Files
openswarm/backend/apps/agents/browser/browser_schema.py
T

1229 lines
66 KiB
Python

"""
Static schema + prompt blob for the browser sub-agent.
One big single-responsibility constant file: tool schema, action map, system
prompt, and the turn/report invariants. Exceeds the 300-LOC soft ceiling on
purpose because it is one cohesive data blob, not multiple responsibilities.
"""
from backend.apps.agents.browser.intervention_copy import (
INTERVENTION_SECTION,
LOOP_AWARENESS_INTERVENTION_PHRASE,
)
# Two prompt levers that A/B-proved out and now ship unconditionally. THINK_SHORTER (no prose beside action tools; ReportProgress IS the thinking) cut per-turn output ~28% and roughly halved narration turns. MERGE_VERIFY (a confirmed `expect` is the proof, skip the re-check) drops a wasted round-trip at the end.
P_THINK_SHORTER = (
"Do NOT write a free-text sentence next to your action tools: your ReportProgress "
"fields ARE your thinking, and a separate prose explanation just repeats them and slows "
"the turn. Don't narrate to the user as you go either. When the task is done you finish by "
"calling the Done tool (never by typing a sentence); every other turn is ReportProgress + "
"tools, no prose.\n"
)
P_MERGE_VERIFY = (
"When that `expect` CONFIRMS (the result says 'Confirmed: ...'), that IS your "
"verification: go STRAIGHT to calling Done. Do NOT spend an "
"extra screenshot or read turn to re-check what the confirmation already proved, that "
"is a wasted round-trip. Only take a separate verification step when `expect` came "
"back 'NOT confirmed' or you forgot to pass one.\n"
)
# The change an action should cause, declared by the agent and CONFIRMED after the action runs (success is observed, never assumed). A hit returns fast; a miss tells the agent it may not have worked instead of letting it claim a false success.
P_EXPECT_DESC = {
"type": "string",
"description": (
"Optional but recommended: LITERAL text that should be VISIBLE on the page "
"after this action; an exact button label, a person's name, the exact text you "
"just typed. Never a description of the change: 'message appears in box' is not "
"page text, can never match, and will always come back NOT confirmed. It's "
"checked right after, so you learn whether it actually worked. REQUIRED for "
"anything you can't undo (Send/Submit/Pay/Post): set it to proof the action "
"landed (for typing, the typed text itself)."
),
}
BROWSER_TOOLS_SCHEMA = [
{
"name": "ReportProgress",
"description": (
"Record your assessment of the previous action and your plan for the "
"next one. You MUST call this BEFORE any browser action tools in every "
"turn (after the very first turn). This is how you reflect on what just "
"happened, track what you've learned about this site, and articulate what "
"you're trying to do next. Skipping it is not allowed and will be rejected."
),
"input_schema": {
"type": "object",
"properties": {
"evaluation_previous": {
"type": "string",
"description": (
"OPTIONAL. Only when the previous action SURPRISED you (failed, "
"landed somewhere unexpected): say what happened and why, briefly. "
"Omit entirely when the attached page state already shows the outcome."
),
},
"working_memory": {
"type": "string",
"description": (
"Short notes about what you've learned about this site so far; "
"selectors that work, keyboard shortcuts, layout quirks, what "
"you've tried that failed. Carry this forward across turns."
),
},
"next_goal": {
"type": "string",
"description": (
"What you're trying to achieve with the action(s) you're about "
"to take next. Be concrete."
),
},
},
"required": ["working_memory", "next_goal"],
},
},
{
"name": "Done",
"description": (
"Call this the moment the task is finished (or you've hit a wall you "
"can't get past) to deliver your final reply to the user. This ends the "
"run. Do NOT type a sentence to finish, always finish by calling Done."
),
"input_schema": {
"type": "object",
"properties": {
"message": {
"type": "string",
"description": (
"What the user reads, so write it like a quick text to a friend: "
"what you did and the proof a person actually cares about (the name, "
"the time, what's now on screen). One or two plain sentences. Use ZERO "
"interface words, no 'button', 'box', 'textbox', 'composer', 'field', "
"'element', index numbers, or coordinates, and don't mechanically repeat "
"the task back. If you couldn't finish, say what's missing in that same "
"plain voice and set success to false."
),
},
"success": {
"type": "boolean",
"description": (
"true if you accomplished what the user asked, false if you couldn't "
"(login wall, missing info, something blocked you). Default true."
),
},
"keep_open": {
"type": "boolean",
"description": (
"Set true ONLY when the result IS the open page and the user will keep "
"using it right now: a video or audio playing, a page you opened for them "
"to read or watch, a download you started, or a place left ready for them "
"to take over. The browser then stays put instead of closing. Leave false "
"(default) for info tasks where you just look something up and report the "
"answer back, since there's nothing left to keep on screen."
),
},
},
"required": ["message"],
},
},
{
"name": "BrowserScreenshot",
"description": (
"Capture a screenshot of the browser page. Returns the screenshot as a "
"base64-encoded PNG image. Use this to see what is currently displayed. "
"Elements from your last BrowserListInteractives come back with numbered "
"colored boxes drawn on them, the same numbers you click with, so you can "
"go straight from what you see to BrowserClickIndex."
),
"input_schema": {
"type": "object",
"properties": {
"annotate": {
"type": "boolean",
"description": (
"Default true. Pass false for a clean, unannotated shot, e.g. "
"when capturing proof of a completed action."
),
},
},
"required": [],
},
},
{
"name": "BrowserGetText",
"description": (
"Get the visible text content of the browser page. Returns up to 15000 characters."
),
"input_schema": {
"type": "object",
"properties": {},
"required": [],
},
},
{
"name": "BrowserExtract",
"description": (
"Pull STRUCTURED data off the current page in one call: say what you "
"want (and optionally the JSON shape) and a fast helper model reads "
"the page text for you, returning just the JSON. Use this instead of "
"BrowserGetText whenever you need specific fields (each result's name "
"+ headline + URL, a listing's price/title/rating, table rows): the "
"page's 15k chars stay out of your context, so it's faster and "
"cheaper than reading it yourself. Read-only. If the data isn't on "
"the page you get {\"not_found\": true}, never a guess."
),
"input_schema": {
"type": "object",
"properties": {
"instruction": {
"type": "string",
"description": (
"What to extract, concretely (e.g. 'every search result: "
"full name, headline, profile URL')."
),
},
"schema": {
"type": "object",
"description": "Optional JSON schema describing the exact output shape.",
},
},
"required": ["instruction"],
},
},
{
"name": "BrowserSaveData",
"description": (
"Save a LARGE dataset you've assembled on the page straight to a file. "
"Use this for 'every comment / all N results / the full list' once you've "
"collected it into a page variable: trying to return hundreds of rows in your "
"reply TRUNCATES, so you'd otherwise waste turns chunking it. Give a JS "
"expression that returns the data as a string (almost always "
"JSON.stringify(window.__yourVar)) plus a filename; the whole dataset is "
"written to the workspace and you get back just the file path, not the data. "
"It only writes your own workspace file (never the web). Put the returned path "
"in your Done message so the user knows where it landed."
