Files
browser/docs/agent-script.md
Francis Bouvier bdf8bd1031 agent: Improve agent script doc (#2692)
* ci: raise cdp bench MAX_VmHWM to 29000

The agent branch grew .text by ~2.8MB (mostly std monomorphization from
the new script/agent modules); none of it executes in serve mode and
anonymous/heap memory is byte-identical to main (716-724KB Private_Dirty
across main, agent-dev, and a build with the agent code compiled out).

The VmHWM delta is file-backed pages of the binary: the kernel maps them
in large readahead/THP chunks around each touched page, so peak RSS
shifts with link layout on every relink. The agent-less test build
measured 4.2MB *below* main while running the identical serve path,
purely from folio/window boundaries moving. At 28000 the gate was
failing on layout noise (28028-28064), not a regression.

* agent: improve docs agent_script

* Revert "ci: raise cdp bench MAX_VmHWM to 29000"

This reverts commit b6b704d43e.

---------

Co-authored-by: Adrià Arrufat <adria.arrufat@gmail.com>
2026-06-11 09:51:50 +02:00

15 KiB

Agent JavaScript scripts

lightpanda agent <script.js> runs a JavaScript file that drives a Lightpanda browser session through a small set of blocking global functions. The format is intentionally plain JavaScript: use normal variables, functions, loops, objects, arrays, JSON.parse, JSON.stringify, and other standard ECMAScript built-ins.

./lightpanda agent session.js

The interactive REPL still uses slash commands. /save writes the recorded browser actions as JavaScript calls so the saved file can be replayed without an LLM.

Runtime Environment

Agent scripts run in their own V8 context. That context is separate from the web page's JavaScript context.

  • It is not the browser page environment. There is no window, document, DOM, localStorage, navigator, or page global state in the agent script. Read page data with extract(...), or explicitly run page JavaScript with evaluate(...) when that is the right tool.
  • It is not Node.js. There is no require, process, fs, path, npm package loading, command-line argument API, or Node network/filesystem API.
  • Page scripts cannot see agent variables or Lightpanda primitives. Agent scripts cannot directly see page variables.
  • The global evaluate(...) primitive runs JavaScript in the page context, distinct from the agent context's own native eval.
  • Agent variables persist for the lifetime of one script run, across navigations and primitive calls. A later lightpanda agent script.js run starts with a fresh agent context.
  • The installed primitives are synchronous and blocking. Do not write an async/await automation contract around them. Scripts compile as classic scripts, so top-level await is a SyntaxError; promise callbacks (.then) only run after the script body finishes, never in between primitive calls.
  • Tool failures throw JavaScript Error exceptions and stop execution unless you catch them.
  • The script's completion value — its last top-level expression — is printed automatically (objects and arrays as JSON; other values coerced). End a script with the bare expression you want as output, e.g. a final extract({ ... }); or results;. console.log(...) is for extra or debug output and does not JSON-format objects.

The agent context includes a small console object:

console.log("printed to stdout");
console.info("printed to stdout");
console.debug("printed to stdout");
console.warn("printed to stderr");
console.error("printed to stderr");

Values And Return Types

Most primitives return the browser tool's result text as a JavaScript string. extract(...) is the exception: it returns extracted data as a normal JavaScript value, so local script logic can use it directly. The result mirrors your schema — an object schema returns an object keyed by your fields (even with a single field), and a bare array schema returns an array:

goto("https://news.ycombinator.com/");

const data = extract({
  title: "title",
  stories: [{
    selector: "tr.athing",
    limit: 5,
    fields: {
      id: { attr: "id" },
      title: ".titleline > a"
    }
  }]
});

data; // printed automatically as JSON

Destructure when a single field is all you need:

const { stories } = extract({
  stories: [{
    selector: "tr.athing",
    limit: 5,
    fields: {
      id: { attr: "id" },
      title: ".titleline > a"
    }
  }]
});

for (const story of stories) {
  console.log(story.title);
}

evaluate(...) still returns the page evaluate tool result text. When page evaluate(...) returns an object or array, that text is JSON.

Primitive arguments must be JSON-serializable. Strings, numbers, booleans, arrays, plain objects, and null work. undefined, functions, symbols, and cyclic objects do not.

