
|
|
946
|
-
description: 'Fill a form in one batch: pass fields with a CSS selector or name/label/placeholder text and the value to set (string, number, or boolean for checkbox/radio; for selects or radio groups pass the option value or visible text). Values are applied with the native setter plus input/change events, so React/Vue controlled inputs update correctly. Optionally submit the containing form. Prefer this over hand-written
|
|
966
|
+
registerSafely(defineTool({
|
|
967
|
+
name: 'dsh_browser_fill',
|
|
968
|
+
description: 'Fill a form in one batch: pass fields with a CSS selector or name/label/placeholder text and the value to set (string, number, or boolean for checkbox/radio; for selects or radio groups pass the option value or visible text). Values are applied with the native setter plus input/change events, so React/Vue controlled inputs update correctly. Optionally submit the containing form. Prefer this over hand-written dsh_browser_execute for form filling; per-field failures are reported instead of throwing.',
|
|
947
969
|
parameters: {
|
|
948
970
|
fields: {
|
|
949
971
|
type: 'array',
|
|
@@ -1001,14 +1023,14 @@ export function apply(ctx, config = {}) {
|
|
|
1001
1023
|
timeoutMs,
|
|
1002
1024
|
isConcurrencySafe: () => false,
|
|
1003
1025
|
async execute(args, exec) {
|
|
1004
|
-
assertAllowed(state, '
|
|
1026
|
+
assertAllowed(state, 'dsh_browser_fill', taskKey(exec));
|
|
1005
1027
|
const browser = ctx.get('browser');
|
|
1006
1028
|
if (browser === undefined)
|
|
1007
1029
|
throw new Error('tool-browser: browser service unavailable');
|
|
1008
1030
|
const session = await ensureSession(browser, state, taskKey(exec), agentOf(exec));
|
|
1009
1031
|
const requested = args.fields ?? [];
|
|
1010
1032
|
if (requested.length > maxFillFields) {
|
|
1011
|
-
throw new Error(`
|
|
1033
|
+
throw new Error(`dsh_browser_fill: ${requested.length} fields in one call exceeds the ${maxFillFields}-field limit; split it into several calls so each one can report per-field results instead of timing out as a batch`);
|
|
1012
1034
|
}
|
|
1013
1035
|
const fields = requested.map((f) => {
|
|
1014
1036
|
// A missing value is a caller mistake, not a request to clear the field. Turning
|
|
@@ -1017,7 +1039,7 @@ export function apply(ctx, config = {}) {
|
|
|
1017
1039
|
// a call can still arrive without it.
|
|
1018
1040
|
if (f.value === undefined) {
|
|
1019
1041
|
const matcher = f.selector ?? f.name ?? f.label ?? f.placeholder ?? '(no matcher)';
|
|
1020
|
-
throw new Error(`
|
|
1042
|
+
throw new Error(`dsh_browser_fill: field "${matcher}" has no "value"; pass one (use "false" to uncheck, or "" to clear deliberately)`);
|
|
1021
1043
|
}
|
|
1022
1044
|
return {
|
|
1023
1045
|
...f.selector !== undefined ? { selector: f.selector } : {},
|
|
@@ -1052,9 +1074,9 @@ export function apply(ctx, config = {}) {
|
|
|
1052
1074
|
},
|
|
1053
1075
|
description: 'Locate the element (css/text/xpath).',
|
|
1054
1076
|
};
|
|
1055
|
-
|
|
1056
|
-
name: '
|
|
1057
|
-
description: 'Set the value of ONE input/textarea/select/contenteditable, located by css/text/xpath. Uses the native setter plus input/change events, so React/Vue controlled inputs update correctly. For selects pass the option value or visible text. For a whole form at once prefer
|
|
1077
|
+
registerSafely(defineTool({
|
|
1078
|
+
name: 'dsh_browser_set_value',
|
|
1079
|
+
description: 'Set the value of ONE input/textarea/select/contenteditable, located by css/text/xpath. Uses the native setter plus input/change events, so React/Vue controlled inputs update correctly. For selects pass the option value or visible text. For a whole form at once prefer dsh_browser_fill.',
