Global site search

Search guides, labs, glossary, and research

Type two or more characters to search.

Start with a channel, artifact, or defense term

Examples include zero-width, metadata, tokenizer, or prompt injection.

    DOM · AX TREE · GEOMETRY · SCREENSHOT

    Browser Automation and Agent-Facing Views

    Reconcile DOM, accessibility semantics, geometry, screenshots, focus, state, timing, and capability checks before allowing browser-agent actions.

    Quick answer

    What does this Machine Tradecraft expansion explain?

    Reliable browser automation combines semantic role and accessible-name targeting with live DOM attachment, accessibility state, geometric hit testing, viewport visibility, timing, and explicit confirmation for consequential actions.

    Semantic view
    The accessibility tree exposes roles, names, states, and relationships efficiently.
    Geometric view
    Layout and hit testing reveal overlap, off-screen targets, and pointer interception.
    Privilege boundary
    A browser agent with cookies, files, network, or script execution is a privileged software component.
    Reviewed implementation boundary

    Know what is measured, approximated, and still external.

    The submitted report remains byte-identical. This separate review, checked 2026-08-26, narrows implementation claims and gives visitors a decision path before they generalize from a local result.

    IMPLEMENTATION MODE Deterministic public fixtures plus repository-owned real-browser tests
    VISIBLE SOURCE PROFILE 6 standards/specifications · 0 research · 0 government · 3 implementation
    SOURCE BODY Preserved; corrections live in this review layer
    Directly computed

    Output produced deterministically by the local runtime.

    • The selected fixture state, static DOM example, expected semantic snapshot, focus model, geometry statement, locator, and confirmation requirement
    • For the MachineTradecraft site itself, the repository harness captures actual local Chromium DOM, geometry, focus, accessibility trees, screenshots, and interactions
    • No public-lab action navigates externally or performs a real side effect
    Bounded approximation

    Useful model or subset that must not be mistaken for full conformance.

    • Public fixture snapshots are authored models, not live accessibility trees from arbitrary submitted pages
    • Accessible-tree exposure, Shadow DOM behavior, top-layer handling, and platform mappings vary by browser and operating system
    • Semantic locators do not eliminate the need for live attachment, visibility, overlap, enabled-state, and confirmation checks
    Requires external verification

    Conclusion that needs an exact implementation, trust system, model, parser, or human review.

    • Cross-browser automation requires the target browser, driver or protocol version, operating system, and application state
    • Closed Shadow DOM is not universally pierceable through ordinary DOM locators; any accessibility exposure is implementation-dependent
    • Destructive actions require application-level capability policy and independent human confirmation
    Decision support

    Choose the next evidence step instead of treating one result as a verdict.

    QuestionWhat the local page can answerWhat it does not establishNext evidence step
    What semantic role and name should identify the fixture control?The deterministic fixture provides an expected locator.Live target stability or cross-browser equivalence.Capture the actual target's DOM and accessibility tree.
    Can the control be clicked safely?The fixture explains geometry and confirmation factors.Current overlap, animation, enabled state, or consequence.Run actionability and policy checks immediately before action.
    Is the interface agent-safe?No global verdict is issued.Prompt injection, privilege, session, or side-effect containment.Threat-model the full browser-agent stack.
    Focused deterministic fixture

    Disclosure, hidden payload, destructive dialog, and overlay set

    Compares semantics, geometry, focus, and capability decisions.

    Expected boundary: Each fixture exposes at least one reason that a single representation is insufficient.

    Open prepared laboratory
    Version and authority checkpoints
    • WebDriverW3C browser-automation specification; target implementation still matters
    • CDP Accessibility domainChromium protocol documentation used by the local harness
    • AccName 1.2W3C Recommendation used for accessible-name terminology
    Compare independent views

    One artifact, several machine-readable representations

    No single view is automatically authoritative. Preserve the source, identify each parser or receiver, and compare their outputs before authorizing a consequential decision.

    01Live DOM

    Current nodes, attributes, attachment, and mutation state.

    02Accessibility tree

    Roles, names, descriptions, states, and properties.

    03Layout geometry

    Bounding boxes, viewport position, overlap, and hit-test target.

    04Screenshot

    The rasterized human-visible representation.

    05Focus and modality

    Current focus, tab order, active dialog, and inert background.

    06Capability context

    The browser profile, cookies, tools, and allowed side effects.