Installed Primitives

Only recorded browser primitives are installed globally:

Primitive Arguments Runs in
goto goto(url[, { timeout }]) Browser session
extract extract(schema) or extract({ schema }) Browser page via extractor; returns a JS object or array
evaluate evaluate(script[, { url, timeout, save }]) Browser page JS context
click click(selector) or click({ selector }) Browser page
fill fill(selector, value) or fill({ selector, value }) Browser page
scroll scroll() or scroll({ x, y }) Browser page
waitForSelector waitForSelector(selector[, { timeout }]) Browser page
waitForScript waitForScript(script[, { timeout }]) Browser page JS context
waitForState waitForState(state[, { timeout }]) Browser page
hover hover(selector) or hover({ selector }) Browser page
press press(selector, key) or press({ key[, selector] }) Browser page
selectOption selectOption(selector, value) or selectOption({ selector, value }) Browser page
setChecked setChecked(selector[, checked]) or setChecked({ selector, checked }) Browser page

goto returns at the load event (a fast snapshot). When a page's content is still loading (rendered by post-load JS), call waitForState("networkidle") before reading. waitForState's state accepts "load", "domcontentloaded", "networkalmostidle", "networkidle", or "done". goto's timeout defaults to 10000 ms; the waitFor* timeouts default to 5000 ms.

The [, { … }] is an optional trailing options object: leading arguments are positional (waitForSelector("#row", { timeout: 2000 })), and the options ride in a final object. Passing a single object with everything (waitForSelector({ selector: "#row", timeout: 2000 })) is equivalent — that's the shape /save records into saved scripts. An option can't be a bare positional, though: waitForSelector("#row", 2000) is an error. A null positional omits that field (press(null, "Enter") presses on the focused element), and setting the same field positionally and in the options object (goto(url, { url: ... })) is an invalid arguments error.

Script primitives address elements by CSS selector only. The tools that hand out backendNodeIds (tree, findElement, nodeDetails) aren't installed in the script context, and a raw node ID wouldn't survive replay anyway. When you're exploring in the REPL and have a backendNodeId — e.g. the leading number on a /tree line, or a /findElement hit — run /nodeDetails backendNodeId=<id> to get a durable CSS selector, then paste that into your script.

Navigation

Use goto(...) to open a page:

goto("https://example.com");

goto({
  url: "https://example.com/app",
  timeout: 15000
});

The call returns a status string and throws if navigation fails. A timeout does not throw: the call returns "Navigation started but the page did not finish loading before the timeout." and the page stays in whatever state it reached. Check the return value — or follow with waitForState(...) / waitForSelector(...) — when completeness matters.

Structured Extraction

Use extract(...) to read data from the current page without writing page-side JavaScript. This is the preferred bridge from page content into local agent logic.

const result = extract({
  heading: "h1",
  links: [{
    selector: "a",
    limit: 10,
    fields: {
      text: "",
      href: { attr: "href" }
    }
  }]
});

The schema forms are:

Schema value Meaning
"<selector>" Text of the first matching element, or null
"" Text of the current matched element inside a fields block
["<selector>"] Text of all matching elements
{ selector: "<selector>", attr: "<name>" } Attribute from the first match
[{ selector: "<selector>", attr: "<name>" }] Attribute from all matches
[{ selector: "<selector>", fields: { ... } }] Array of records, with fields resolved relative to each matched element
limit: N Cap array extraction to N matches

Return shape follows the top-level schema:

  • extract({ title: "h1" }) returns { title: "..." }.
  • extract({ title: "h1", links: [{ selector: "a" }] }) returns an object with both fields.
  • extract({ links: [{ selector: "a" }] }) returns { links: [...] } — an object schema always returns an object, even with a single field.
  • extract([{ selector: "a" }]) is shorthand for a single anonymous array extraction and returns the array directly.

Every value is a string (trimmed text or a raw attribute) or null — parse numbers in script logic. An array field that matches nothing yields [] without complaint (a page with zero comments is a valid result), but if every field in the schema misses, extract(...) throws no schema selector matched any element — treat that as "my selectors are wrong", not "the page is empty".

extract(...) reads only the current page. For list-to-detail scraping — capture a list, then visit each row for more — capture the list, then loop in the script: goto each row's URL and extract the detail. The local agent context keeps the data across navigations, so the assembly happens in plain JavaScript. See the complete example below.

When passing an object directly to extract(...), the runtime serializes it as the extractor schema. These forms are equivalent:

extract({ title: "h1" });
extract({ schema: { title: "h1" } });
extract('{ "title": "h1" }');

The wrapped form accepts only schema: the REPL's save= option does not exist in scripts (extract({ schema: ..., save: ... }) is rejected). Keep results in local variables instead.