|
|
1058
1080
|
parameters: {
|
|
1059
1081
|
target: targetParam,
|
|
1060
1082
|
value: { type: 'string', required: true, description: 'The value to set (string form; numbers/booleans accepted).' },
|
|
@@ -1073,7 +1095,7 @@ export function apply(ctx, config = {}) {
|
|
|
1073
1095
|
timeoutMs,
|
|
1074
1096
|
isConcurrencySafe: () => false, // mutates a field; exclusive within a task
|
|
1075
1097
|
async execute(args, exec) {
|
|
1076
|
-
assertAllowed(state, '
|
|
1098
|
+
assertAllowed(state, 'dsh_browser_set_value', taskKey(exec));
|
|
1077
1099
|
const browser = ctx.get('browser');
|
|
1078
1100
|
if (browser === undefined)
|
|
1079
1101
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1090,8 +1112,8 @@ export function apply(ctx, config = {}) {
|
|
|
1090
1112
|
return { method: result.method, value: result.value };
|
|
1091
1113
|
},
|
|
1092
1114
|
}));
|
|
1093
|
-
|
|
1094
|
-
name: '
|
|
1115
|
+
registerSafely(defineTool({
|
|
1116
|
+
name: 'dsh_browser_check',
|
|
1095
1117
|
description: 'Check (or uncheck, with checked=false) a checkbox or radio button, located by css/text/xpath.',
|
|
1096
1118
|
parameters: {
|
|
1097
1119
|
target: targetParam,
|
|
@@ -1104,7 +1126,7 @@ export function apply(ctx, config = {}) {
|
|
|
1104
1126
|
timeoutMs,
|
|
1105
1127
|
isConcurrencySafe: () => false, // mutates a field; exclusive within a task
|
|
1106
1128
|
async execute(args, exec) {
|
|
1107
|
-
assertAllowed(state, '
|
|
1129
|
+
assertAllowed(state, 'dsh_browser_check', taskKey(exec));
|
|
1108
1130
|
const browser = ctx.get('browser');
|
|
1109
1131
|
if (browser === undefined)
|
|
1110
1132
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1121,8 +1143,8 @@ export function apply(ctx, config = {}) {
|
|
|
1121
1143
|
return { checked: result.checked };
|
|
1122
1144
|
},
|
|
1123
1145
|
}));
|
|
1124
|
-
|
|
1125
|
-
name: '
|
|
1146
|
+
registerSafely(defineTool({
|
|
1147
|
+
name: 'dsh_browser_select',
|
|
1126
1148
|
description: 'Select one option of a <select>, located by css/text/xpath — by option value, visible text, or 0-based index (provide exactly one).',
|
|
1127
1149
|
parameters: {
|
|
1128
1150
|
target: targetParam,
|
|
@@ -1144,7 +1166,7 @@ export function apply(ctx, config = {}) {
|
|
|
1144
1166
|
timeoutMs,
|
|
1145
1167
|
isConcurrencySafe: () => false, // mutates a field; exclusive within a task
|
|
1146
1168
|
async execute(args, exec) {
|
|
1147
|
-
assertAllowed(state, '
|
|
1169
|
+
assertAllowed(state, 'dsh_browser_select', taskKey(exec));
|
|
1148
1170
|
const browser = ctx.get('browser');
|
|
1149
1171
|
if (browser === undefined)
|
|
1150
1172
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1163,8 +1185,8 @@ export function apply(ctx, config = {}) {
|
|
|
1163
1185
|
return { value: result.value, text: result.text };
|
|
1164
1186
|
},
|
|
1165
1187
|
}));
|
|
1166
|
-
|
|
1167
|
-
name: '
|
|
1188
|
+
registerSafely(defineTool({
|
|
1189
|
+
name: 'dsh_browser_clear',
|
|
1168
1190
|
description: 'Clear an input/textarea/contenteditable, or uncheck a checkbox/radio, located by css/text/xpath.',
|
|
1169
1191
|
parameters: {
|
|
1170
1192
|
target: targetParam,
|
|
@@ -1176,7 +1198,7 @@ export function apply(ctx, config = {}) {
|
|
|
1176
1198
|
timeoutMs,
|
|
1177
1199
|
isConcurrencySafe: () => false, // mutates a field; exclusive within a task
|
|
1178
1200
|
async execute(args, exec) {
|
|
1179
|
-
assertAllowed(state, '
|
|