),
"input_schema": {
"type": "object",
"properties": {
"expression": {
"type": "string",
"description": (
"JS that returns the data as a string, e.g. "
"JSON.stringify(window.__rows) or a CSV string you build."
),
},
"filename": {
"type": "string",
"description": (
"Plain data filename, e.g. results.json or comments.csv "
"(allowed: .json .ndjson .csv .tsv .txt .md)."
),
},
},
"required": ["expression", "filename"],
},
},
{
"name": "BrowserGetConsole",
"description": (
"Read the page's OWN recent JavaScript console warnings and errors "
"(uncaught exceptions, failed resource loads like a 403/500, React "
"errors). Use this when an action isn't working and the page looks "
"fine: it tells you WHY the page is broken (an API call failed, the "
"app crashed) so you fix the real cause instead of retrying blindly. "
"Read-only; returns nothing if the page logged no warnings or errors."
),
"input_schema": {
"type": "object",
"properties": {},
"required": [],
},
},
{
"name": "BrowserNavigate",
"description": (
"Navigate the browser to a URL. Use a normal page URL a person would see. "
"Do NOT point it at a raw JSON/API endpoint (a /api/... or search-JSON URL "
"like Instagram's web/search/topsearch): that paints an unreadable data wall "
"in the card. To READ a site's own API, use BrowserReplayRoute, which fetches "
"the JSON silently without disturbing the page."
),
"input_schema": {
"type": "object",
"properties": {
"url": {"type": "string", "description": "The URL to navigate to."},
},
"required": ["url"],
},
},
{
"name": "BrowserClick",
"description": "Click an element identified by a CSS selector. Use BrowserGetElements first to discover valid selectors.",
"input_schema": {
"type": "object",
"properties": {
"selector": {"type": "string", "description": "CSS selector of the element to click."},
"expect": P_EXPECT_DESC,
},
"required": ["selector"],
},
},
{
"name": "BrowserType",
"description": "Type text into an input element. Clears existing value first.",
"input_schema": {
"type": "object",
"properties": {
"selector": {"type": "string", "description": "CSS selector of the input element."},
"text": {"type": "string", "description": "The text to type."},
},
"required": ["selector", "text"],
},
},
{
"name": "BrowserEvaluate",
"description": "Evaluate a JavaScript expression in the browser page and return the result.",
"input_schema": {
"type": "object",
"properties": {
"expression": {"type": "string", "description": "JavaScript expression to evaluate."},
},
"required": ["expression"],
},
},
{
"name": "BrowserGetElements",
"description": (
"Get a list of interactive elements on the page with CSS selectors. "
"Call this BEFORE clicking or typing so you know which selectors are valid."
),
"input_schema": {
"type": "object",
"properties": {
"selector": {
"type": "string",
"description": "Optional CSS selector to scope the search (e.g. 'form', '#main'). Defaults to 'body'.",
},
},
"required": [],
},
},
{
"name": "BrowserScroll",
"description": (
"Scroll the page up or down. Automatically finds the correct scrollable "
"container (works on SPAs like Notion, Gmail, etc. that use nested scroll "
"containers instead of window-level scrolling). Returns scroll position info "
"including whether top/bottom has been reached."
),
"input_schema": {
"type": "object",
"properties": {
"direction": {
"type": "string",
"enum": ["up", "down"],
"description": "Scroll direction. Defaults to 'down'.",
},
"amount": {
"type": "number",
"description": "Pixels to scroll. Defaults to 500.",
},
},
"required": [],
},
},
{
"name": "BrowserListInteractives",
"description": (
"Get a NUMBERED LIST of interactive elements on the page using the "
"browser's accessibility tree. Returns elements like [1]<button \"Like\">, "
"[2]<link \"Settings\">, etc. Use this BEFORE BrowserClickIndex. This is "
"the PREFERRED way to discover clickable elements on hostile sites "
"(Tinder, Instagram, TikTok) where CSS selectors fail because the page "
"uses unlabeled <div>s; the accessibility tree sees roles and names "
"even when raw HTML doesn't expose them. Much more reliable than "
"BrowserGetElements (which uses CSS selectors). Numbers are STABLE "
"across looks (same number = same element as your previous list, so "
"you can act on a remembered index without re-listing if the page "
"hasn't changed). When several rows share a label, each carries "
"ctx=\"...\" naming the card/section it sits in, so pick the row whose "
"ctx matches your target (the right person's \"Message\" button). "
"NOTE: every mutating action's result already ends with '[page state "
"after action]', a fresh copy of this list, so right after acting you "
"do NOT need this; call it only to re-orient or when the attached "
"list was truncated."
),
"input_schema": {
"type": "object",
"properties": {},
"required": [],
},
},
{
"name": "BrowserClickIndex",
"description": (
"SOLO click, reserved for two cases: (1) the final IRREVERSIBLE step "
"(Send/Submit/Pay/Post), always alone with `expect` proof, and (2) "
"filling a text box via `text` (focused directly by node, so it works "
"inside messaging/compose overlays where coordinate clicks miss). "
"Routine clicks belong inside BrowserBatch as click_index sub-actions. "
"Uses native OS-level mouse events (event.isTrusted=true) so it works "
"on sites that filter synthetic JS events. If the click returns "
"'index no longer valid', the page changed; re-list and retry."
),
"input_schema": {
"type": "object",
"properties": {
"index": {
"type": "integer",
"description": "The numeric index from BrowserListInteractives (1-based).",
},
"text": {
"type": "string",
"description": (
"Optional. If the index is a text box, focus it and type this "
"whole string in one call (no character-by-character). Use this "
"to fill a compose/message box reliably."
),
},
"expect": P_EXPECT_DESC,
},
"required": ["index"],
},
},
{
"name": "BrowserActVerified",
"description": (
"Run a short SEQUENCE of dependent UI steps (2-4) in one call, where each "
"step must take effect before the next: open a menu then pick an item, "
"fill a field then the next one, expand a section then click inside it. "
"Each step names its target ELEMENT BY NAME (resolved fresh against the "
"live page at act time, so a stale index can't bite) and is VERIFIED in "
"code (did the expected change actually happen), with one automatic "
"re-aim on a miss. Steps:\n"
"- { action: 'click', target: '<element name>', role?: 'button'|'link'|..., "
"expect?: 'appeared:<text>'|'gone:<text>'|'url_changed'|'changed' }\n"
"- { action: 'fill', target: '<field name>', text: '<text to type>' } "
"(auto-verifies the text committed)\n"
"Execution stops at the first step that can't be verified and you get "
"per-step results plus what went wrong. NEVER put an irreversible action "
"(send/submit/post/pay/delete/confirm) here; those stay SOLO clicks with "
"an `expect` proof, as always."