    Bounded method

    Analysis workflow

    The workflow preserves evidence before transformation and keeps structural inspection separate from execution, remote verification, or model behavior.

    1. Locate by role and accessible name where possible.
    2. Re-query the live node immediately before action.
    3. Confirm visible, stable, enabled, non-inert, and unobscured geometry.
    4. Respect modal, focus, and top-layer state.
    5. Require a transaction preview and human approval for consequential actions.
    6. Record the semantic target and actual hit-test result.
    Defense in depth

    Controls carried into implementation

    These controls are contextual. They reduce a defined risk; they do not guarantee safety, truth, attribution, or resistance to every adaptive attack.

    01

    Use native semantics and names that match visible purpose.

    02

    Avoid hidden or off-screen focusable controls.

    03

    Keep untrusted page content separated from agent policy and tools.

    04

    Use isolated ephemeral browser contexts for untrusted tasks.

    05

    Disable arbitrary script execution and unrestricted file or network tools.

    Shared vocabulary

    Key terms

    Definitions are linked into the site-wide glossary and back to the full report.

    WebDriver

    The W3C remote-control interface for browsers.

    DevTools Protocol

    A browser instrumentation protocol exposing domains such as DOM, Page, Network, and Accessibility.

    Locator

    A query strategy that identifies an interface object for repeated live evaluation.

    Hit testing

    Determining which rendered object receives an input at a coordinate.

    Stale element

    A prior node reference that no longer identifies an attached live element.

    Semantic control

    A native or correctly mapped widget exposing purpose and state to accessibility APIs.

    ARIA snapshot

    A serialized accessibility-tree representation used for testing or agent observation.

    Actionability check

    A pre-action verification of attachment, visibility, stability, enabled state, and event reception.

    Continue with primary material

    External standards and research

    These links are provided for visitors who want the governing specification, paper, framework, or implementation documentation. Links open in a new tab; the site does not fetch them during runtime analysis.

    Primary standard W3C specification

    WebDriver

    Cross-browser remote-control standard.

    Normative or first-party specification material.
    www.w3.org
    Primary standard W3C Recommendation

    WAI-ARIA 1.2

    Roles, states, and properties.

    Normative or first-party specification material.
    www.w3.org
    Primary standard W3C Recommendation

    Core-AAM 1.2

    Platform accessibility mappings.

    Normative or first-party specification material.
    www.w3.org
    Primary standard W3C Recommendation

    AccName 1.2

    Accessible-name and description computation.

    Normative or first-party specification material.
    www.w3.org
    Implementation reference Protocol documentation

    Chrome DevTools Protocol

    Chromium instrumentation domains.

    Tool, vendor, or implementation documentation; behavior is version-specific.
    chromedevtools.github.io
    Implementation reference Protocol documentation

    CDP Accessibility domain

    Accessibility-tree types and commands.

    Tool, vendor, or implementation documentation; behavior is version-specific.
    chromedevtools.github.io
    Implementation reference Implementation documentation

    Playwright accessibility testing

    Tool-specific accessibility testing guidance.

    Tool, vendor, or implementation documentation; behavior is version-specific.
    playwright.dev
    Implementation reference Implementation documentation

    Playwright ARIA snapshots

    Serialized semantic snapshot testing.

    Tool, vendor, or implementation documentation; behavior is version-specific.
    playwright.dev
    Implementation reference Implementation documentation

    Playwright MCP

    Browser automation exposed through an agent tool protocol.

    Tool, vendor, or implementation documentation; behavior is version-specific.
    playwright.dev
    Continue the investigation

    Read the evidence, then test the bounded model

    The full submitted report is preserved byte-for-byte in the governed research library and in durable repository documentation. The laboratory turns selected concepts into deterministic local output without external calls or hidden persistence.

    Detailed report

    Browser Automation, Accessibility Trees, and Agent-Facing Web Representations

    A detailed treatment of WebDriver, CDP, live DOM state, accessibility mappings, accessible names, geometry, screenshots, locators, timing, dialogs, prompt injection, capability safeguards, and semantic testing.

    Read governed report
    Focused laboratory

    Browser Agent-View Simulator

    Compare deterministic fixture DOM, accessibility snapshot, focus order, geometry, hidden state, locator strategy, and confirmation requirements without external navigation or real side effects.

    Open bounded laboratory