1201
|
+
assertAllowed(state, 'dsh_browser_clear', taskKey(exec));
|
|
1180
1202
|
const browser = ctx.get('browser');
|
|
1181
1203
|
if (browser === undefined)
|
|
1182
1204
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1192,8 +1214,8 @@ export function apply(ctx, config = {}) {
|
|
|
1192
1214
|
return { cleared: true };
|
|
1193
1215
|
},
|
|
1194
1216
|
}));
|
|
1195
|
-
|
|
1196
|
-
name: '
|
|
1217
|
+
registerSafely(defineTool({
|
|
1218
|
+
name: 'dsh_browser_get_value',
|
|
1197
1219
|
description: 'Read the current value of one input/textarea/select/contenteditable, located by css/text/xpath. Use to VERIFY that a fill worked (e.g. before submitting).',
|
|
1198
1220
|
parameters: {
|
|
1199
1221
|
target: targetParam,
|
|
@@ -1219,7 +1241,7 @@ export function apply(ctx, config = {}) {
|
|
|
1219
1241
|
timeoutMs,
|
|
1220
1242
|
isConcurrencySafe: () => true,
|
|
1221
1243
|
async execute(args, exec) {
|
|
1222
|
-
assertAllowed(state, '
|
|
1244
|
+
assertAllowed(state, 'dsh_browser_get_value', taskKey(exec));
|
|
1223
1245
|
const browser = ctx.get('browser');
|
|
1224
1246
|
if (browser === undefined)
|
|
1225
1247
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1239,9 +1261,9 @@ export function apply(ctx, config = {}) {
|
|
|
1239
1261
|
};
|
|
1240
1262
|
},
|
|
1241
1263
|
}));
|
|
1242
|
-
|
|
1243
|
-
name: '
|
|
1244
|
-
description: 'Capture the current shared-browser page as a screenshot (PNG default, JPEG optional). This is for models that can read images: layout checks, charts, designs, CAPTCHAs, or locating an element by eye before clicking its coordinates. A model without image input gains nothing from it —
|
|
1264
|
+
registerSafely(defineTool({
|
|
1265
|
+
name: 'dsh_browser_screenshot',
|
|
1266
|
+
description: 'Capture the current shared-browser page as a screenshot (PNG default, JPEG optional). This is for models that can read images: layout checks, charts, designs, CAPTCHAs, or locating an element by eye before clicking its coordinates. A model without image input gains nothing from it — dsh_browser_snapshot, dsh_browser_a11y and dsh_browser_content carry the same page as text, and dsh_browser_scrape extracts structured data. Supports full-page capture, save-to-file, JPEG encoding, and downscaling (maxWidth/maxHeight) to cut vision-tool token cost. JPEG needs a browser whose CDP encoder works — the self-hosted browser and an installed Chrome/Edge; some carriers (the desktop sidebar) return PNG instead. Downscaling works on every carrier.',
|
|
1245
1267
|
parameters: {
|
|
1246
1268
|
fullPage: { type: 'boolean', description: 'Capture the full scrollable page instead of the viewport (default false).' },
|
|
1247
1269
|
savePath: { type: 'string', description: 'Absolute file path to also save the image to (e.g. for read_image vision location). Must resolve inside the configured downloadDir (default: the system Downloads folder, localized names such as ~/下载 included); an existing file is never overwritten.' },
|
|
@@ -1291,8 +1313,8 @@ export function apply(ctx, config = {}) {
|
|
|
1291
1313
|
},
|
|
1292
1314
|
}));
|
|
1293
1315
|
if (config.tabTools !== false) {
|
|
1294
|
-
|
|
1295
|
-
name: '
|
|
1316
|
+
registerSafely(defineTool({
|
|
1317
|
+
name: 'dsh_browser_list_tabs',
|
|
1296
1318
|
description: 'List the shared-browser session\'s tabs with their URLs and which is active.',
|
|
1297
1319
|
parameters: {},
|
|
1298
1320
|
output: {
|
|
@@ -1332,9 +1354,9 @@ export function apply(ctx, config = {}) {
|
|
|
1332
1354
|
return { tabs: tabs.map(t => ({ id: t.id, url: t.url, active: t.active })) };