),
"input_schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"maxItems": 4,
"items": {
"type": "object",
"properties": {
"action": {"type": "string", "enum": ["click", "fill"]},
"target": {"type": "string"},
"role": {"type": "string"},
"text": {"type": "string"},
"expect": {"type": "string"},
},
"required": ["action", "target"],
},
},
},
"required": ["steps"],
},
},
{
"name": "BrowserBatch",
"description": (
"Low-level action batch for INDEPENDENT actions (navigate, scroll, press) "
"or when you must act by index/point. For a sequence of DEPENDENT named "
"click/fill steps (open menu then pick item, fill then fill), prefer "
"BrowserActVerified: it self-verifies each step and re-aims in code. "
"Every mutation (navigate, "
"click, type, press, scroll) is a sub-action in this array, 1-5 per "
"call; an array of one is fine when that is genuinely all you know. "
"Each sub-action executes in order with the URL captured before/after. "
"If the URL changes mid-batch (the page navigated), the rest is "
"aborted and you get a partial result plus fresh state, so a "
"conservative batch costs nothing.\n\n"
"Sub-action types and their params:\n"
"- click_index: { index: int }\n"
"- press_key: { key: str }\n"
"- click_point: { xPercent: number, yPercent: number, hold_ms?: int } "
"(tap a screen point; for canvas apps/games)\n"
"- type: { selector: str, text: str }\n"
"- click: { selector: str }\n"
"- scroll: { direction?: 'up'|'down', amount?: int }\n"
"- wait: { milliseconds?: int }\n"
"- navigate: { url: str }\n"
"- list_interactives: { } (read the page; ONLY valid as the LAST sub-action)\n\n"
"End a batch with list_interactives to fold a click -> wait -> read into "
"ONE turn: e.g. click a button, wait for it to settle, then read the "
"result, all without a second round-trip.\n"
"Example: { actions: [{type: 'click_index', params: {index: 1}}, "
"{type: 'wait', params: {milliseconds: 4000}}, "
"{type: 'list_interactives', params: {}}] }"
),
"input_schema": {
"type": "object",
"properties": {
"actions": {
"type": "array",
"maxItems": 5,
"items": {
"type": "object",
"properties": {
"type": {
"type": "string",
"enum": ["click_index", "press_key", "click_point", "type", "wait", "scroll", "navigate", "click", "list_interactives"],
},
"params": {"type": "object"},
},
"required": ["type", "params"],
},
},
},
"required": ["actions"],
},
},
{
"name": "BrowserPressKey",
"description": (
"Press a keyboard key (or key combination) on the page using a real native "
"input event. Use this for keyboard shortcuts when JS-dispatched events get "
"ignored; sites like Tinder, Slack, Notion, Gmail listen for trusted key "
"events. Examples: 'ArrowLeft', 'ArrowRight', 'Enter', 'Escape', 'Tab', "
"'Space', single letters like 'a'. Prefer this over BrowserEvaluate with "
"dispatchEvent for keyboard shortcuts."
),
"input_schema": {
"type": "object",
"properties": {
"key": {
"type": "string",
"description": (
"The key to press. Use JS KeyboardEvent.key names like "
"'ArrowUp', 'ArrowDown', 'Enter', 'Escape', 'Tab', 'Space', "
"'Backspace', or a single character like 'a'."
),
},
},
"required": ["key"],
},
},
{
"name": "BrowserClickPoint",
"description": (
"Tap/click at a point on the screen using a real native mouse event, "
"WITHOUT needing a DOM element. This is the way to operate a <canvas> "
"app or game (the kind with no clickable HTML elements): you click a "
"spot the way a person does. Give the position as a PERCENT of the view "
"(xPercent/yPercent, 0-100, with 0,0 = top-left and 50,50 = center), "
"read off the screenshot. Optional hold_ms presses and holds (e.g. a "
"charge-up or a platformer jump). For element-based pages prefer "
"BrowserClickIndex; use this when there is nothing in the element list "
"to click."
),
"input_schema": {
"type": "object",
"properties": {
"xPercent": {
"type": "number",
"description": "Horizontal position as a percent of view width (0=left, 100=right).",
},
"yPercent": {
"type": "number",
"description": "Vertical position as a percent of view height (0=top, 100=bottom).",
},
"hold_ms": {
"type": "number",
"description": "Optional. Milliseconds to hold the button down before releasing (default 0 = a tap). Max 5000.",
},
"button": {
"type": "string",
"enum": ["left", "right", "middle"],
"description": "Mouse button; defaults to left.",
},
},
"required": ["xPercent", "yPercent"],
},
},
{
"name": "BrowserWait",
"description": (
"Wait for the page to be READY after navigation or an action. This is SMART: "
"it returns as soon as the page settles visually (its DOM stops changing), so "
"the duration is just an upper bound, not a fixed sleep, pass a generous cap "
"(e.g. 4000) without worrying about wasted time. Best of all, pass `until` with "
"the thing you expect to appear (a button label, result text, the compose box) "
"and it returns the INSTANT that shows up, so you wait for what you actually "
"need instead of guessing. Min 100, max 10000."
),
"input_schema": {
"type": "object",
"properties": {
"milliseconds": {
"type": "number",
"description": "Upper-bound wait in milliseconds. Defaults to 1000.",
},
"until": {
"type": "string",
"description": (
"Optional. A specific button label, visible text, or CSS selector "
"you expect to appear (e.g. the person's name you searched, a "
"placeholder like 'Write a message', or 'button[type=submit]'). The "
"wait ends the moment it's present and visible. Be specific, not a "
"generic word like 'Message'."
),
},
},
"required": [],
},
},
{
"name": "BrowserListRoutes",
"description": (
"List the site's own API endpoints that were captured while you "
"browsed it (GET routes, safe to call directly). When you need to "
"re-fetch data you already loaded once (search results, a list, a "
"detail page), calling the API with BrowserReplayRoute is far faster "
"than re-navigating and re-scraping the UI. Returns nothing until "
"you've actually used the page. Read-only."
),
"input_schema": {"type": "object", "properties": {}, "required": []},
},
{
"name": "BrowserReplayRoute",
"description": (
"Directly call one of the site's captured GET endpoints (from "
"BrowserListRoutes) and get the raw response, skipping the UI. "
"ONLY safe read-only GET/HEAD requests on the current site are "
"allowed; anything that changes data (add to cart, send, delete, "
"post) must be done through the UI by clicking. Use this to read "
"data fast, not to perform actions."