Use local variables to keep extracted data available to later script logic:

const page = extract({ title: "title" });

Page JavaScript

evaluate(...) is the explicit escape hatch into the current page's JavaScript context. Its script string runs where window and document exist.

goto("https://example.com");

const title = evaluate("document.title");
console.log(title);

Keep the boundary clear:

const selector = "h1";

// Good: local agent logic builds an extract schema.
const data = extract({ heading: selector });

// Bad: page evaluate cannot see local agent variables.
evaluate("document.querySelector(selector).textContent");

Page evaluate(...) cannot call goto, extract, or other agent primitives. Agent scripts cannot access document directly. If you need page DOM data, prefer extract(...); use evaluate(...) only for page behavior that extraction cannot express.

waitForScript(...) also evaluates in the page context, repeatedly, until the expression is truthy or the timeout expires:

waitForScript("document.querySelectorAll('.row').length >= 5");

Interaction Primitives

The action primitives operate on the current page. Most take one object whose fields match the browser tool schema:

click({ selector: "a.login" });
fill({ selector: "input[name='acct']", value: "$LP_HN_USERNAME" });
fill({ selector: "input[name='pw']", value: "$LP_HN_PASSWORD" });
press({ key: "Enter" });

waitForSelector("#logout");

hover({ selector: "#menu" });
selectOption({ selector: "select[name='country']", value: "FR" });
setChecked({ selector: "input[name='terms']", checked: true });
setChecked({ selector: "input[name='newsletter']", checked: false });

scroll({ y: 600 });
scroll();

setChecked defaults checked to true when the field is omitted (setChecked("#chk") checks the box). press's leading positional is the optional selector, not key: a bare press("Enter") binds "Enter" to selector and fails. Press on the focused element with press({ key: "Enter" }) or press(null, "Enter"); target an element with press("#search", "Enter") or press({ key: "Enter", selector: "#search" }).

$LP_* placeholders in string arguments are resolved inside the Lightpanda process. This keeps credentials out of recorded scripts and LLM prompts. In recordings, resolved LP_* values are scrubbed back to placeholders.

Recording Format

The REPL remains slash-command based:

> /goto https://example.com
> /click selector='a.login'
> /save

/save writes JavaScript by default:

goto("https://example.com");
click({ selector: "a.login" });

Only replayable browser actions are recorded:

  • Recorded: goto, click, fill, scroll, hover, press, selectOption, setChecked, waitForSelector, waitForScript, waitForState, evaluate, and extract — the same set installed as script primitives.
  • Not recorded: read-only exploration tools such as tree, markdown, html, links, findElement, consoleLogs, getUrl, getCookies, and getEnv.
  • Natural-language prompts are written as // comments above the actions they produced.

This is the deterministic --no-llm transcription. When /save runs with an LLM it instead synthesizes an idiomatic script from the session and is asked to emit JavaScript only, so those // comments generally won't appear.

Error Handling

Primitive failures throw JavaScript exceptions:

try {
  waitForSelector({ selector: "#dashboard", timeout: 1000 });
} catch (err) {
  console.error("dashboard did not appear:", err.message);
  throw err;
}

Common failures:

Error Meaning
ReferenceError: document is not defined You tried to use browser DOM APIs in the agent context. Use extract(...) or page evaluate(...).
ReferenceError: require is not defined Agent scripts are not Node.js scripts.
no page loaded - run goto(url) first A page-dependent primitive ran before navigation.
invalid arguments A primitive received the wrong number or shape of arguments, or a non-JSON-serializable value.
extract: no schema selector matched any element Every field in the schema missed. Fix the selectors; an empty page section yields null/[] per field, not this error.

Complete Example

This script opens Hacker News, extracts five stories, visits each comments page, and prints one JSON object. The looping and object assembly happen in the local agent script, not in the page.

const HN = "https://news.ycombinator.com";

goto(HN);

const { stories } = extract({
  stories: [{
    selector: "tr.athing",
    limit: 5,
    fields: {
      id: { attr: "id" },
      title: ".titleline > a",
      url: { selector: ".titleline > a", attr: "href" }
    }
  }]
});

for (const story of stories) {
  story.comments = [];
  if (!story.id) continue;

  goto(`${HN}/item?id=${story.id}`);
  const { comments } = extract({
    comments: [{
      selector: "tr.athing.comtr:has(.commtext)",
      limit: 3,
      fields: {
        author: ".hnuser",
        text: ".commtext"
      }
    }]
  });
  story.comments = comments;
}

stories; // printed automatically as JSON