|
|
1333
1355
|
},
|
|
1334
1356
|
}));
|
|
1335
|
-
|
|
1336
|
-
name: '
|
|
1337
|
-
description: 'Switch the shared browser to a tab by id (from
|
|
1357
|
+
registerSafely(defineTool({
|
|
1358
|
+
name: 'dsh_browser_switch_tab',
|
|
1359
|
+
description: 'Switch the shared browser to a tab by id (from dsh_browser_list_tabs).',
|
|
1338
1360
|
parameters: {
|
|
1339
1361
|
tabId: { type: 'string', required: true, description: 'The tab id to switch to.' },
|
|
1340
1362
|
},
|
|
@@ -1345,7 +1367,7 @@ export function apply(ctx, config = {}) {
|
|
|
1345
1367
|
timeoutMs,
|
|
1346
1368
|
isConcurrencySafe: () => false, // mutates the active tab; exclusive within a task
|
|
1347
1369
|
async execute(args, exec) {
|
|
1348
|
-
assertAllowed(state, '
|
|
1370
|
+
assertAllowed(state, 'dsh_browser_switch_tab', taskKey(exec));
|
|
1349
1371
|
const browser = ctx.get('browser');
|
|
1350
1372
|
if (browser === undefined)
|
|
1351
1373
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1354,8 +1376,8 @@ export function apply(ctx, config = {}) {
|
|
|
1354
1376
|
return { switched: true };
|
|
1355
1377
|
},
|
|
1356
1378
|
}));
|
|
1357
|
-
|
|
1358
|
-
name: '
|
|
1379
|
+
registerSafely(defineTool({
|
|
1380
|
+
name: 'dsh_browser_close_tab',
|
|
1359
1381
|
description: 'Close a tab in the shared browser by id. Closing the active tab activates the next.',
|
|
1360
1382
|
parameters: {
|
|
1361
1383
|
tabId: { type: 'string', required: true, description: 'The tab id to close.' },
|
|
@@ -1367,7 +1389,7 @@ export function apply(ctx, config = {}) {
|
|
|
1367
1389
|
timeoutMs,
|
|
1368
1390
|
isConcurrencySafe: () => false, // mutates the tab list; exclusive within a task
|
|
1369
1391
|
async execute(args, exec) {
|
|
1370
|
-
assertAllowed(state, '
|
|
1392
|
+
assertAllowed(state, 'dsh_browser_close_tab', taskKey(exec));
|
|
1371
1393
|
const browser = ctx.get('browser');
|
|
1372
1394
|
if (browser === undefined)
|
|
1373
1395
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1376,8 +1398,8 @@ export function apply(ctx, config = {}) {
|
|
|
1376
1398
|
return { closed: true };
|
|
1377
1399
|
},
|
|
1378
1400
|
}));
|
|
1379
|
-
|
|
1380
|
-
name: '
|
|
1401
|
+
registerSafely(defineTool({
|
|
1402
|
+
name: 'dsh_browser_reset',
|
|
1381
1403
|
description: 'Close every tab in the shared browser and start fresh with one blank tab.',
|
|
1382
1404
|
parameters: {},
|
|
1383
1405
|
output: {
|
|
@@ -1387,7 +1409,7 @@ export function apply(ctx, config = {}) {
|
|
|
1387
1409
|
timeoutMs,
|
|
1388
1410
|
isConcurrencySafe: () => false, // closes every tab; exclusive within a task
|
|
1389
1411
|
async execute(_args, exec) {
|
|
1390
|
-
assertAllowed(state, '
|
|
1412
|
+
assertAllowed(state, 'dsh_browser_reset', taskKey(exec));
|
|
1391
1413
|
const browser = ctx.get('browser');
|
|
1392
1414
|
if (browser === undefined)
|
|
1393
1415
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1397,8 +1419,8 @@ export function apply(ctx, config = {}) {
|
|
|
1397
1419
|
},
|
|
1398
1420
|
}));
|
|
1399
1421
|
}
|
|
1400
|
-
|
|
1401
|
-
name: '
|
|
1422
|
+
registerSafely(defineTool({
|
|
1423
|
+
name: 'dsh_browser_history',
|
|
1402
1424
|
description: 'List the shared browser session\'s recorded operation history (navigate/execute/click/type), newest last, with per-step success/error. Use to understand what the agent did and to pick a step to replay.',
|
|
1403
1425
|
parameters: {
|
|
1404
1426
|
verbose: { type: 'boolean', description: "Include each operation's parameters (default false). They are what you just sent, so they are omitted unless asked for." },