),
"input_schema": {
"type": "object",
"properties": {
"url": {"type": "string", "description": "The endpoint URL to GET (from BrowserListRoutes; same site only)."},
"method": {"type": "string", "enum": ["GET", "HEAD"], "description": "Defaults to GET."},
},
"required": ["url"],
},
},
{
"name": "BrowserApiWrite",
"description": (
"Perform a WRITE on the CURRENT site through the site's OWN API using your "
"already-logged-in session, instead of clicking the UI. This is deterministic "
"and far more reliable: no captcha, no button to miss, no selector to drift, "
"and it hands back the site's REAL receipt (the new post/comment's id and "
"permalink) as proof it landed. Only some sites are supported so far "
"(currently Reddit: comment, reply, post, edit, delete). If the current site "
"has no built-in adapter, you can still do it the GENERAL way: set action='route' "
"with the site's own write endpoint (method + url + body) taken from BrowserListRoutes, "
"and it replays that request with your session. If neither works you get a clean miss, "
"just do the write through the UI instead. This IS a real write: call it ONCE, and the "
"receipt is your confirmation, do not re-check or re-fire it."
),
"input_schema": {
"type": "object",
"properties": {
"action": {"type": "string", "description": "comment, reply, post, edit, delete (built-in adapter), or route (general: replay a captured write endpoint)."},
"parent_id": {"type": "string", "description": "comment/reply: fullname of the post/comment you're replying to (e.g. t3_abc, t1_xyz)."},
"thing_id": {"type": "string", "description": "edit/delete: fullname of your OWN post/comment (e.g. t1_xyz)."},
"text": {"type": "string", "description": "The body text (comment/reply/post/edit)."},
"subreddit": {"type": "string", "description": "post: the subreddit name, without the r/ prefix."},
"title": {"type": "string", "description": "post: the post title."},
"url": {"type": "string", "description": "post: a link URL; OR route: the write endpoint's full URL from BrowserListRoutes."},
"method": {"type": "string", "description": "route: the endpoint's HTTP method (POST, PUT, PATCH, DELETE)."},
"body": {"type": "object", "description": "route: the JSON body to send, matching the endpoint's captured shape, with YOUR content in the text field(s)."},
},
"required": ["action"],
},
},
{
"name": "BrowserRepeatFlow",
"description": (
"Repeat a mechanical flow you JUST did, fast, for many inputs, without "
"re-screenshotting or re-analyzing the page each time. After you've done "
"ONE item the slow way (e.g. searched and read one person), call this "
"with the step template and the remaining inputs and they run at machine "
"speed: zero screenshots, zero extra thinking. Write the steps using "
"{{value}} wherever the input varies. Each iteration is verified; any "
"item whose page doesn't match falls back and is reported so you can "
"handle it yourself, it never pretends. It HANDS BACK each item's read "
"data (the last step's output, capped), keyed by value, so a read loop "
"actually delivers ('Read 5 of 5: - ada: ...'). Use this for SEARCH / "
"READ / NAVIGATE loops. It REFUSES irreversible steps (Send, Submit, "
"Connect, Post, Pay, Delete, message composers): do those one at a time. "
"For reading data, a 'replay_route' step (hit a captured API endpoint) is "
"far faster and cheaper than navigating the UI per item."
),
"input_schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"description": "The flow for ONE item, in order. Put {{value}} where the input varies.",
"items": {
"type": "object",
"properties": {
"action": {"type": "string", "enum": ["navigate", "get_text", "evaluate", "type", "click", "press_key", "scroll", "replay_route"]},
"url": {"type": "string", "description": "for navigate / replay_route"},
"selector": {"type": "string", "description": "for type"},
"text": {"type": "string", "description": "for type"},
"role": {"type": "string", "description": "for click (e.g. button, link)"},
"name": {"type": "string", "description": "for click: the visible text"},
"key": {"type": "string", "description": "for press_key (e.g. Enter)"},
"expression": {"type": "string", "description": "for evaluate (JS)"},
"direction": {"type": "string", "description": "for scroll"},
"amount": {"type": "integer", "description": "for scroll"},
},
"required": ["action"],
},
},
"values": {
"type": "array",
"items": {"type": "string"},
"description": "The inputs to run the flow for (the remaining items after the one you already did).",
},
},
"required": ["steps", "values"],
},
},
{
"name": "BrowserDetectWebMCP",
"description": (
"Check whether the current page declares its own WebMCP tools "
"(navigator.modelContext). If a site exposes tools this way, they "
"are the fastest, most reliable path; prefer them over scraping the "
"UI. Most sites don't support this yet, so it usually reports none; "
"in that case just use the normal browser tools. Read-only."
),
"input_schema": {"type": "object", "properties": {}, "required": []},
},
{
"name": "BrowserListSkills",
"description": (
"List the shortcuts (learned skills) you already have for the CURRENT "
"site. Each is a previously-completed task you can repeat fast. Useful "
"when a task feels familiar: a near-match skill may let you adapt "
"instead of figuring the site out from scratch. Returns task summaries "
"+ how many times each has been reused. Read-only."
),
"input_schema": {"type": "object", "properties": {}, "required": []},
},
{
"name": "BrowserDeprecateSkill",
"description": (
"Throw away a learned shortcut for this site that has gone stale (the "
"page changed, or replaying it no longer works), so it stops being "
"used. Pass the task text exactly as shown by BrowserListSkills. Use "
"this when you realize a saved shortcut is wrong; the correct version "
"will be re-learned the next time you do the task successfully."
),
"input_schema": {
"type": "object",
"properties": {
"task": {"type": "string", "description": "The skill's task text (from BrowserListSkills) to remove."},
},
"required": ["task"],
},
},
{
"name": "RequestHumanIntervention",
"description": (
"Request the user's help when you encounter an obstacle you cannot solve "
"programmatically; captchas, login prompts, cookie consent walls, "
"two-factor authentication, or any blocking popup. The agent will pause "
"until the user resolves the issue and clicks Continue."
),
"input_schema": {
"type": "object",
"properties": {
"problem": {
"type": "string",
"description": (
"One short sentence describing the obstacle. Keep it under "
"15 words. Example: 'Login required; please sign in to X/Twitter.'"
),
},
"instruction": {
"type": "string",
"description": (
"One short sentence telling the user what to do. Keep it under "
"15 words. Example: 'Log in with your credentials, then click Done.'"
),
},
},
"required": ["problem", "instruction"],
},
},
{
"name": "BrowserDeleteItem",
"description": (
"Remove ONE item visible on the CURRENT page, named by a distinctive snippet of its "
"own text. Navigate to where the item lives FIRST (your profile, the thread) so it is "
"on screen, then call this. It opens that item's overflow / 'More' menu, clicks "
"Delete/Remove, confirms, and verifies the item is gone. Use this instead of clicking "
"the '...' menu yourself, that menu is small and lazy-rendered and hand-clicking it is "
"unreliable. Only your OWN items expose a Delete option. Returns whether it was "
"verifiably removed; if it says the item isn't on the page, navigate to where it lives "
"and retry."
),
"input_schema": {
"type": "object",
"properties": {
"target_text": {
"type": "string",
"description": (
"A distinctive exact snippet of the target item's own text, long enough "
"to match only that one item (a phrase or id from the post/comment)."
),
},
},
"required": ["target_text"],
},
},
]
# Schema-forced batching: the model ignored every prompt-level batching invitation (0 adoptions across 8 measured runs), so the single-step mutating tools are not offered to it at all; acting means a BrowserBatch array, and the one deliberate solo path is BrowserClickIndex (irreversible step with expect, or a text-box fill). Executors and replay still support everything.
P_SOLO_MUTATORS_HIDDEN = {"BrowserNavigate", "BrowserClick", "BrowserType", "BrowserScroll", "BrowserPressKey", "BrowserClickPoint"}
MODEL_VISIBLE_TOOLS = [t for t in BROWSER_TOOLS_SCHEMA if t["name"] not in P_SOLO_MUTATORS_HIDDEN]
ACTION_MAP = {
"BrowserScreenshot": "screenshot",
"BrowserGetText": "get_text",
"BrowserGetConsole": "get_console",
"BrowserNavigate": "navigate",
"BrowserClick": "click",
"BrowserType": "type",
"BrowserEvaluate": "evaluate",
"BrowserGetElements": "get_elements",
"BrowserScroll": "scroll",
"BrowserWait": "wait",
"BrowserPressKey": "press_key",
"BrowserListInteractives": "list_interactives",
"BrowserClickIndex": "click_index",
"BrowserClickPoint": "click_point",
"BrowserBatch": "batch",
"BrowserDetectWebMCP": "detect_webmcp",
"BrowserListRoutes": "list_routes",
"BrowserReplayRoute": "replay_route",
# Internal replay primitive (skill replay calls it directly; not in the LLM-facing schema). Re-resolves a click target by role+name.
"BrowserClickByName": "click_by_name",
# Internal: structural composer finder (send-script fallback; not LLM-facing). Ranks editable regions, marks the winner, optionally fills+verifies in-page.
"BrowserFindComposer": "find_composer",
}
# --- App agent: driving an OpenSwarm-built app via its native bridge ---------
# Apps expose window.OPENSWARM_APP = { describe(), getState(), invoke(name,args) }.
# The agent reads structure/state and acts through that bridge in single
# executeJavaScript round-trips, no screenshots or accessibility tree. These
# three tools are translated to BrowserEvaluate in execute_browser_tool, so they
# need no frontend command-handler changes.
APP_TOOLS_SCHEMA = [
{
"name": "AppDescribe",
"description": (
"Read the app's rules and CURRENT list of actions, straight from the "
"app itself (window.OPENSWARM_APP.describe()). Returns "
"{rules, controls, __rev}: rules is what the app is and its objective, "
"controls is an array of {name, args, description, keys}, and __rev is "
"a revision number. (Older apps may return a bare array of controls.) "
"These are ALREADY front-loaded into your first message, so you rarely "
"need to call this. The app's controls are DYNAMIC: call this again "
"ONLY when AppGetState reports a changed __rev (e.g. after an AppInvoke "
"added or removed actions). Returns null if the app exposes no bridge "
"(then fall back to BrowserListInteractives/BrowserScreenshot)."
),
"input_schema": {"type": "object", "properties": {}, "required": []},
},
{
"name": "AppGetState",
"description": (
"Read a small JSON snapshot of the app's current state "
"(window.OPENSWARM_APP.getState()). Use it to check what's on screen "
"and to verify an action landed. The snapshot includes __rev, the "
"controls revision: if it differs from the __rev you were given, the "
"controls changed, so call AppDescribe to refresh them. Returns null "
"if the bridge is absent."
),
"input_schema": {"type": "object", "properties": {}, "required": []},
},
{
"name": "AppInvoke",
"description": (
"Perform one app action by name with arguments "
"(window.OPENSWARM_APP.invoke(name, args)). The name MUST be one that "
"AppDescribe just returned; never invent or modify actions, only invoke "
"the ones the app exposes. Returns the action's result, or an "
"{__error__} if it threw. After invoking, re-AppDescribe if the action "
"may have changed which controls exist."
),
"input_schema": {
"type": "object",
"properties": {
"name": {"type": "string", "description": "Action name from AppDescribe."},
"args": {
"type": "object",
"description": "Arguments object for the action; {} if none.",
},
},
"required": ["name"],
},
},
]
# Bridge tool names -> the JS the app exposes. execute_browser_tool maps these to
# a BrowserEvaluate. Each expression returns a JSON string (so it round-trips as
# text) and never throws (errors are returned as JSON).
APP_BRIDGE_TOOLS = {"AppDescribe", "AppGetState", "AppInvoke"}
# Lean toolset for app mode: bridge tools first, then a UI-driving fallback for
# apps that don't expose the bridge. Built by name-selecting the shared defs so
# their schemas stay in one place.
P_APP_FALLBACK_TOOL_NAMES = [
"ReportProgress", "Done",
"BrowserScreenshot", "BrowserGetText",
"BrowserListInteractives", "BrowserClickIndex", "BrowserBatch",
# Human-style native input for apps with no clickable DOM (canvas games):
# press real keys and tap real screen points the way a person plays.
"BrowserPressKey", "BrowserClickPoint",
]
p_app_fallback_tools = [t for t in BROWSER_TOOLS_SCHEMA if t["name"] in P_APP_FALLBACK_TOOL_NAMES]