|
|
@@ -1497,9 +1519,9 @@ export function apply(ctx, config = {}) {
|
|
|
1497
1519
|
return { entries: rendered };
|
|
1498
1520
|
},
|
|
1499
1521
|
}));
|
|
1500
|
-
|
|
1501
|
-
name: '
|
|
1502
|
-
description: 'List pages the shared browser has visited, newest first, from the PERSISTENT browsing history: it survives closing the browser and restarting DSH, so it is how you (or the human) find a page again afterwards. Reopen one by passing its url to
|
|
1522
|
+
registerSafely(defineTool({
|
|
1523
|
+
name: 'dsh_browser_visited',
|
|
1524
|
+
description: 'List pages the shared browser has visited, newest first, from the PERSISTENT browsing history: it survives closing the browser and restarting DSH, so it is how you (or the human) find a page again afterwards. Reopen one by passing its url to dsh_browser_open. Not to be confused with dsh_browser_history, which is only the current session\'s operation log (navigate/click/type) and disappears with the session.',
|
|
1503
1525
|
parameters: {
|
|
1504
1526
|
limit: { type: 'number', description: 'Maximum entries to return (default 30, capped at 200).' },
|
|
1505
1527
|
domain: { type: 'string', description: 'Only visits whose hostname contains this text (case-insensitive).' },
|
|
@@ -1576,9 +1598,9 @@ export function apply(ctx, config = {}) {
|
|
|
1576
1598
|
};
|
|
1577
1599
|
},
|
|
1578
1600
|
}));
|
|
1579
|
-
|
|
1580
|
-
name: '
|
|
1581
|
-
description: 'Replay one recorded browser operation by its history sequence number (from
|
|
1601
|
+
registerSafely(defineTool({
|
|
1602
|
+
name: 'dsh_browser_replay',
|
|
1603
|
+
description: 'Replay one recorded browser operation by its history sequence number (from dsh_browser_history). Navigate/click/type are re-issued against the current page; execute re-runs its script. The replayed step is appended to history as a new entry.',
|
|
1582
1604
|
parameters: {
|
|
1583
1605
|
seq: { type: 'number', required: true, description: 'The history entry sequence number to replay.' },
|
|
1584
1606
|
},
|
|
@@ -1589,7 +1611,7 @@ export function apply(ctx, config = {}) {
|
|
|
1589
1611
|
timeoutMs,
|
|
1590
1612
|
isConcurrencySafe: () => false,
|
|
1591
1613
|
async execute(args, exec) {
|
|
1592
|
-
assertAllowed(state, '
|
|
1614
|
+
assertAllowed(state, 'dsh_browser_replay', taskKey(exec));
|
|
1593
1615
|
const browser = ctx.get('browser');
|
|
1594
1616
|
if (browser === undefined)
|
|
1595
1617
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1598,8 +1620,8 @@ export function apply(ctx, config = {}) {
|
|
|
1598
1620
|
return { replayed: true };
|
|
1599
1621
|
},
|
|
1600
1622
|
}));
|
|
1601
|
-
|
|
1602
|
-
name: '
|
|
1623
|
+
registerSafely(defineTool({
|
|
1624
|
+
name: 'dsh_browser_download',
|
|
1603
1625
|
description: 'Download a URL to a local file, keeping the browser session\'s cookies and login state. Use for fetching files behind authentication or from the current page context. Available on the self-hosted browser; the desktop shell delegates downloads to the real browser UI.',
|
|
1604
1626
|
parameters: {
|
|
1605
1627
|
url: { type: 'string', required: true, description: 'The URL to download.' },
|
|
@@ -1612,7 +1634,7 @@ export function apply(ctx, config = {}) {
|
|
|
1612
1634
|
timeoutMs,
|
|
1613
1635
|
isConcurrencySafe: () => false,
|
|
1614
1636
|
async execute(args, exec) {
|
|
1615
|
-
assertAllowed(state, '
|
|
1637
|
+
assertAllowed(state, 'dsh_browser_download', taskKey(exec));
|
|
1616
1638
|