# ReportProgress + Done lead, then the bridge tools, then UI fallback.
APP_VISIBLE_TOOLS = (
[t for t in p_app_fallback_tools if t["name"] in ("ReportProgress", "Done")]
+ APP_TOOLS_SCHEMA
+ [t for t in p_app_fallback_tools if t["name"] not in ("ReportProgress", "Done")]
)
SYSTEM_PROMPT = (
"You are a website-agnostic browser automation agent. You can operate on ANY "
"website the user is signed into; social media, dating apps, email, productivity "
"tools, dashboards, ecommerce, anything. Assume the user has already logged in.\n\n"
"## Plan once up front, then execute tersely (this is how you stay fast)\n"
"Your first turn or two, once you can see the starting page, is your ONE planning "
"window: lay out the whole route in a few lines, the navigations, which buttons and "
"links you expect to click, what you'll read, and roughly where they sit. Treat it as "
"a guideline, not a contract: the live page may differ and you'll adjust, that's fine. "
"This plan stays in the conversation history, so you never re-derive it.\n"
"After that, go TERSE. Every later turn is mostly action, not narration: fire the next "
"step (batched whenever the sequence is known, see BrowserBatch) with a one-line note, "
"and lean on the plan already in context instead of re-explaining it. Re-plan out loud "
"ONLY when the page clearly contradicts your plan. Verbose per-turn prose is the single "
"biggest thing that slows you down, so once the plan exists, keep execution turns short.\n\n"
"## Jump straight to deep URLs\n"
"The fastest navigation is a URL you construct yourself. Most sites expose search "
"and entities as URL patterns: a LinkedIn people search is "
"linkedin.com/search/results/people/?keywords=NAME, a Google search is "
"google.com/search?q=..., a YouTube search is youtube.com/results?search_query=..., "
"an Amazon search is amazon.com/s?k=.... When the task names a thing to find, "
"BrowserNavigate DIRECTLY to the site's search-results URL for it instead of loading "
"the homepage and driving its search UI; one navigate replaces three clicks and two "
"waits. Fall back to the UI only when you don't know the site's URL pattern.\n\n"
"## Required output structure: ReportProgress before every action\n"
"Before ANY action tool (BrowserBatch, BrowserClickIndex, BrowserEvaluate), "
"you MUST call the ReportProgress tool in the SAME turn. "
"ReportProgress takes these short fields:\n"
"- evaluation_previous (OPTIONAL): include ONLY when the last action surprised you "
"(failed, wrong page); when the attached page state already tells the story, omit it.\n"
"- working_memory: what have you learned about this site? what worked, what didn't?\n"
"- next_goal: what specifically are you trying to do with the next action?\n"
"After your first planning turn, keep every field TELEGRAPHIC, a few words each, "
"not sentences (e.g. next_goal: 'click result 1'). "
"Terse means fewer WORDS, never fewer FACTS: always keep the one detail the next step "
"needs (the exact selector, index, or value). Each token you write is generated one at a "
"time and is the main thing that slows a turn, so write the fewest that still carry the "
"plan forward. Only write working_memory when you learn something NEW this turn; else 'none'.\n"
+ P_THINK_SHORTER +
"Emit ReportProgress and your action tool(s) together in the same response. "
"If you skip ReportProgress, your action tools will be REJECTED with an error "
"and you will have to retry. This is not optional. Read-only tools "
"(BrowserScreenshot, BrowserGetText, BrowserGetConsole, BrowserGetElements, BrowserWait) do not "
"require ReportProgress.\n\n"
"## Act and confirm: trust only what you observe\n"
"Success is OBSERVED, never assumed. On any action that changes the page (click, "
"type, navigate), add `expect`: the change it should cause (a label, text, or the "
"element you expect to see). It's confirmed right after, a hit comes back fast and "
"you move on; a 'NOT confirmed' means it may not have worked, so check the page "
"instead of pressing on. For anything you CANNOT undo (Send, Submit, Pay, Post): "
"first make sure the goal isn't already done (e.g. your message isn't already the "
"last one in the thread), pass `expect` set to proof it landed, and NEVER fire it a "
"second time unless you have verified the first did NOT go through. This is how you "
"avoid both ghost-successes and double-sends.\n"
"When you arrive (or wake) with the target's message THREAD or composer ALREADY OPEN, "
"that IS your thread, commit to it. The open thread's header name (and your own earlier "
"messages to that person sitting in it) ARE the recipient proof; do NOT navigate away to "
"open their profile or re-search 'just to be sure', that round-trips for nothing and can "
"even land you on the wrong surface. The plan from an open composer is fixed and short: "
"type the message, click Send, then verify ONCE. Execute it; do not re-derive it, re-open "
"anything, or re-confirm with extra screenshots/DOM/JS probes what the open thread already "
"shows. That re-verification is the single biggest waste of turns.\n"
"If you already saw the Send button (or Submit/Post) at an index, REMEMBER that "
"number: typing into the composer does NOT move it, so after you type, click that "
"same remembered index directly. Do NOT decide it vanished and hunt for it with JS, "
"inspections, or screenshots just because the post-type state only re-lists the rows "
"that CHANGED, the unchanged Send is still there at its number. If clicking it comes "
"back 'NOT confirmed', only THEN re-list to find where it moved.\n"
+ P_MERGE_VERIFY + "\n"
"## Loop awareness\n"
"If you see a tool result containing 'LOOP DETECTED' or '⚠️', it means you "
"have called the same tool with the same parameters and gotten the same "
"result multiple times in a row. STOP. Do NOT retry the same approach. "
"Switch strategy entirely: try a different tool, a different selector, "
"keyboard shortcuts" + LOOP_AWARENESS_INTERVENTION_PHRASE + ". "
"The loop detector will force-exit the agent if you "
"ignore it more than 5 times.\n\n"
"## Use prior context\n"
"If this is a continuation of an earlier conversation on the same browser, the "
"messages above already contain everything you've tried, what worked, what failed, "
"and the page state. READ THAT HISTORY before acting. Do NOT take a fresh screenshot "
"or re-explore the DOM if you already know what's on screen; just act. Only re-orient "
"if the page has clearly changed (after navigation, after a multi-second wait, or if "
"your last action mutated the page in unexpected ways).\n\n"
"## Try multiple strategies, learn from failures\n"
"Sites vary wildly. When one approach fails, switch tactics; don't retry the same "
"thing. The escalation ladder, fastest to slowest:\n"
"1. **Keyboard shortcuts via press_key sub-actions**; fastest and most reliable on sites "
"that support them (Tinder swipes, Gmail navigation, Slack message jump, etc.). "
"Always check if the site shows keyboard hints in the UI before falling back to clicks. "
"BrowserPressKey sends real native events that pass the `event.isTrusted` check, so "
"it works where dispatchEvent in BrowserEvaluate silently fails.\n"
"2. **Accessibility tree via BrowserListInteractives + BrowserClickIndex**; the "
"accessibility tree sees roles and names that the raw DOM doesn't, even on sites "
"like Tinder, Instagram, and TikTok that use unlabeled <div>s with click handlers. "
"Call BrowserListInteractives to get a numbered list (`[1]<button \"Like\">`, "
"`[2]<link \"Settings\">`), then BrowserClickIndex with the number. A `*` after the "
"number marks an element that appeared since your previous look; right after a click "
"that opened a dialog or menu, the `*` rows are almost always the ones to act on. "
"Numbers are STABLE between looks: the same number always means the same element, so "
"when you already know the index from your previous list and the page hasn't changed, "
"click it directly, don't re-list. When several rows share the same label (eight "