const browser = ctx.get('browser');
|
|
1617
1639
|
if (browser === undefined)
|
|
1618
1640
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1621,8 +1643,8 @@ export function apply(ctx, config = {}) {
|
|
|
1621
1643
|
return { path: result.path };
|
|
1622
1644
|
},
|
|
1623
1645
|
}));
|
|
1624
|
-
|
|
1625
|
-
name: '
|
|
1646
|
+
registerSafely(defineTool({
|
|
1647
|
+
name: 'dsh_browser_session',
|
|
1626
1648
|
description: 'Show THIS task\'s browser session: its id and open tabs. Each task (DSH session) has its own browser session, so this reflects what your task drives. The window is shared with the human and other tasks, but tab sets and history are isolated per task.',
|
|
1627
1649
|
parameters: {},
|
|
1628
1650
|
output: {
|
|
@@ -1662,9 +1684,9 @@ export function apply(ctx, config = {}) {
|
|
|
1662
1684
|
return { session, tabs: tabs.map(t => ({ id: t.id, url: t.url, active: t.active })) };
|
|
1663
1685
|
},
|
|
1664
1686
|
}));
|
|
1665
|
-
|
|
1666
|
-
name: '
|
|
1667
|
-
description: 'Reset THIS task\'s browser session: close it entirely so the next
|
|
1687
|
+
registerSafely(defineTool({
|
|
1688
|
+
name: 'dsh_browser_reset_session',
|
|
1689
|
+
description: 'Reset THIS task\'s browser session: close it entirely so the next dsh_browser_* call starts a fresh session with one blank tab. Other tasks\' sessions are untouched. Use when a session is in a bad state or you want a clean slate.',
|
|
1668
1690
|
parameters: {},
|
|
1669
1691
|
output: {
|
|
1670
1692
|
schema: { type: 'object', additionalProperties: false, properties: { reset: { type: 'boolean', required: true } } },
|
|
@@ -1673,7 +1695,7 @@ export function apply(ctx, config = {}) {
|
|
|
1673
1695
|
timeoutMs,
|
|
1674
1696
|
isConcurrencySafe: () => false, // closes the whole session; exclusive within a task
|
|
1675
1697
|
async execute(_args, exec) {
|
|
1676
|
-
assertAllowed(state, '
|
|
1698
|
+
assertAllowed(state, 'dsh_browser_reset_session', taskKey(exec));
|
|
1677
1699
|
const browser = ctx.get('browser');
|
|
1678
1700
|
if (browser === undefined)
|
|
1679
1701
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1692,9 +1714,9 @@ export function apply(ctx, config = {}) {
|
|
|
1692
1714
|
return { reset: true };
|
|
1693
1715
|
},
|
|
1694
1716
|
}));
|
|
1695
|
-
|
|
1696
|
-
name: '
|
|
1697
|
-
description: 'Restrict which browser actions are allowed, to prevent stray clicks/navigation. Pass a list of browser tool names (e.g. ["
|
|
1717
|
+
registerSafely(defineTool({
|
|
1718
|
+
name: 'dsh_browser_restrict',
|
|
1719
|
+
description: 'Restrict which browser actions are allowed, to prevent stray clicks/navigation. Pass a list of browser tool names (e.g. ["dsh_browser_snapshot","dsh_browser_content","dsh_browser_click"]) — any other dsh_browser_* call is refused. Pass an empty list or omit to lift the restriction. Never blocked, whatever the list says: the observing tools (snapshot/a11y/content/scrape/screenshot/get_value/wait/challenge/list_tabs/session/history/visited) and the three that lift a restriction or recover from a bad state (restrict/reset/reset_session) — so a restriction can always be read around and undone. IMPORTANT: this is a SOFT guardrail against accidental actions, NOT a security boundary — you (the model) can lift it yourself with an empty list.',
|
|
1698
1720
|
parameters: {
|
|
1699
1721
|
allowed: {
|
|
1700
1722
|
type: 'array',
|
|
@@ -1711,9 +1733,9 @@ export function apply(ctx, config = {}) {
|
|
|
1711
1733
|
async execute(args, exec) {
|
|
1712
1734
|
// Always allowed so the guard can be lifted.