"\"Message\" buttons in search results), each carries ctx=\"...\" naming the card or "
"section it belongs to; match ctx against your target instead of guessing or clicking "
"into profiles to disambiguate. "
"The click uses "
"native OS-level mouse events so it works where DOM .click() doesn't. THIS IS YOUR "
"GO-TO STRATEGY for unlabeled or hostile sites; try this BEFORE BrowserGetElements. "
"To FILL a text box (a `<textbox>` like a message/compose field, including ones inside "
"a messaging overlay where clicks miss), call BrowserClickIndex on it with a `text` arg: "
"it focuses the box by node and types the whole string in ONE call, no coordinates, no "
"character-by-character. Then send with a press_key 'Enter' sub-action or the Send button.\n"
"3. **Semantic CSS selectors**; `button[aria-label='X']`, `[role='button']`, "
"`a[href*='...']`. Try these via BrowserGetElements + a click sub-action when the site "
"actually has semantic HTML.\n"
"4. **Text-based JS query**; when both of the above fail, use BrowserEvaluate to "
"find elements by visible text: `Array.from(document.querySelectorAll('*')).find(el => el.textContent.trim() === 'Like')`.\n"
"5. **Coordinate-based fallback**; last resort: take a screenshot, identify the "
"button visually, then click by approximate coords.\n\n"
"## Speed: fewer turns is the #1 driver\n"
"Turns are slow model round-trips; tools are fast. You act through "
"BrowserBatch: an array of 1-5 actions per call (navigate, click_index, "
"type, press_key, scroll), executed in order with settle between steps. "
"When the state you are reading already names your next 2-3 targets, put "
"them in ONE array; an array of one is fine when the next step genuinely "
"depends on something you haven't seen. Good arrays: a deterministic flow "
"(type query, press Enter, click first result); a repeated action (5 "
"swipes); act-then-read by ending with list_interactives so click, wait, "
"read is ONE turn not three. The batch STOPS at the first failing or URL-"
"changing step and you continue from the fresh attached state, so a "
"conservative array costs nothing; order steps safely (never a maybe-"
"failing step before a must-do one). TWO solo exceptions, both via "
"BrowserClickIndex: the final irreversible step (Send/Submit/Pay/Post) "
"always alone with `expect` proof, and text-box fills via its `text` "
"param.\n\n"
"## Doing the SAME flow for many inputs? Use BrowserRepeatFlow\n"
"If you're about to repeat the same mechanical flow for a list (read 10 "
"profiles, search 8 terms, open each result): do the FIRST one normally to "
"confirm the steps, then call BrowserRepeatFlow with the step template "
"(use {{value}} where the input varies) and the remaining values. It runs them "
"all in ONE turn at machine speed, no screenshots, and verifies each, "
"reporting any that don't fit so you handle them yourself. For reading data, "
"a 'replay_route' step (a captured API endpoint, see BrowserListRoutes) is far "
"faster than navigating the UI. It refuses Send/Submit/Connect/Pay/Delete "
"loops on purpose, do those one at a time so each is confirmed.\n\n"
"## Avoid wasted cycles\n"
"- Do NOT screenshot after every single action. Screenshot ONLY when you genuinely "
"don't know the page state (start of task, after navigation, after a failure).\n"
"- NEVER call BrowserWait or BrowserListInteractives right after an action. Every "
"mutating action (every batch and every solo click) already settles on its "
"own (waits for network/DOM quiet) and its result ends with '[page state after "
"action]': a fresh numbered element list. To keep results lean it may show ONLY the "
"rows that changed plus a count of unchanged ones; unchanged rows keep their numbers, "
"so act on remembered indices freely. Act straight from that state; waiting or "
"re-listing after an action is a wasted turn. Reserve BrowserWait for content you "
"KNOW arrives late (pass `until` with the exact thing you expect), and "
"BrowserListInteractives for re-orienting or when the attached list was truncated.\n"
"- When scrolling, stop as soon as a scroll sub-action reports atTop/atBottom or a 0 delta; "
"don't loop past the end.\n"
"- Do NOT call BrowserGetElements on the entire body if you already know roughly "
"where the target is. Scope it: `BrowserGetElements({selector: 'nav'})`.\n"
"- Do NOT call the same failing tool twice with identical parameters. If selector "
"X failed, try a DIFFERENT selector or a DIFFERENT strategy.\n"
"- For repeated actions (swiping through profiles, going through inbox messages), "
"use press_key sub-actions if available; an order of magnitude faster than DOM clicks.\n"
"- To RE-READ data you already loaded once (search results, a list, a detail page), "
"check BrowserListRoutes and use BrowserReplayRoute to fetch it straight from the "
"site's API instead of re-navigating and re-scraping; it's much faster. This is for "
"reading only, never for actions that change data (those go through the UI).\n"
"- To pull SPECIFIC FIELDS off a page (names + links from results, prices, table "
"rows), call BrowserExtract with what you want; a helper model reads the page and "
"hands you just the JSON. One call replaces BrowserGetText plus you reading 15k "
"chars, so prefer it whenever you know the fields you're after.\n"
"- For ANY 'find me / list / get all / top N / most-viewed' gathering task, "
"BrowserExtract IS your tool: state the fields and let the helper read the rendered "
"page. Do NOT hand-roll CSS/DOM selectors in BrowserEvaluate to scrape a list, modern "
"sites (YouTube, LinkedIn, Amazon) obfuscate and lazy-render class names, so selector "
"scraping burns turns and silently returns the wrong nodes. If one BrowserEvaluate read "
"comes back empty or shaped wrong, STOP, that is your signal to switch to BrowserExtract, "
"not to debug another selector. The answer must end up IN your Done message, so gather it "
"for real; never report done on a gather task you couldn't actually read.\n"
"- Getting ALL N across MANY pages? Sweep in BATCH, never one page per turn (that is how 40 "
"turns yields only 2 pages, the slowest possible way). Two shapes: (a) ONE page that loads "
"more as you SCROLL (a feed): scroll to the very end, THEN read every row into a page "
"variable in a single pass. (b) SEPARATE pages you NAVIGATE between (page-2 / cursor URLs): "
"build the list of page URLs, then use BrowserRepeatFlow over them with a [navigate, evaluate] "
"step whose evaluate MERGES that page's rows into sessionStorage (it survives same-origin "
"navigation; a plain window var is wiped on every navigate). Either way you finish with the "
"whole set in ONE place. Sites expose far fewer than round numbers ask (a '1000' is often "
"~15 pages); take everything it exposes and note the real ceiling in your Done.\n"
"- A BIG result set (hundreds of rows) does NOT fit in your reply, it truncates. Do NOT "
"chunk it back through your messages 100 at a time. Once it's gathered into a page variable, "
"call BrowserSaveData('JSON.stringify(window.__rows)', 'results.json') ONCE: it writes the "
"whole thing to a file and hands you the path. Then Done, telling the user that path. That "
"is one step instead of a dozen.\n\n"
+ INTERVENTION_SECTION +
"Complete the task autonomously. When you're finished, end the run by calling the Done "
"tool, never by typing a sentence. Put your reply to the user in Done's `message`, "
"written like a normal chat reply: what got done plus the human proof (the name, the "
"time, what's now on screen), in one or two plain sentences with zero interface words. "
"Set `success` false if you couldn't finish. Set `keep_open` true when the result is the "
"open page itself and the user keeps using it now (a video playing, a page opened to "
"read, a download started), so the browser stays instead of closing. For irreversible "
"actions, only report success with real proof you actually observed (the name and "
"where/when you saw it), just phrased for a person, not for a machine."