|
|
1713
1735
|
const allowed = args.allowed ?? [];
|
|
1714
|
-
const unknown = allowed.filter((t) => !t.startsWith('
|
|
1736
|
+
const unknown = allowed.filter((t) => !t.startsWith('dsh_browser_'));
|
|
1715
1737
|
if (unknown.length > 0) {
|
|
1716
|
-
throw new Error(`
|
|
1738
|
+
throw new Error(`dsh_browser_restrict: unknown tool name(s) ${unknown.map(t => `"${t}"`).join(', ')} (must start with "dsh_browser_")`);
|
|
1717
1739
|
}
|
|
1718
1740
|
// Empty list (or omitted) lifts this task's restriction; a non-empty list is its
|
|
1719
1741
|
// new allow-list. Per task, not per plugin: this state object is shared by every
|
|
@@ -1727,8 +1749,8 @@ export function apply(ctx, config = {}) {
|
|
|
1727
1749
|
return { restrictedTo: current === undefined ? [] : [...current] };
|
|
1728
1750
|
},
|
|
1729
1751
|
}));
|
|
1730
|
-
|
|
1731
|
-
name: '
|
|
1752
|
+
registerSafely(defineTool({
|
|
1753
|
+
name: 'dsh_browser_auth',
|
|
1732
1754
|
description: 'Export or restore the browser session\'s cookies (login state). Use "flush" to get a JSON cookie list (save it to a private file to persist logins), or "restore" with that list to put logins back (e.g. after the browser host restarted). Available on all three carriers: the self-hosted browser reads its own session, while the desktop sidebar and an installed Chrome/Edge go through CDP. Refused entirely when the credential switch is off. Exported cookies are LIVE CREDENTIALS: treat them as secrets — do not echo them into the conversation, keep them out of logs, and store the list in a private file.',
|
|
1733
1755
|
parameters: {
|
|
1734
1756
|
action: { type: 'string', required: true, enum: ['flush', 'restore'], description: 'flush = export cookies; restore = import cookies.' },
|
|
@@ -1748,7 +1770,7 @@ export function apply(ctx, config = {}) {
|
|
|
1748
1770
|
timeoutMs,
|
|
1749
1771
|
isConcurrencySafe: () => false,
|
|
1750
1772
|
async execute(args, exec) {
|
|
1751
|
-
assertAllowed(state, '
|
|
1773
|
+
assertAllowed(state, 'dsh_browser_auth', taskKey(exec));
|
|
1752
1774
|
const browser = ctx.get('browser');
|
|
1753
1775
|
if (browser === undefined)
|
|
1754
1776
|
throw new Error('tool-browser: browser service unavailable');
|
|
@@ -1761,7 +1783,7 @@ export function apply(ctx, config = {}) {
|
|
|
1761
1783
|
// empty call returned {restored: 0} and rendered 'Restored 0 cookies.', which
|
|
1762
1784
|
// reads as success and would leave the caller believing the login was restored.
|
|
1763
1785
|
if (!Array.isArray(args.cookies) || args.cookies.length === 0) {
|
|
1764
|
-
throw new Error('
|
|
1786
|
+
throw new Error('dsh_browser_auth: action "restore" needs a non-empty "cookies" array (get one from action "flush" on the same profile)');
|
|
1765
1787
|
}
|
|
1766
1788
|
const list = args.cookies;
|
|
1767
1789
|
const restored = await browser.restoreAuth(session, list);
|