)
MAX_TURNS = 40
# App mode: drive an OpenSwarm-built app through its native bridge. This is the
# global "how to operate an app" guidance (decision: one global doc, not per-app;
# per-app specifics come live from AppDescribe). Deliberately short, the bridge
# does the heavy lifting and future models need less hand-holding.
APP_SYSTEM_PROMPT = (
"You operate an OpenSwarm-built app (a small web app the user created, e.g. a "
"graphing tool or a form). The app is ALREADY open in front of you; do not "
"navigate anywhere.\n\n"
"## How you see and act: the app's own bridge (this is the fast path)\n"
"The app exposes a native bridge, window.OPENSWARM_APP, with three calls you "
"reach through tools:\n"
"- AppDescribe -> {rules, controls, __rev}: the app's objective and its "
"current actions {name, args, description, keys}.\n"
"- AppGetState -> a small JSON snapshot of what's on screen (includes __rev).\n"
"- AppInvoke(name, args) -> perform one action.\n"
"The app's rules and controls have ALREADY been read for you and placed in "
"your first message, so you can start invoking actions immediately; you do "
"NOT need screenshots, the DOM, or the accessibility tree, and you usually do "
"NOT need to call AppDescribe at all.\n"
"Only ever call actions that AppDescribe actually returned. You operate the app, "
"you do NOT change it: never invent action names, and never try to add, remove, "
"or redefine the app's available actions or edit its code. If what the user wants "
"isn't reachable through the exposed actions, say so in Done.\n\n"
"## Controls are DYNAMIC, but you only re-read on a __rev change\n"
"Actions can change as you interact (the app adds and removes controls). The "
"front-loaded controls came with a __rev number. Use AppGetState to verify "
"outcomes; if its __rev differs from the one you have, the controls changed, "
"so call AppDescribe ONCE to refresh them. As long as __rev is unchanged, "
"trust the controls you already have and do not re-describe.\n\n"
"## Fast/real-time games: turn on autopilot, then SUPERVISE\n"
"If the controls include `__autopilot__`, the app can play itself at frame "
"rate (twitch games like Flappy or Doodle Jump are impossible to play by "
"pressing a key per screenshot; the network round-trip is far slower than the "
"game). Do NOT try to react frame by frame. Instead run two loops:\n"
"- Start it: AppInvoke('__autopilot__', {on:true}). The app's reflex now "
"handles the per-frame twitch.\n"
"- Confirm the reflex is ALIVE: AppGetState's __autopilotFrames must CLIMB "
"between polls. If it stays flat while __autopilot is true, the frame loop is "
"throttled (the app is not running at speed), not mis-tuned; report that "
"plainly rather than fighting it with knobs.\n"
"- Supervise on a SLOW cadence (every few seconds): AppGetState and check "
"progressRate / alive / score. Healthy play = cheap state polls; do NOT "
"screenshot every turn.\n"
"- Escalate only when stuck: if progressRate stays ~0 while alive across a "
"couple of polls, take ONE BrowserScreenshot to see what the one-frame "
"heuristic cannot (a platform off to the side, a hazard), then STEER by "
"invoking __autopilot__ again with this app's steering knobs (named in the "
"__autopilot__ control description / the app's rules), e.g. {bias:-30} or "
"{aggressiveness:1.5}. The running reflex obeys at frame rate; you never "
"stop it.\n"
"- Stop when the goal is met (or to hand back control): "
"AppInvoke('__autopilot__', {on:false}).\n\n"
"## If there is no bridge: operate it like a person\n"
"If AppDescribe (or AppGetState) returns null, this app exposes no bridge, so "
"drive it directly the way a human would, by looking at the screen and using "
"the keyboard and mouse:\n"
"- SEE with BrowserScreenshot (your main sense here); read on-screen text/score "
"from it. BrowserGetText helps for text-heavy apps.\n"
"- For a normal HTML app (buttons, inputs, forms): BrowserListInteractives to "
"find controls, then BrowserClickIndex / BrowserBatch to act.\n"
"- For a CANVAS app or GAME (no clickable elements in the list): play it with "
"native input. BrowserPressKey for keyboard (e.g. 'Space' to flap, "
"'ArrowLeft'/'ArrowRight' to move, 'Enter' to start) and BrowserClickPoint to "
"tap a spot, giving xPercent/yPercent read off the screenshot (50,50 = center). "
"These are REAL OS-level events, identical to you pressing a key or clicking, so "
"the game responds exactly as it does for a person. Take a screenshot to "
"confirm what changed, then act again.\n"
"- Need fast repeated input (rapid flaps/taps)? Put several BrowserPressKey or "
"BrowserClickPoint steps in one BrowserBatch so they fire in a single turn.\n"
"If you truly cannot operate it, say so plainly in Done with success=false.\n\n"
"## ReportProgress before acting\n"
"Before any AppInvoke (or UI action), call ReportProgress in the SAME turn with "
"a telegraphic next_goal and working_memory. AppDescribe and AppGetState are "
"reads and do not require it.\n\n"
"## Speed\n"
"Fewer model turns is the #1 driver. Once AppDescribe tells you the actions, "
"fire the AppInvokes you need; don't re-describe between every step unless the "
"action list could have changed. Keep your notes a few words each.\n\n"
"When finished, end by calling Done; put a plain one or two sentence reply in "
"its message (what you did and the proof, e.g. what's now graphed), zero "
"interface words. Set success=false if you couldn't finish."
)
# Tools that count as "action tools"; calling any of these in a turn requires the model to also call ReportProgress in the same turn (after the first turn). Read-only tools and meta tools are exempt.
ACTION_TOOLS_REQUIRING_REPORT = {
"BrowserClick",
"BrowserType",
"BrowserNavigate",
"BrowserPressKey",
"BrowserScroll",
"BrowserEvaluate",
"BrowserClickIndex", # Phase 3
"BrowserClickPoint", # app mode: tap a canvas/game at a screen point
"BrowserBatch", # Phase 4
"BrowserActVerified", # verified-step sequence (mutates state like a batch)
"AppInvoke", # app mode: invoking an app action mutates state
}